@onlyworlds/sdk 4.178.1 → 4.179.1

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/CHANGELOG.md DELETED
@@ -1,2805 +0,0 @@
1
- # Changelog
2
-
3
- All notable changes to `@onlyworlds/sdk`. Maintained from 3.1.0 onward;
4
- earlier history lives in git log only.
5
-
6
- ## [4.178.1] - 2026-11-17
7
-
8
- ### Fixed
9
-
10
- - **Installing the package into a plain JavaScript project no longer installs TypeScript.** TypeScript was a required peer, so npm placed the newest compiler beside the package in every project, used or not: two of the four packages a first `npm install @onlyworlds/sdk` added. It is an optional peer now: a TypeScript project brings its own (4.5 or later, as before) and a JavaScript project gets the package and `fflate`, nothing else.
11
- - **`@onlyworlds/sdk/waymark` has types in a project on the older module resolution.** With `moduleResolution` set to `node` (or `node10`), which does not read a package's `exports`, the import answered TS2307, "Cannot find module", while the package root's types were found. The package now carries a `typesVersions` entry for the subpath. Projects on `node16`, `nodenext` or `bundler` were not affected and read the same files as before. Read with TypeScript 4.5, 4.7, 5.0, 5.9, 6.0 and 7.0 against the packed build.
12
-
13
- ### Tests
14
-
15
- - **What a TypeScript project is handed is compiled, under each module resolution.** A new test type-checks a consumer of the built package's root and `/waymark` subpath on `node`, `node16` and `bundler`, and holds every typed subpath of `exports` to its `typesVersions` entry and the peer to optional.
16
-
17
- - **Node 18 is tested.** `package.json` admits Node 18 and later, and the suite ran on 20 and 22 only. It now runs on 18 as well, with one flag: Sign in with OnlyWorlds needs Web Crypto as a global, which Node has from 19 on and Node 18 only when started with `--experimental-global-webcrypto`. Without it `signInWithOnlyWorlds` and `pkceChallenge` refuse with a `TypeError` that names Web Crypto, storing and opening nothing (a new test holds that on every Node), and the rest of the package runs.
18
- - **Node 24 is tested**, beside 18, 20 and 22.
19
- - **The build and the tests install from a committed lockfile.** `package-lock.json` was ignored, so every CI run and every release installed whatever the version ranges resolved to that day, and `npm ci || npm install` hid that there was no lockfile to install from. It is committed now and CI and the release run `npm ci`. Nothing changes for a project that installs the package: a lockfile is not published.
20
- - **happy-dom 21.0.4**, which only the tests use (the Waymark's DOM), past a published advisory on 21.0.3 and earlier (GHSA-834v-8h66-3h79). It is a development dependency and never part of the published package.
21
-
22
- ## [4.178.0] - 2026-11-17
23
-
24
- ### Added
25
-
26
- - **`by_role` on every row of a draught's `work`**, which `GET /character/{id}/context` now serves: the ids of her Titles that the work item names, each once, in the item's order, and an empty list when none does. Work is for her by name (the item's `actor`) or by a role she holds (the item's `x_packet_for` names a Title whose `holders` name her), and a row that reached her only through a Title has `via` `['holders']`. Read as her (`mode: 'voice'`) a work item reaches her only through a field of its own that names her, so a caller asks with `mode: 'act'` for what is hers by role.
27
-
28
- ### Fixed
29
-
30
- - **The message for a write the host refused says the whole way out.** The same command passes in a fenced code block whose first line names a language, or with no backtick, semicolon or pipe directly before it; the message named only the backtick.
31
-
32
- ## [4.177.0] - 2026-11-17
33
-
34
- ### Added
35
-
36
- - **The charter and strand packets say each element's type** (schema-dist serial 133). `charterOf` answers `type: 'law'` on the Charter and `type: 'construct'` on the Guide and on each Domain; `strandOf` answers `type` on every filled slot (`narrative`, `event`, `location`, `character`, and `location` for `now`): the kind the slot holds, singular and lowercase, never the spelling stored. A packet is handed to a reader that may hold nothing else of the world, and an id is fetched by its type. `OwReadingAidsCharter`, `OwReadingAidsGuide`, `OwReadingAidsDomain` and `OwStrandSlot` declare the key, so `OwServedReadingAids` reads it as onlyworlds.com's `reading_aids` serves it. The schema pin moves to `v0.34.1-dist.133`; the charter fixture holds 47 cases and the strand fixture 29.
37
-
38
- ### Fixed
39
-
40
- - **A Domain's hub is an element of the kind its field links to, in the reader's view.** An id of another kind in a hub field, or of an element the reader does not see, names no hub. Wherever the charter and strand readers compare an element's type to a kind, the type is trimmed and compared without regard to case, as the room profile defines it.
41
- - **`getAccessKey`'s deprecation note names `ai('ask', ...)` and `ai('sort', ...)`**, where it sent a reader to `ask()` and `sort()`, themselves deprecated since 4.139.0. The example at the head of the ask module calls `ai('ask', ...)` too. The by-links-3d frame comment says every candidate is tested, and the README names the fixtures without a case count.
42
-
43
- ### Deprecated
44
-
45
- - **The `version` key on `layoutByKind`'s and `layoutAround`'s answers** leaves in 5.0. It was always 1, and the layout profile's mark carries no version key: read `method`.
46
- - **Five token types that serve only methods already deprecated** (`TokenConsumeParams`, `TokenConsumeResponse`, `AccessKeyResponse`, `RevokeSessionResponse`, `RevokeAllSessionsResponse`) are marked and leave in 5.0 with those methods.
47
-
48
- ## [4.176.0] - 2026-11-16
49
-
50
- ### Added
51
-
52
- - **`front_door.charter_element` and `front_door.charter_skill_elements`**, which `GET /me`, `PUT /me/character` and an AI seat's join now serve: the Charter and its skills typed as `{type, id}`, so a reader knows which route fetches each. `OwFrontDoorCharter` (`type: 'law'`, or null where `charter` is) and `OwFrontDoorSkill` (`type: 'object'`, in `charter_skill`'s order) are exported.
53
-
54
- ### Fixed
55
-
56
- - **`OwSeat`'s descriptions of the Character right** follow the platform's: the owner may give a seat its Character without the right; once the right is cleared, sending the same Character again keeps it cleared, and only the owner's `may_change_character` true gives it back; and `may_change_character_note` also says why a seat can change its Character all the same (it made it, or it is a co-builder), not only why it cannot.
57
-
58
- ## [4.175.0] - 2026-11-16
59
-
60
- ### Added
61
-
62
- - **`OwApiError.isHostBlocked`: a write the host refused before the API saw it.** The host in front of onlyworlds.com answers some writes with a 403 whose body is an HTML page titled "Blocked", not the API's JSON envelope: text that holds a shell command such as `python -m`, `python -c`, `perl -e`, or `curl` or `wget` with an address right after a backtick, a semicolon or a pipe. Such an error read "OnlyWorlds API error 403" until now. It now says the host refused the request and nothing was written, and how to write the command so it passes (a fenced code block whose first line names a language, or no backtick). `code` stays null, as it is no API code, and `detail` keeps the page. Read on the page's title and on the body not being JSON, so the API's own 403s read as before.
63
-
64
- ### Fixed
65
-
66
- - **`removeSeat`'s description** says when the Character a seat plays goes with it: only while it is as the seat's join left it (made by the join, named as the seat, supertype `Agent` alone, last written by the join, linked from nothing and played by no one else). Any other stays, one given to the seat to play included.
67
-
68
- ## [4.174.0] - 2026-11-16
69
-
70
- ### Added
71
-
72
- - **`BY_LINKS_3D_VERSION`: the version of `layoutByLinks3d`'s answers.** It starts at 1 and is raised exactly when the answer to some case of the method's fixture changes (a constant, a rule, or a fix that moves a point), never for a release that moves none. A tool that keeps a picture to hand back as `held` can store it beside the points and read a different value as no picture kept, so a change to the method lays the world out fresh instead of holding points the method no longer gives. A test pins every fixture case's answer by its sha256 under the version (`test/fixtures/layout/by-links-3d.answers.json`), and `scripts/pin-layout-answers.mjs` refuses to pin a moved answer unless the version has been raised. No point of the method moves.
73
-
74
- ## [4.173.1] - 2026-11-16
75
-
76
- ### Fixed
77
-
78
- - **A bundle that never lays out in 3D no longer carries the 3D method's golden angle.** It was computed at the module's top level, and since the multiplication is not marked pure, every bundle kept it (25 bytes). It is now written as the double it is, and a test holds the literal to `Math.PI * (3 - Math.sqrt(5))`. No point of the method moves.
79
-
80
- ## [4.173.0] - 2026-11-16
81
-
82
- ### Added
83
-
84
- - **`unplacedByLinks3d`: what a saved `by-links-3d` layout map does not place.** `unplacedByLinks3d(graph, placed)` answers the ids of the graph's nodes that are not in `placed` (the ids the Map's Pins stand for, a list or a Set), by id in code point order. Every node counts, of any type and with or without a link, because the method places every node it is handed. A member of `placed` that is not a non-empty string is passed over; a `placed` that is not a list or a Set, or a graph that is not `{ nodes, edges }`, throws a TypeError, and two nodes with one id throw as `layoutByLinks3d` does.
85
-
86
- ### Changed
87
-
88
- - **The method's shared fixture gains eight cases and an `unplaced` suite of eight** (72 cases and 8), the same bytes as the Python package's copy, its sha256 pinned in the test. No answer of the method changes.
89
-
90
- ## [4.172.0] - 2026-11-16
91
-
92
- ### Added
93
-
94
- - **A seat row says whether the seat may change its Character.** `listSeats()` rows (`OwSeat`) carry three fields the API serves on every row since 2026-11-16: `may_change_character` (true when the seat may change, never delete, the Character it plays), `may_change_character_note` (a sentence saying why it may not, else null) and `follows_sponsor` (true for a seat its member made to play their own Character). Where a Character is kept from you the row reads `character` null, `may_change_character` false and no note. The rows already came back whole; this release types them.
95
-
96
- ## [4.171.1] - 2026-11-16
97
-
98
- ### Fixed
99
-
100
- - **`layoutByLinks3d` no longer throws on a Map whose width or height is under 0.** A Map's `width` and `height` are any whole number in the standard. One under 0 made the call throw from inside the method ("the square root of a negative area") when that Map held the most pinned elements, and with both under 0 the Map became a plate with every place mirrored. A width or height that is not over 0 is now no size: such a Map is no plate, no other Map is tried in its stead, and the world is laid out by its links alone; a Map over a place with such a size is no small plate. The method's fixture gains four cases for it (64).
101
- - **Held points cost what they move.** The force steps worked out the push between every two fixed elements, 300 times, though neither can move and the push was never read. They no longer do; no number of any answer changes. Laying out a made world of 9,000 elements with every point handed back took 1,147 ms and takes 269 ms, the same answer to the bit, so a layout that keeps a picture still now costs less than one from nothing.
102
-
103
- ### Changed
104
-
105
- - **The `held` list's words say what false means.** A pinned element stands at its Pin and is never answered as held, so `held: false` does not mean new. What is new is every id that was not handed in.
106
-
107
- ## [4.171.0] - 2026-11-16
108
-
109
- ### Added
110
-
111
- - **`layoutByLinks3d`: a 3D layout of a whole graph by its links.** `layoutByLinks3d(graph, elements?, options?)` takes `worldGraph`'s answer and, optionally, the world's elements (only Map and Pin elements are read, by `mapSpace` and `layoutOf`). The body of the graph is cut into groups round its centres, the elements with most links (twelve at most, more when a group that holds over half the body is split), packed apart in directions read from a table baked into the package, and relaxed for 300 seeded force steps; what hangs off the body is fanned out from it, and what is not joined to the main piece stands on a shelf under the picture. The world's Map that holds the most elements of the graph by a Pin (three or more) is a flat plate with each of them at its Pin, and a Map over a place on it a small plate above it; a Pin's z is not read, and a layout map is never the plate. The answer is one row per node in the order handed in: `points` (x, y, z in a `Float64Array`, in units where a link at rest is `BY_LINKS_3D_REST`, 60, long), and per element `how`, `group`, `part` and `held`, with the centres' ids; the words of `how` and `part` are the exported types `OwLayout3dHow` and `OwLayout3dPart`. The same input gives the same points in every runtime: to the last bit for every centre, grouped and pinned element, and within a billionth of a rest for the rest, which are placed with the runtime's cosine and sine.
112
- - **Held points.** `options.held` hands back points an earlier answer gave (element id to `[x, y, z]`, a Map or a plain object), so the picture keeps still as the world changes; `options.heldUnit` says the step they were rounded to. A held element that is still a node and is not pinned stands at its held point, and the answer's `held` says which. A held value that is not a Map or a plain object, a point that is not three finite numbers, held points wider than 768 rests, and a `heldUnit` outside 0 to an eighth of a rest are refused before any step of the layout runs, with the codes `held-not-object`, `held-not-point` (TypeError), `held-box` and `held-unit` (RangeError). An unknown option key is refused by name.
113
- - **The word `by-links-3d` is the tool's own until the method's page is published, and until then a release may move its points.** The method's shared fixture is the test, its sha256 pinned, and the directions table is held to the sha256 the fixture names.
114
-
115
- ## [4.170.1] - 2026-11-16
116
-
117
- ### Fixed
118
-
119
- - The comment on `unplacedByKind` said a saved layout map changes only when a person asks. The map-space profile's rule 8 says something else: reading or drawing a layout map never changes it, and laying it out again is a write like any other. The comment now says the first. No behaviour changes.
120
-
121
- ## [4.170.0] - 2026-11-16
122
-
123
- ### Added
124
-
125
- - **`unplacedByKind` and `unplacedAround`: what a saved default layout does not place.** Given the graph and the ids of the elements a layout Map's Pins stand for (a list or a Set), each answers the ids its method would place that the Map has no Pin for, by id in code point order: for `by-kind` every node of the graph, for `around` the focus and every node within its steps. An id among the placed ones that the method would not place is no part of the answer. A tool shows the answer's length as its count; neither function places anything.
126
-
127
- ### Changed
128
-
129
- - **The schema pin moves to `v0.34.1-dist.132`** (canonical 00.34.01 unchanged; no schema file moved). The distribution gains the layout profile: the page for the two default methods, `by-kind` and `around`, and for what a saved layout does not place, with its fixture. The map-space profile's rule 8 names it, and says that reading or drawing a layout map changes nothing of its layout.
130
- - **A method's word is its version.** The layout profile rules that a method which changes takes a new word, so a saved layout says what made it by `method` alone. The answers of `layoutByKind` and `layoutAround` keep `version: 1` as it is; their comments now say to read `method`.
131
- - Tests: the layout tests read the profile's own fixture from the pinned distribution (`profiles/layout.fixture.json`); the package's copy, the same bytes, is removed.
132
- - Tests: the layout fixture gains eight cases for `layoutByKind` and `layoutAround` (an index of ten or more ordered by number, an edge with no field, one with no index, an edge to an id that is no node, the same link twice, names where code point order and UTF-16 order differ, a null name beside an absent one, two steps past sixty neighbours) and a suite for the two new readings. No answer of a released function changes.
133
-
134
- ## [4.169.2] - 2026-11-15
135
-
136
- ### Changed
137
-
138
- - **The schema pin moves to `v0.34.1-dist.131`** (canonical 00.34.01 unchanged; no schema file moved). The room profile now says once what simple case folding is: each code point replaced by its mapping in the C and S rows of Unicode's CaseFolding.txt at 15.1, never a language's lower-casing and never full folding, so `MEſſAGE` (two long s) is the word `message` and `Meßage` is not. The room and message profiles say a value that is not an object (`null`, a list, a string) is no element: every reader answers for it as for a body that is not a message, and none fails. The readers here already answered so; no source file moves. The fixtures gain the cases that hold it: room 38 to 48 (the two folds, the Kelvin sign, the dotless and dotted i, a type with a ligature, every White_Space at both ends, three bodies that are no object), message 103 to 120, lens 235 to 242 (the first line and a paragraph separator, and a description that is one separator alone) and calendar 77 to 78 (`v2-except-offset-leap-mid-year`, 229 steps).
139
- - Tests: the message and room readers on a value that is no object.
140
-
141
- - Comments only: an address on an agent link is a label, never matched to an account (the platform, 2026-11-15). `OwFrontDoor.sponsor` and `my_seats`, a roster row's `sponsor_member` and the account members row's `sponsor` no longer say a seat made for an email address names nobody: it answers to the link's maker and is listed in their seats.
142
-
143
- ## [4.169.1] - 2026-11-15
144
-
145
- ### Changed
146
-
147
- - **The schema pin moves to `v0.34.1-dist.130`** (canonical 00.34.01 unchanged; no schema file moved): the message profile says a body that is not a message answers false from the unseen-recipients reading whatever the key, and its fixture gains one case for it (`keyless-non-message-is-not-unseen`), which `hasUnseenRecipients` already answered; the profile writes the keyless readers as the package takes them (`isToEveryone(message, { keyless })`). `hasUnseenRecipients`' description says so.
148
-
149
- ## [4.169.0] - 2026-11-15
150
-
151
- ### Added
152
-
153
- - `readReportLine(line, { world })` and `readReportLines(text, { world })`: a line of a report's `consequences` as the report profile reads it. The line's address is the last one that names an element of this world (`address`, written `ow://<type>/<id>`, with its `reading`, `start` and `end`), and `words` is what a reader shows: the line without its address when nothing the writer wrote is lost (canonical, no moment, no lens, standing apart), else the whole line. `readReportLines` splits a text at every carriage return and line feed and trims each line of White_Space. Both packages are held to the report profile's second fixture, `profiles/report-lines.fixture.json` (101 lines and 16 texts), byte for byte.
154
-
155
- ### Changed
156
-
157
- - **The schema pin moves to `v0.34.1-dist.129`** (canonical 00.34.01 unchanged; no schema file moved): the report profile's rule 10, a line of `consequences` read in three parts, and its second fixture.
158
-
159
- ## [4.168.0] - 2026-11-14
160
-
161
- ### Added
162
-
163
- - `DECLARED_LINKS`: the extension fields a published profile reads as links, each a list of ids in this world (`x_lens_leads_to`, `x_spine_depends_on`), the list `worldGraph` draws edges from. A writer that builds one of these lists names each element once, as the folder format 0.3.24 says of a multi-link, and can read the fields from here rather than copy them. A test holds it to the list schema-dist's own folder lint reads (`LINK_EXTENSIONS`), so a serial that grows the list reaches the package at its repin.
164
-
165
- ## [4.167.1] - 2026-11-14
166
-
167
- ### Fixed
168
-
169
- - `findOwLinks` and `readOwLinkAt` pass over a moment the parser refuses (more than 11 characters, a leading zero, past the 32-bit range) without the parse it would fail, so a line of addresses each carrying a 60-digit moment reads in about 5 ms where it took about 31 (200 KB), and one of moments with a leading zero in about 14 where it took about 103. `parseOwLink` reads a moment through the same function. No answer changes: 200,000 random texts aimed at moments answer the same as 4.167.0.
170
-
171
- ## [4.167.0] - 2026-11-14
172
-
173
- ### Added
174
-
175
- - `layoutByKind(graph)`: a world nobody laid out, laid out from its graph as `worldGraph` answers it. Each kind gets a point on a circle, links pull, neighbours push apart, for 320 seeded steps (160 past 2,000 elements), and the points are fitted into a frame of -1 to 1 on the longer axis. Answers `{ method: 'by-kind', version: 1, points }`, one `{ id, x, y }` per node in node order (`OwLayoutByKind`). The kinds' points on the circle are a table of numbers baked into the package, never computed, so the same graph gives the same points to the last bit in every browser and in the Python package; the shared fixture `test/fixtures/layout/by-kind.json` (12 cases, 3,013 points) is compared exactly. The graph is first put in one order (nodes by id; links by source, field, index, then target, in code point order), so the order a caller hands the nodes and links in moves no point; the points come back in the caller's node order.
176
- - `layoutAround(graph, focus, { steps })`: one element at the centre, its neighbours on rings round it in kind order and then by name in code point order (never by locale), up to 60 on the first ring, and with `steps: 2` their own neighbours further out, near the neighbour that reached them. Answers `{ method: 'around', version: 1, points }`, one `{ id, depth, x, y }` per element within the steps, in the caller's node order, after the same one order as `layoutByKind` (`OwLayoutAround`). A focus that is not a node, a `steps` that is not 1 or 2, or an unknown option is refused with a TypeError.
177
-
178
- ## [4.166.0] - 2026-11-14
179
-
180
- schema-dist v0.34.1-dist.128 (canonical 00.34.01 unchanged; `ONLYWORLDS_VERSION` unchanged, no schema file moved). The fixtures' new cases pass: the lens fixture's 235, the message fixture's 102 (its two `recipients` cases), the address fixture's two `prose-all` cases, the landing fixture's 33.
181
-
182
- ### Changed
183
-
184
- - `worldGraph` reads a multi-link that names one element twice once: one edge, at its first entry, whose `index` is that entry's place in the list as served; the entries after the repeat keep their own places (in `[c1, c1, c2]` the edge to `c1` has index 0 and the edge to `c2` index 2). Schema links and the declared extension links (`x_lens_leads_to`, `x_spine_depends_on`) alike; ids compare as written, so `a` and `A` are two.
185
- - `worldGraph` counts a repeat to an element that is not a node once in `edgesCut`, not once per entry.
186
- - `degreeOf` and `landingOf` follow the graph without a change of their own: a list naming one element twice counts 1 at each end, so an element linked only that way is a near orphan.
187
- - `recipientsOf` answers each recipient once, in its first place; ids compare as written.
188
- - `recipientsOf`, `isAddressedTo`, `isToEveryone` and `hasUnseenRecipients` answer for a message only: any other Narrative, a Story or a Note, has no recipients and is addressed to nobody, whatever its `characters` holds (a Story's `characters` are the people it is about).
189
-
190
- ## [4.165.2] - 2026-11-14
191
-
192
- ### Fixed
193
-
194
- - `findOwLinks` and `readOwLinkAt` pass over an id that follows a word that is not one of the 22 types (`ow://thing/<id>`) without the parse it would fail, so a line of such starts reads in about 4 ms where it took about 63 (200 KB). No answer changes: 100,000 random texts and 714 hand-made shapes answer the same as 4.165.1.
195
-
196
- ## [4.165.1] - 2026-11-14
197
-
198
- ### Fixed
199
-
200
- - `findOwLinks` and `readOwLinkAt` read a start that holds no address faster: an end is tested for an id only where an id's four hyphens stand, so a long line of `ow://` starts and no ids (200 KB) reads in about 5 ms where it took about 8 to 18. No answer changes: 40,000 random texts answer the same as 4.165.0.
201
-
202
- ## [4.165.0] - 2026-11-14
203
-
204
- schema-dist v0.34.1-dist.127 (canonical 00.34.01 unchanged; `ONLYWORLDS_VERSION` unchanged, no schema file moved). The fixtures' new cases pass: the address profile's glued starts (216 cases), the views fixture's own-banner table, three layout-mark cases and four presence cases.
205
-
206
- ### Changed
207
-
208
- - `readOwLinkAt` answers null when the address would begin inside a word: after a letter of any script, a combining mark or an underscore (`elbow://object/<id>` is prose), the same set that cuts a run after it; a hyphen or a digit before `ow://` does not glue. A letter above U+FFFF before the index is read whole.
209
-
210
- ### Added
211
-
212
- - `findOwLinks(text)`: every address in a text, in order, each `{start, end, reading}` (`OwAddressFound`), read by `readOwLinkAt` at each place either prefix starts; none overlap, and a start that reads nothing is passed. Linear in the text: each prefix's next place is searched once, and `readOwLinkAt` now passes over every path end that is not an id without a parse (a refused parse costs an exception), so a megabyte of bare `ow://` reads in tens of milliseconds where it took seconds. No answer of `readOwLinkAt` changes but the start rule's.
213
-
214
- ## [4.164.0] - 2026-11-14
215
-
216
- ### Added
217
-
218
- - `signInWithOnlyWorlds({ scope: 'world:read' })`: a read-only sign-in, for a tool that only reads. The token it gives reads everything the account reads, and every write (a world key minted, a world made, another token minted, an AI call) is refused with 403 `permission_error`, except revoking the token itself. Any other `scope` is refused before anything is kept or sent; left out, the token writes as the account does, as before.
219
- - `completeSignIn` answers `scope`: `'world:read'` when the server granted a read-only token, null when its answer carries none, so a tool that asked for read-only reads what it holds. A read-only sign-in gets no AI door token: `aiUses` beside one is refused before the code is spent, and the same return then completes without it.
220
- - `OwAccountToken` (and `OwMintedAiToken`) declare `scope`, present only on a token a read-only sign-in minted, as `GET /account/tokens` lists it.
221
-
222
- ## [4.163.1] - 2026-11-14
223
-
224
- ### Fixed
225
-
226
- - The descriptions of `isRoomElement` and `isCharterElement` say what 4.163.0 reads: an empty list or an empty object is no mark. They still said any value but null counted, and an editor shows them beside the code. No code changes.
227
-
228
- ## [4.163.0] - 2026-11-14
229
-
230
- ### Changed
231
-
232
- - Schema repinned to schema-dist `v0.34.1-dist.126` (canonical 00.34.01 unchanged, no schema file moved): six profile pages and five fixtures.
233
- - A mark read by presence (`x_room_of`, `x_charter_of`, through `isRoomElement` and `isCharterElement`) reads an empty list or an empty object as absent, as the review mark already does: a compact read leaves both out, so one element read in its full and its compact form must not count as marked in one and unmarked in the other. `false` is not null and still counts. If your code relied on `[]` or `{}` counting as a mark, it no longer does.
234
- - The fixtures' new cases pass as they came: three on each of the room and charter fixtures, seven on the layout profile's empty-string ids and links (a row without a non-empty string id is no element, an empty-string link is null; every reading here already answered so), and four prose cases in the address fixture (after a slug, and an upper-case type word, which names no element: the world's own address).
235
-
236
- ## [4.162.0] - 2026-11-13
237
-
238
- ### Changed
239
-
240
- - Schema repinned to schema-dist `v0.34.1-dist.124`, canonical **00.34.01**: `ONLYWORLDS_VERSION` is now `'00.34.01'`, a published literal type. One description moved and no field: Map's `hierarchy` orders the maps that stand side by side (floors, decks, layers), the maps with the same `parent_map`, without one the maps of the same `location`, without either the maps of the world itself; lower is deeper or earlier, and nesting is said by `parent_map` and `location`, not by the number. No reader in the package changes.
241
-
242
- ## [4.161.0] - 2026-11-13
243
-
244
- ### Added
245
-
246
- - `readOwLinkAt(text, index)`: the address that starts exactly at `index` in a text, found in prose, and where it ends: `{ reading, end }`, or null when no address starts there. After `/w/` the run is read whole, once: exactly an id's form (36 characters, hex in either case, four hyphens) is the world's id, any other run is a slug if it is a legal one, else there is no address; it is never cut back to an id or a shorter slug. A run followed by a letter of any script, a combining mark or an underscore is cut inside a word and is no address. Past its path the address continues only with `@` and a moment and `?lens=` and a lens; any other query key, a fragment or a full stop is prose. `index` and `end` count JavaScript string indexes (UTF-16 code units), so they match `text.slice` and `indexOf`. `parseOwLink` is unchanged and still reads a string whole. Holds the address fixture's 57 prose cases.
247
-
248
- ### Changed
249
-
250
- - Schema repinned to schema-dist `v0.34.0-dist.123` (canonical 00.34.00 unchanged, no schema file moved): times that are read and not ordered, and an address found in prose. Eighty-four new fixture cases across the address, lens, message, record, strand and checks profiles, each passing here (this package has no checks reader).
251
- - `readMarkOf`: a lone mark is the mark. Among several that count, the one with the newest readable `updated_at` is the mark, wherever it stands in the list; one whose time cannot be read never beats one that can. Among marks equal as instants, however each is written, or when none can be read, the higher id by code point, as written, is the mark, never the earlier one in the list.
252
- - The lens readers: an element's clock is the newest readable of its `updated_at`, `server_updated_at` and `local_updated_at`, decided element by element; a value that cannot be read is ignored. The budget orders newest first by that clock, an element with no readable clock comes last of its type, and ties by id go by code point, each id as written. `changed_within` cuts by the same clock and keeps an element with no readable clock rather than cutting it `stale`. A message whose `created_at` cannot be read is cut by the mail window, counted `window`.
253
- - The record readers read `played_at` by the message profile's time grammar: a lower-case `t` and `z`, an offset, no zone and the year 0000 now read; a time without its seconds does not, so a record written without seconds is no record. Records are still ordered to the millisecond.
254
- - `openersFor`: an element whose own id is held in upper case is one no address can name, so its menu now has no rows instead of throwing, and a surface still draws its mark. An id that differs from the address's by more than case still throws, and the empty answer comes after every other check, so a bad tools row is still refused. The shared openers fixture's held-upper case now wants no rows.
255
- - `TOOLS` is baked at the tools list's `aa00969`, which adds the OnlyWorlds mapper: its world and element rows need an account, open a public world without one, and skip map space.
256
-
257
- ## [4.160.3] - 2026-11-13
258
-
259
- ### Changed
260
-
261
- - `list`'s parameters move into its doc comment's first sentence. The docs' reference takes only that sentence, so 4.160.2's wording never reached it.
262
-
263
- ## [4.160.2] - 2026-11-13
264
-
265
- ### Changed
266
-
267
- - `list`'s doc comment names its parameters (`limit`, `cursor`, `filter`, `ids`, `createdBy`, `expand`, `fields`, `viewAs`, `signal`), so the reference built from it shows them.
268
-
269
- ## [4.160.1] - 2026-11-13
270
-
271
- ### Changed
272
-
273
- - The Waymark draws the OnlyWorlds mark in true curves: `MARK_PATH` is the logo redrawn with exact arcs (the ring and parallels circular, the meridians and the W elliptical) in place of the traced outline of short straight lines, filled with `fill-rule="nonzero"`. It scales without faceting; at icon sizes it is the same mark.
274
-
275
- ## [4.160.0] - 2026-11-13
276
-
277
- ### Changed
278
-
279
- - Schema repinned to schema-dist `v0.34.0-dist.122` (canonical 00.34.00 unchanged, no schema file moved). The address profile's "The same id": inside a world an id is written one way, lowercase, and two ids are the same id when they are written the same. A reader reads an id in either case, and takes it in lowercase, only at an address and in a text a profile searches for ids (the charter profile's Charter description). What it is matched against is read as written. The lens key in an address is `lens` exactly. Sixteen new fixture cases across the address, charter, checks, lens, record and report profiles, each passing here. `VIEWS`' conversation layout orders its lines as the message profile orders messages.
280
- - `openersFor` reads the element's own `id` as written, as the same-id rule says. The address's id is still read in either case. An element held in upper case is now refused (`its id differs`) whatever the address's case. An element held in lowercase opens under an address in either case. The shared openers fixture's id-case case is now two: `skips-id-case-held-upper` and `skips-id-case-address-upper`.
281
- - `TOOLS` is baked at the tools list's `53a05e4`, which adds Windrose, a map of one world in the browser. It only reads, so it offers no Open with row.
282
-
283
- ## [4.159.1] - 2026-11-13
284
-
285
- ### Changed
286
-
287
- - Schema repinned to schema-dist `v0.34.0-dist.121` (canonical 00.34.00 unchanged, no schema file moved). The report profile states what rule 9 left unsaid: a parent that is no string, padded, or one id given twice; a thread whose report has a date alone; a ring of reports, which hides no report. The message profile states the whole time grammar, a `created_at` with no offset read as UTC and the years 0000 to 9999 included. `reportThreadsOf` and every message reader already answered each new case as ruled, so no reader changed: the report fixture's 27 thread cases and the message fixture's 93 cases are their own tests.
288
- - `instantOf`'s notes say what a year is: any four of the digits 0 to 9, 0000 among them, on the Gregorian calendar run on in both directions, so 0000 is a leap year and an offset may carry an instant past either end.
289
- - Comments only: the generated types' header and the codegen README name every server-managed field on the shared base (`doors` and `hidden` were left out of the header's list, `doors` and `x_served_trimmed` out of the README's).
290
-
291
- ## [4.159.0] - 2026-11-13
292
-
293
- ### Added
294
-
295
- - **`instantOf(value)` and `compareInstants(a, b)`: the time reader the package's readers sort by, exported.** `instantOf` gives the instant a `created_at` (or `updated_at`) names, `{ seconds, microseconds }`, or `null` when it cannot be read. It follows the message profile's grammar: the whole value `YYYY-MM-DDTHH:MM:SS`, an optional fraction, then `Z`, `+HH:MM`, `-HH:MM` or nothing (read as UTC), `T` and `Z` in either case, each part in its range, and a fraction counted to its sixth digit, later digits dropped. A date alone, a space for the `T`, a time without seconds, an offset without its colon and a trailing line break are not read. `compareInstants` orders two of them earlier first, with `null` after every readable one and 0 for equal instants, so a stable sort keeps the order given. Both are the very function the message, report, charter and lens readers order by, so a tool that sorts with them orders exactly as those readers do.
296
-
297
- ## [4.158.1] - 2026-11-13
298
-
299
- ### Changed
300
-
301
- - **`TOOLS` carries Subs.** The tools list is repinned at `0a7539c` (from `9452529`): Subs, a convoy-escort game whose Character captains the lead escort, is a listed row with an element door for a Character (`needs: "account"`; it skips room, hidden and proposed elements), so `openersFor` offers it beside a Character, and on no other element and no world. The Python package bakes the same list.
302
-
303
- ## [4.158.0] - 2026-11-12
304
-
305
- ### Added
306
-
307
- - **`reportThreadsOf(narratives)`**, the report profile's rule 9 (schema-dist serial 120): reports grouped into threads by the `parent_narrative` they name, read as written. For each report, what its parent is (`none`, `report`, `board`, `missing`, `other` or `self`; `none` when the field is absent, null, no string or empty once trimmed, and a parent with content read as written); only reports are read, so a message that names a report is in no part of the answer; one group per parent that is a report (its thread) or a board entry (the talks under it), its reports ordered by `created_at` as messages are (an instant; equal times keep the list's order; unreadable times last); and a lint on a report that names itself or whose parent names another report the reader holds (a report that names itself reads as naming no parent, so one under it is no chain). A report and a board entry are each their pair and the room's mark; the reader never walks further up a chain and never orders by id. Its types `OwReportThreads`, `OwReportGroup`, `OwReportLint` and `OwReportParentKind`. The profile's fixture, 20 thread cases, is its own test, with edges the fixture leaves out.
308
-
309
- ### Changed
310
-
311
- - Schema repinned to schema-dist `v0.34.0-dist.120` (canonical 00.34.00 unchanged, no schema file moved, `views.json` still version 3). The report profile: a report that continues earlier work names the first report of that work in `parent_narrative`, and rule 9 tells a report's thread from a closed talk's by the parent's kind, a report or a board entry; its first fixture holds 20 thread cases, all read by `reportThreadsOf`. The views fixture's two references into the report page move down one line; its cases are replayed as before.
312
- - The list filter's notes name the Pin list's `element=<id>` (served since 2026-11-12): `list('pin', { filter: { element: id }, expand: ['map'] })` gives an element its maps in one call, and `element: ''` lists the Pins placing nothing. It was already sent as given; only the words are new, and the unknown-key hint names a single link and a Pin's element.
313
-
314
- ## [4.157.2] - 2026-11-12
315
-
316
- ### Changed
317
-
318
- - Schema repinned to schema-dist `v0.34.0-dist.119` (canonical 00.34.00 unchanged, no schema file moved, `views.json` still version 3). The map-space profile gains rule 11: a layout map is never a candidate when a tool picks the Map that stands for the world's space, and a Pin or a Marker on a layout map is a place in that drawing, never in the world. `mapSpace` and `layoutOf` answer as before: they still read a layout map, its Pins and its Markers, and the rule is on the tool that picks or places from their answers. The fixture gains two layout cases (23) and one map-space case (35), all read the same by this release. Reading H now says a row's id is a non-empty string, as both readers already read it. The folder lint the distribution ships knows folder format 0.3.22.
319
-
320
- ## [4.157.1] - 2026-11-11
321
-
322
- ### Changed
323
-
324
- - Schema repinned to schema-dist `v0.34.0-dist.118` (canonical 00.34.00 unchanged, no schema file moved, `views.json` still version 3). The views table's `conversation` layout now ends with the case of a tool that shows one element and no thread: it draws that element as a document, and the layout asks nothing of it; `VIEWS_TABLE.layouts.conversation` carries the new words, and no answer of `viewOf` moved. The checks profile gains two rules with eight cases (a work item and a report carry the room's mark, read by `present`) and now says that `present` trims by the room profile's White_Space set; the room profile says a sitting's own mark holds the world's id and that a sitting names its winning spec in `x_star`. The package has no checks reader, and `isRoomElement` reads as before.
325
-
326
- ## [4.157.0] - 2026-11-11
327
-
328
- ### Changed
329
-
330
- - **Six answer fields admit `null`, as the API document says they may**: a `/changes` upsert row's `updated_at` and a delete row's `deleted_at` (`OwChangeUpsert`, `OwChangeDelete`), a `/bulk` slot's `id`, `created_at` and `updated_at` (`OwBulkItemResult`: a failed slot carries no id), and a member row's `x_served_trimmed` (`OwMemberLenses`). The SDK typed them as never null; nothing at runtime changes, but code that reads one as a plain string now has to handle `null`.
331
-
332
- ## [4.156.0] - 2026-11-11
333
-
334
- ### Changed
335
-
336
- - **`OwProposedNewElement.evidence` and `.expires` are `unknown`**: the world's queue serves a new element's review mark's `evidence` and `expires` as stored, whatever JSON value they hold (a list of ids, a number, an object), where it served a non-string as its text before. A string still arrives as written and null as null; anything else arrives as itself. Code that read either as a string narrows first (`typeof row.evidence === 'string'`). Whether a row has lapsed is its `state`; an `expires` that is not a date or a date-time never lapses.
337
-
338
- ## [4.155.1] - 2026-11-11
339
-
340
- ### Changed
341
-
342
- - Schema repinned to schema-dist `v0.34.0-dist.117` (canonical 00.34.00 unchanged, no schema file moved, `views.json` still version 3). The views table's own description now says that a tool's own words for a heading or a fold are its drawing, and that a fold's label is the word a tool uses where it has none of its own; `VIEWS` is baked without that description, so no value it holds moved. The views fixture gains a 46th case, a Pin whose generic link has one key holding and the other White_Space alone, which holds nothing; `viewOf` already read it so, and the replayed fixture now holds it.
343
-
344
- ## [4.155.0] - 2026-11-11
345
-
346
- ### Changed
347
-
348
- - **`proposedMarkOf`'s `by` is read as stored, never trimmed** (the review profile's Who `by` names, schema-dist serial 116): a `by` that is not a string, or is White_Space alone, names nobody (`null`), and White_Space around a word or an id makes it neither. 4.154.0 trimmed it, so `" owner "` came back as `owner`. `OwProposedMark.by` says how a reader shows it: as the mark's word ("the mark names Ada"), never as a fact ("proposed by Ada"); a name in its place only where the platform resolves it to a member on its roster or to the owner, comparing ids as ids, or where it is the id of a Character of the same world this reader is served, said as that Character; otherwise the raw value, apart.
349
- - **`OwProposedNewElement.by`'s words**: the mark's own `by`, the writer's word for who proposed it, a claim, null when it names nobody; never a record of who wrote the element or the mark.
350
- - **Schema repinned to schema-dist v0.34.0-dist.116** (from dist.115; same canonical 00.34.00, no schema file and no fixture moved; only `profiles/review.md`).
351
-
352
- ## [4.154.0] - 2026-11-11
353
-
354
- ### Added
355
-
356
- - **`OwFileAddedBy.owner`**: true when the world's current owner added the file, through any door (the site, the API, or an agent with the owner's own keys; a seat the owner sponsors is the seat's, false); false for everyone else, on every file of a world with no owner, and on a past owner's files once the world has another owner. onlyworlds.com serves it on every row of `listFiles` and `worldFiles`. `membership` is null both for the owner's files and for a file whose membership no longer exists, and `owner` tells the two apart: a null `membership` with `owner` false is a former member, and such a file's `name` is "a former member" (it was an empty string for a member whose account was deleted).
357
-
358
- ## [4.153.0] - 2026-11-11
359
-
360
- ### Changed
361
-
362
- - **A lens's `first_line` row carries the first line that holds something**: the description is split at every carriage return and every line feed, and its first line with something left after trimming White_Space is the row's `description`, carried as stored with its own spaces. A description with no such line (empty, blank lines, White_Space alone) gets no `description` key. 4.152.0 cut at the first line break, so a description opening with a blank line gave an empty `description`, and one of White_Space alone was carried as it stood.
363
- - **Schema repinned to schema-dist v0.34.0-dist.115** (from dist.114; same canonical 00.34.00, no schema file moved): the lens fixture at 220 cases (ten new, all first-line rows), the charter fixture at 31 (four new), `views.json` version 3 and its 45 cases, all replayed.
364
- - **`VIEWS.version` is `3`**, the version of `views.json` baked at this release; codegen reads version 3 only, so a build on a version 2 file stops rather than drawing a Pin by the old rule.
365
- - **`viewOf`'s `order` draws a Pin's `element`**: a generic link is one field of the order, under its schema name and at its place (a Pin's Details: `map`, `element`, `x`, `y`, `z`), read as the two keys a served Pin carries (`element_type`, `element_id`, from the walk); it is drawn only when both are strings with something left after trimming. 4.151.0 never drew it, since no key `element` is served.
366
-
367
- ## [4.152.0] - 2026-11-11
368
-
369
- ### Added
370
-
371
- - **`OwApiError.isReviewByAnother`**: true for 403 `review_by_another`. A write from a key below co-builder (a contributor's or a guest's, a person's or an AI seat's) that sets an `x_review_proposed` or `x_review_checked` mark is refused when the mark's `by` is not the writer's own membership id and the mark is not exactly the one the element holds; `param` names the mark, and nothing is written. Send your own membership id as `by`, or the mark as the element holds it. A Character's id is refused too.
372
-
373
- ## [4.151.0] - 2026-11-10
374
-
375
- ### Added
376
-
377
- - **`viewOf` returns `order`: the fields an element's page draws, in the order drawn** (`views.json` version 2): the row's `lead`, then the rest of the type's section fields in the schema's own order, then each `fold`; each field once, and only a field that holds something (not absent, null, a string with nothing after trimming, an empty list or an empty object), so an empty lead field is skipped and an empty fold draws no fold. The base fields are the page's head and appear only where a row names one. With no row, the type's section fields.
378
-
379
- ### Changed
380
-
381
- - **The `x_served_trimmed` banner shows only when the list names a field**: at least one string with something left after trimming. A list of blanks, or of anything but strings, shows none (4.147.0 showed it for any non-empty list).
382
- - **`VIEWS.version` is `2`**, the version of `views.json` baked at this release; codegen reads version 2 only.
383
- - Where two rows match on one rung, the first in the table wins and the page is drawn, as before; the file now says so, and the fixture holds the case.
384
- - **A lens's short rows carry `supertype` and `subtype`**: a `name` or `first_line` row from `resolveLens` now holds each where the element as served holds it as a string with something left after trimming White_Space, as stored (never trimmed); absent, null, empty, White_Space alone or a value that is not a string adds no key. A known-of stub's row carries them too, since a stub is served both. Rows under `full` are unchanged.
385
- - **Schema repinned to schema-dist v0.34.0-dist.114** (from dist.113; same canonical 00.34.00, no schema file moved): `views.json` version 2 and its 41 cases; the lens fixture at 210 cases (nine for the short rows), all replayed; the review profile's rule 5 (an object with at least one key is a mark, whatever its keys: `proposedMarkOf` already read it so, and its tests now name the rule's two examples); the report profile's rule 8 (no reader here reads a report's `consequences`).
386
-
387
- ## [4.150.0] - 2026-11-10
388
-
389
- ### Added
390
-
391
- - **`search(q, { type, limit, viewAs })`**: the world's elements whose name or text fields hold every word of `q` as a whole word, ignoring case, as `GET /search` finds them (`OwSearchResults`: `query`, `results`, `count`, `truncated`; each `OwSearchResult` with its `type`, `id`, `name`, `supertype`, `subtype`, the `field` the words were found in, a `snippet` of it, and `known_of` on a row the reader knows of without seeing it whole). A word is a run of letters, digits or underscores, so `APP-142` is searched as two words; a snippet is a piece of one field, so read the element to keep only those that hold a key whole. No pages: `truncated` says more matched than `limit` (1 to 200) let through. On a client with `world` set it searches that world, like the other world reads. An unknown option or a `q` that is not a string sends nothing.
392
-
393
- ## [4.149.0] - 2026-11-10
394
-
395
- ### Changed
396
-
397
- - **`lenses_from` can be `none`** on `OwMe` and `OwMember`: a member or seat that never chose its lenses and that nothing brings a lens to (no Character, or Titles that name none), told apart from `chosen`, which an empty list chosen on purpose still reads. What a row reads is unchanged; the owner's row and a removed row are always `chosen`. A TypeScript switch over the field gains a case.
398
-
399
- ## [4.148.0] - 2026-11-10
400
-
401
- ### Added
402
-
403
- - **A file on an element can be an image**: `OwFileKind` gains `png`, `jpg`, `webp` and `avif`, and `FILE_EXTENSIONS` (so `fileKindOf`, `createFileTicket` and `uploadFile`) reads `.png`, `.jpg`, `.jpeg` (kind `jpg`), `.webp` and `.avif`, as onlyworlds.com now takes them. `uploadFile(type, id, file)` with `screen.png` sends instead of refusing before the request. An image file is private to the element's readers like every file, and is not the element's picture (`image_url`). `.svg` and `.gif` are still refused before sending.
404
-
405
- ## [4.147.0] - 2026-11-10
406
-
407
- ### Added
408
-
409
- - **`viewOf(element)` and `VIEWS`: how an element is shown by its kind**, from the distribution's new `views.json` (presentation, not the standard). `viewOf` returns the row that matches (a profile's pair with its mark, by that profile's own reader, then a type with a supertype, then a type alone; types and supertypes trimmed and folded as the room profile defines), the one sentence for a card, a list row or a hover (the first field the row names whose value is a string with something in it, trimmed, cut at its first line break, trimmed again), and the marks to show at the head of its page, in order (`known_of` and `hidden` when true, `x_served_trimmed` when it names a field, the review marks when an object with something in it, then the row's own). An element no row matches has no row and draws as its type; the banners still show. Today's rows: a report, a charter, a message and a line in a room. `VIEWS` is the table baked at this release, frozen; codegen refuses a row column this reader does not read, a pair its profile's reader does not match, a field its type lacks, an unknown layout and two rows holding one kind. The fixture's 28 cases are replayed.
410
-
411
- ### Changed
412
-
413
- - **Schema repinned to schema-dist v0.34.0-dist.113** (from dist.112; same canonical 00.34.00, no schema file moved): `views.json` and `views.fixture.json` new; the room profile says its pairs' order carries no meaning (no pair moved); two examples reworded. The room and checks fixtures are replayed unchanged.
414
- - `lensesInUse`'s notes say what a null membership means, in the lens profile's words since serial 112: there is no membership (a folder world), and then the Character's `x_lens_uses` is read. Behaviour unchanged.
415
-
416
- ## [4.146.0] - 2026-11-10
417
-
418
- ### Changed
419
-
420
- - **Schema repinned to schema-dist v0.34.0-dist.112** (from dist.111; same canonical 00.34.00, no schema file moved). `ROOM_PAIRS` gains the room profile's tenth pair, `narrative`/`Report`: a report (the new report profile, `profiles/report.md`: what a session or a closed talk leaves behind, a Narrative with supertype `Report` and the room's mark) is a room element, so `isRoomElement` counts a marked report and every reader that leaves the room out of a lore view leaves it out too (`ask`, `ai`, the lens readers unless a lens's `types` names `narrative/report`, the Open with menu's `room` mark). A Narrative with supertype `Report` and no mark stays lore. The room fixture's 35 cases (3 new) and the lens fixture's 201 are replayed.
421
-
422
- ## [4.145.0] - 2026-11-10
423
-
424
- ### Added
425
-
426
- - `err.refusals` on an `OwApiError` that `isAnswerNotParsed`: when an AI use the server reads itself (`subs-captain`, `lens`) gives up with 422 `answer_not_parsed`, each answer it refused, in order, as an `OwAiRefusal` (`word`, the provider's `stop_reason`, `output_tokens`). An answer cut off at `max_tokens` now reads apart from one that finished in the wrong shape. Never the answer's text; a replay under the same idempotency key carries the same list. Undefined for any other error or when none is sent; a row not in that shape is left out.
427
- - `OwAiRefusal`, exported, and `refusals` on `OwErrorBody`, as the API document declares the error envelope (a `/bulk` slot's error included).
428
-
429
- ## [4.144.0] - 2026-11-10
430
-
431
- ### Added
432
-
433
- - `OwAiUse.can_pay`: true when a call by this account would be paid now, by the same check the call runs (`on`, the server's daily cap, today's token rating against `rating_need`, or a saved AI key that runs the use). Offer a use only when it is true. The server's fifth use, `lens` (a lens proposed from a sentence, served switched off), is called through `ai('lens', { world_id, sentence })` and answers as served; it is not typed yet.
434
-
435
- ### Changed
436
-
437
- - Schema repinned to schema-dist v0.34.0-dist.111 (canonical 00.34.00 unchanged, no schema file moved): the map-space fixture gains four layout cases, 21 in all: an id above U+FFFF beside one in U+E000..U+FFFF, a repeated Pin id where the last row wins, a Pin row without a string id, and a Map row without a string id; the page says reading H holds for the Map row too. `layoutOf` already answered all four.
438
-
439
- ### Tests
440
-
441
- - The layout cases are replayed by their `map` id, or, for a Map row with no string id, at each position the case's `map_index` lists, the answer the case's `want` for each; a case with neither fails by name. The count is held at 21.
442
-
443
- ## [4.143.0] - 2026-11-10
444
-
445
- ### Changed
446
-
447
- - `TRAIL_TIERS`, `trailByPlace`'s default floors: each default tier is named by its own kinds: `location, object, zone` at 0, `character, collective, creature, species` at 240, `event` at 480, and the other eleven kinds at 720. The heights and the kinds are unchanged, so every point is the same; what moves is the text a tier's name reaches, each hanging Pin's `by` (`tier event; 3 of 7 on the spiral round it`) and the recipe's `settings.tiers`. The old names were group names for sets of kinds that no source holds. A caller that passes its own `tiers` sees no change.
448
-
449
- ### Tests
450
-
451
- - The shared trail fixture moves with it (the same bytes in both packages, its sha held in each); its main input passes these tiers, and the hero case runs on the default.
452
-
453
- ## [4.142.1] - 2026-11-10
454
-
455
- ### Fixed
456
-
457
- - `mintAiToken`'s `name` (at most 255) and `ai`'s `idempotencyKey` (1 to 200) are counted in code points, as the server counts them, not in UTF-16 units. A name of 255 characters outside the Basic Multilingual Plane (an emoji, say) was refused before sending, though the server takes it.
458
-
459
- ### Tests
460
-
461
- - Both caps are held at their edge with characters above U+FFFF: 255 pass and 256 are refused for the name, 201 are refused for the key.
462
-
463
- ## [4.142.0] - 2026-11-10
464
-
465
- ### Added
466
-
467
- - `layoutOf(map, maps, pins)` and its answer type `OwLayoutOf`: the map-space profile's layout reading (rules 8 to 10) for one Map, `{ layout, ground, hand }`. `layout` is `x_layout.method` when `x_layout` is an object whose `method` is a non-empty string (`"hand"` for a board; a method no page names answers itself), else null. `ground`, only on a layout map, is the id in `x_layout.settings.ground.map` when it names a Map in `maps` other than this one, else null; its size is not returned. `hand` is the ids, in code point order, of this Map's Pins a new lay-out keeps: every Pin on a board, the Pins whose `x_hand` is exactly `true` on any other layout map, none on a plain Map. `pins` may hold every Pin the caller has: the reader keeps those whose `map` names this Map, reads a row without an id as no Pin and the last row served for an id, and counts a Pin whose element it cannot see. The method's own settings are never returned. `mapSpace` is unchanged.
468
-
469
- ### Tests
470
-
471
- - The layout reading replays all 17 of the fixture's `layout_cases`, whole answers in key order, the count held; edge tests beyond the fixture hold code point order (ids above U+FFFF), the row readings and a ground row without an id. `layoutOf` joins the tree-shaking checks.
472
-
473
- ### Changed
474
-
475
- - Schema repinned to schema-dist v0.34.0-dist.110 (canonical 00.34.00 unchanged, no schema file moved): the map-space profile gains rules 8 to 10, `x_layout` on a Map, its ground and `x_hand` on a Pin, and a layout reading with its own fixture section, `layout_cases`. The map-space reading and its 34 cases do not change.
476
-
477
- ## [4.141.0] - 2026-11-10
478
-
479
- ### Added
480
-
481
- - `OwSignInError.param`: the field the server named when it refused the code exchange, such as `ai_uses` or `ai_lifetime_minutes` on a 422 for an AI door token asked with `completeSignIn`, so that refusal can be told from any other 422 there.
482
-
483
- ## [4.140.0] - 2026-11-10
484
-
485
- ### Added
486
-
487
- - `OwAiUse.rating_need`: the rating tokens a call to the use needs left today to run, the output cap or the use's worst case when it is larger. `provider` names `anthropic` or `jev`.
488
- - `err.isAnswerNotParsed`: 422 `answer_not_parsed` (param `answer`), a use whose model answered off its promised shape twice; billed to the use's own budget, not the token rating.
489
- - `ai()`'s notes: a refusal the provider billed (`answer_not_parsed`, or a 502 after a billed answer) is kept under the idempotency key and replayed like an answer.
490
-
491
- ### Tests
492
-
493
- - A layout-only bundle (`trailByPlace`, `mapSpace`) carries none of the zip library, with `readWorldFile` as the control.
494
-
495
- ## [4.139.0] - 2026-11-10
496
-
497
- ### Deprecated
498
-
499
- - `ask()` and `sort()`, in favour of `ai('ask', body)` and `ai('sort', body)` (since 4.137.0): the same answers, with the use's meter (`use_remaining`) and an idempotency key, under which a retry is answered from the first call and billed nothing. The body takes the wire's spelling: `world_id` where these took `worldId`. Both keep working, unchanged, until 5.0.
500
-
501
- ## [4.138.0] - 2026-11-10
502
-
503
- ### Added
504
-
505
- - An AI door token: an `ow_a_` token that opens only `aiUses()` and the AI uses it names, as the person and on their token rating, until midnight UTC at the latest. Another use answers 403 `use_not_granted`, every other route 401 `wrong_audience`.
506
- - `completeSignIn({ ..., aiUses, aiLifetimeMinutes })` asks the sign-in's code exchange for one beside the account token; the result's new `aiToken` (`OwAiToken`: `token`, `tokenId`, `tokenType`, `audience`, `aiUses`, `expiresAt`) carries it, and is null when none was asked or the server answered none. The exchange spends its code whatever it answers, so a list the SDK can see is wrong (not 1 to 8 names, a name that is not one path segment, a lifetime outside 1 to 1440 minutes or without `aiUses`) is refused before anything is sent and before the pending sign-in is touched. A name the server does not run is its 422, and the code is then spent. Both options are optional, and every field the result had stays as it was.
507
- - `mintAiToken(uses, { lifetimeMinutes, name })`: `POST /account/tokens` with `ai_uses`, trading an account token down to an AI door token (`OwMintedAiToken`, the token's row with the token once). It belongs to the token that minted it: revoking that one revokes it too. The uses are required, so a call without them never reaches the route that would mint a full account token.
508
- - `OwApiError.isUseNotGranted` (403 `use_not_granted`) and `isWrongAudience` (401 `wrong_audience`).
509
- - `OwAccountToken.ai_uses`, on an `ai` token's row.
510
-
511
- ## [4.137.0] - 2026-11-09
512
-
513
- ### Added
514
-
515
- - `aiUses()`: `GET /account/ai`, the AI uses onlyworlds.com runs, each a named call a tool makes: its owner, provider, model and effort, the hash of its system prompt, what its body may hold, its answer cap, its price, its daily ceiling, this account's calls left today, and whether it is on (`OwAiUse`). Reading it spends nothing. Needs an account token.
516
- - `ai(use, body, { idempotencyKey, signal })`: `POST /account/ai/{use}`, one call of a listed use, metered on the account's token rating and then its own saved AI key, as `ask()` and `sort()` are. The body is the use's own, in its wire spelling, sent as given (`{world_id, question, context}` for `ask`, `{world_id, notes, slice}` for `sort`); typed for those two (`OwAiAskBody`, `OwAiSortBody`, answered as `OwAiAskAnswer` and `OwAiSortAnswer`) and open for any other use. The answer adds `use_remaining` (null when the use has no daily ceiling) and `wasReplay`, read from `Idempotent-Replay`. The use's name is checked as one path segment and the key as 1 to 200 characters before anything is sent; the client's name goes as `X-OW-Client`.
517
- - With an `idempotencyKey`, a retry is safe: the same key and body answer the first answer again and bill nothing. A key whose first call is still running answers 409 `idempotency_in_flight`; retry with the same key and body a few times at most, and never mint a new key to get past it, since a key whose answer could not be stored stays in flight for a day although the call ran and was billed. The client never retries on its own.
518
- - `OwApiError.isUseCeiling` (429 `use_ceiling`: the use's calls for the day are made) and `isUseInFlight` (429 `use_in_flight`: another call to this use is running; wait `retryAfter`). A 503 `ask_unavailable` with `param` `use` is still `isAskUnavailable`.
519
- - `ask()` and `sort()` keep calling their own routes, unchanged.
520
-
521
- ## [4.136.0] - 2026-11-09
522
-
523
- ### Added
524
-
525
- - `trailByPlace`: the ground Map's element id, `ground.map`, travels in the recipe: given in the input, the recipe's `settings.ground` carries it back as `{map, width, height}`, so a tool reading a laid-out Map finds the picture its pins stand on. Optional; a ground given without it carries none. A `map` that is not a non-empty string is refused. The shared fixture's whole-world case now names its ground map. The type is `OwTrailGround`.
526
-
527
- ## [4.135.2] - 2026-11-09
528
-
529
- ### Fixed
530
-
531
- - A tool that never calls `trailByPlace` no longer carries its default floors. `TRAIL_TIERS` was built by a top-level call a bundler could not drop, so 4.135.0 and 4.135.1 added about 450 bytes to every bundle. Its freezes are now marked pure, and a bundle that imports only `mapSpace`, `worldGraph` or `OwV2Client` comes out the size it was on 4.134.1. The table is the same, frozen as before.
532
-
533
- ## [4.135.1] - 2026-11-09
534
-
535
- ### Fixed
536
-
537
- - `trailByPlace` no longer runs forever when a row of the frame has no free point left: more Pins on one row (one y and z) than the frame is wide plus one. It is refused with a TypeError naming the element, its row and the frame's width. A taken point tries every x on its row first, as before. The shared fixture gains the boundary pair (two on one point in a frame 1 wide fit; a third is refused) and a pin given as whole floats, whose words read as whole numbers.
538
-
539
- ## [4.135.0] - 2026-11-09
540
-
541
- ### Added
542
-
543
- - `trailByPlace(input)`: the first layout method, a world laid out along a route on a ground map, answered as a layout in the standard's own fields: a Map frame (`width`, `height`, `depth`), one Pin per element (whole numbers in the Map's pixels, which `mapSpace` reads back), the route, and the recipe (`method: 'trail-by-place'` and the settings it ran with, so the same layout runs again from the Map alone). The places come in travel order, each at its own pin on the ground map plus a margin, evenly between two pinned places, or on a ring inside a parent place. Every other element hangs on one spiral round the place it belongs to (the mean of its places' points), the hero first, then by floor and id, at its floor's height. Nothing shares a place's own point, and no two Pins share a point: a taken point moves along x from its own point, one pixel right, one left, two right, and on, by element id order. Each Pin carries `by`, the caller's words followed by the method's own. The method places and does not read: which elements are in, the travel order and what hangs where are the caller's. Points round half to even. An input it cannot place is a TypeError naming what, an unknown key at any level included. `TRAIL_TIERS` is the default floors; `roundHalfEven` the rounding.
544
-
545
- ## [4.134.1] - 2026-11-08
546
-
547
- ### Fixed
548
-
549
- - `OwWorldKeyRow` and `OwWorldKeyMember` are exported. 4.134.0 named them in `madeWorldKey`'s signature and in its notes, but left them out of the package's export list, so `import { type OwWorldKeyRow }` failed to compile. A test now reads each built entry's type declarations against its export list.
550
- - `mintWorldKey`'s comment now says to mint again with a new idempotency key after a lost answer. The same key sent again makes a second key, and `madeWorldKey` keeps naming the first, so the second could never be found. Words only.
551
-
552
- ## [4.134.0] - 2026-11-08
553
-
554
- ### Added
555
-
556
- - `madeWorldKey(worldId, idempotencyKey, { signal })`: the key a `mintWorldKey` call sent with that idempotency key made, as its row without the key itself (`OwWorldKeyRow`: id, name, last four, scope, last used, revoked, kept, and `member` on a member-bound key), or null when it made none. A revoked key still answers, with `revoked: true`. Null too from a server before the lookup, which answers that address 405.
557
- - `mintWorldKey` takes `idempotencyKey`, sent as the Idempotency-Key header, and `signal`. The mint is never replayed (a stored answer would keep the key), so a second mint with the same key makes a second key. The server remembers the idempotency key on the row instead. When an answer is lost: keep the idempotency key on the device before the send, ask `madeWorldKey` after a network error or a timeout, revoke the id it names, and mint again. A 4xx made no key, and "none made" is worth believing only a couple of minutes after the send. An `idempotencyKey` that is not a non-empty string is a TypeError before anything is sent.
558
- - `OwWorldKeyMember`, the membership a member-bound key is bound to; `OwMintedKey.member` is typed with it (it was a plain record).
559
-
560
- ## [4.133.2] - 2026-11-08
561
-
562
- ### Changed
563
-
564
- - Schema repinned to schema-dist `v0.34.0-dist.109` (canonical 00.34.00 unchanged, no field moved). The calendar profile now refuses an era whose `year_base` is present and not 0 or 1, as its field table always said, and its fixture gains `refuse-year-base-five` (77 cases, 43 refusals). The calendar reader already refused such an era; the replayed fixture now holds it.
565
-
566
- ### Fixed
567
-
568
- - `mintWorldKey`'s comment said that without `keep`, revoking the token takes the keys it minted. That holds for a token a tool got through Sign in with OnlyWorlds; a token from `signInWithPassword` or one made by hand never takes its keys, as `revokeAccountToken`'s comment already said. Words only.
569
- - `createAccountWorld`'s comments said an account with a PIN must leave `pin` out (a 422) and that the route takes no Idempotency-Key. Since 2026-11-07 the server checks a `pin` sent by an account that has one before it makes the world (a wrong one is a 401 with `param: 'pin'`, nothing made) and takes an Idempotency-Key, whose replay answers the world without its key. The call is unchanged: it sends `pin` as given and no Idempotency-Key. Words only.
570
-
571
- ## [4.133.1] - 2026-11-06
572
-
573
- ### Changed
574
-
575
- - `TOOLS` is the tools list at `9452529`: Atlas's row now says `opens_public: true` (its links open a public world with nothing signed in, and `needs` stays `account` for a private one), so `openersFor` with `TOOLS` offers Atlas as a live row on a world's public page.
576
-
577
- ## [4.133.0] - 2026-11-06
578
-
579
- ### Added
580
-
581
- - `openersFor` reads a tools-list row's new `open.opens_public`: `true` when the tool's links open a public world for someone who is not signed in, although its `needs` is `account` for a private one. On a world's public page (the `public` audience) such a row is a live row with no word, instead of "Needs an account"; anywhere else it reads as its `needs` says, since only a public page knows the world is public. Absent or `false` reads as before, so a list without the key gives the same rows. The Waymark draws the row live there. `OwToolOpen` types the key.
582
- - `openersFor` refuses an `opens_public` that is not `true` or `false`, and `true` beside a `needs` other than `account` (a row that needs nothing already opens a public world, and a row that needs the device never reaches a public page), as the tools list's own check does. Seven new cases in the shared openers fixture (110 in all, the same bytes as the Python package's).
583
-
584
- ## [4.132.4] - 2026-11-05
585
-
586
- ### Changed
587
-
588
- - The schema pin moves to schema-dist `v0.34.0-dist.108` (canonical `00.34.00` unchanged, so `ONLYWORLDS_VERSION` is the same). Three description fixes across 19 fields, no field added or removed: `Event.constructs` says concepts, systems or built entities (no longer laws); the seventeen time fields say in one phrasing that they count in the World's `time_basic_unit`; `Map.hierarchy` says it is an integer to order or group the maps that share a parent, read as each tool chooses; three map descriptions lose a trailing space. The generated types' field comments and `SCHEMA.md` carry the new words.
589
-
590
- ## [4.132.3] - 2026-11-04
591
-
592
- ### Changed
593
-
594
- - `openersFor`'s refusal of a bad `element_types` lists the 22 type words in the standard's own order, as the Python package and the game SDKs do (the message only; what is refused is unchanged).
595
- - `me()` has a doc comment: what `GET /me` answers, that the client's PIN is sent there (the connection test), and that an account token is refused before sending.
596
- - The doc comments of `agent`, `setMyLenses`, `setMemberLenses`, `createWorldLink`, `members`, `ask` and `allowance` name the server as onlyworlds.com or the server.
597
-
598
- ### Tests
599
-
600
- - The openers fixture gains three refusals of `element_types` (103 cases): a type word given twice, a word in another case, and a word that is no string. The same bytes as the Python package's copy.
601
-
602
- ## [4.132.2] - 2026-11-03
603
-
604
- ### Changed
605
-
606
- - The schema pin moves to schema-dist `v0.34.0-dist.107` (canonical `00.34.00` unchanged, so `ONLYWORLDS_VERSION` is the same). Two profile sentences changed and no fixture: the charter profile names what open means to a public reader (the World's `public_read`), and the map-space profile's whole-number range sentence is dated. No reader changed.
607
-
608
- ## [4.132.1] - 2026-11-03
609
-
610
- ### Changed
611
-
612
- - **`TOOLS` repinned at the tools list's `60a590e`: Tactical Tangle opens a world or a Character.** Its row gains a world door and an element door (`needs: "account"`, `element_types: ["character"]`, skipping rooms, hidden and proposed elements), so `openersFor` now offers it on a world's menu and on a Character's, and on no other type. No code changed.
613
-
614
- ## [4.132.0] - 2026-11-03
615
-
616
- ### Added
617
-
618
- - **A tool that opens one kind of element is offered only on that kind.** The tools list's `open` gains `element_types`, the element types its element door opens (type words such as `character`), or `null` for every type. On an element's menu, `openersFor` leaves out a row whose `element_types` does not hold the address's type word, whether or not the row has a world door: the tool opens other kinds, so its world is not offered from this element either. The type is read from the address, so no element body is needed; a world's menu reads no element types. `OwToolOpen.element_types` is typed. Refused by name, as a row that is not one: a word outside the 22, an empty list, a repeat, and `element_types` on a row with no element door. A list without the key reads as before.
619
- - `TOOLS` repinned at the tools list's `d0e52c4`, where every row carries `element_types: null`: no menu changes. The shared openers fixture has 100 cases (from 92), the same bytes as the Python package's.
620
-
621
- ### Tests
622
-
623
- - The shared openers fixture's 24 type refusals each name their kind beside the name: `unknown_key` (a key the object does not take, refused by name) or `bad_value` (a key it takes whose value it refuses). Each package's test tells the two apart by its own message, so a reader that refuses a known key as unknown, or the other way round, fails the case where before the name alone passed it. The same bytes in both packages; nothing shipped changes.
624
-
625
- ## [4.131.0] - 2026-11-03
626
-
627
- ### Added
628
-
629
- - **Your own AI key.** An account can now save its own Anthropic or OpenAI key on its account page; once the day's token rating is spent, `ask` and `sort` run on that key, billed by its provider. `OwAskAnswer` and `OwSortAnswer` carry `paid_by` (`OwPaidBy`: `'rating'` or `'own_key'`), so a tool can say which paid. `OwAllowance` carries `own_key`: `{ provider: 'anthropic' | 'openai' }` when a key is saved, else `null`; the counts keep the rating's meaning.
630
- - **`isOwnKeyRefused`** on `OwApiError`: 422 `own_key_refused`, the own key's provider refused it (revoked, wrong, or out of credit). Nothing was counted and the call was not retried on the platform's key: send the person to the account page, and do not retry. A 429 `quota_exceeded` now means the rating is spent and no own key is saved.
631
-
632
- ## [4.130.0] - 2026-11-02
633
-
634
- ### Changed
635
-
636
- - **A signed-out visitor sees the tools that need an account, greyed.** `openersFor` with `audience: 'public'` now gives a row for a tool whose `open.needs` is `account`, with the new state word `"Needs an account"` (whatever `signedIn` says: the public is the signed-out reader); before, the public saw no such row. A tool that needs the device is still not there for the public. The word comes after "You are here" and "Needs a connection", and before "Opens the world". `OwOpenerState` gains the word, so a `switch` over it that lists every case meets a new one. `mountWaymark` draws the row like one that needs a connection: grey, a link with no href, `aria-disabled="true"`, in the arrow and Tab order, doing nothing when pressed. The shared openers fixture has 92 cases (from 90): seven public cases gain their account rows, and two are new (a visitor outside a browser, a visitor on a tool that needs an account).
637
-
638
- ## [4.129.2] - 2026-11-02
639
-
640
- ### Fixed
641
-
642
- - **`mapSpace` reads a number past 2^53 as not whole** (schema-dist `profiles/map-space.md`, rule 1: a reader holds whole numbers within the range it holds exactly). JavaScript parses a larger integer to the nearest double, so `-9223372036854775809` arrived as -2^63 and was kept as a point at that rounded value. A coordinate, `order`, `width`, `height` or `depth` past 2^53 - 1 now reads as not a whole number, as one with a fraction does: an `x` or `y` leaves the row out (`no_point`), a `z` reads as 0. The profile's fixture keeps every number inside that range, so no fixture answer moves.
643
-
644
- ## [4.129.1] - 2026-11-02
645
-
646
- ### Fixed
647
-
648
- - **`mapSpace` reads two rows of one type with one id as the last one served** (schema-dist `profiles/map-space.md`, serial 106, reading I: an element is its id). Markers and Pins now follow the rule Maps and Zones already did: until now each row of a repeated Marker or Pin id was read as a Marker or Pin of its own. The last row wins even when it reads less (no point, or a map that is not served), and is then counted under its reason.
649
-
650
- ### Changed
651
-
652
- - **Schema repinned to schema-dist `v0.34.0-dist.106`** (from dist.105); the standard stays canonical **00.34.00** and no schema field moved. Serial 106 words the map-space page's readings I to K and four sharpened rules (a measured frame from placed points only, a `z` that is not a whole number read as 0, `no_markers` for a Zone with no placed Marker, a Pin with neither a served map nor a point counted once as `no_map`), and its fixture holds 34 cases. Apart from reading I, `mapSpace` already answered every new case as ruled.
653
-
654
- ## [4.129.0] - 2026-11-02
655
-
656
- ### Added
657
-
658
- - **`mapSpace(elements)`: the map-space profile** (schema-dist `profiles/map-space.md`, serial 105). It reads a world's map space from its elements as the reader is served them: `{ maps, zones, pins, out }`, all four keys always present. Each Map reports its frame (`width`, `height`, `depth`, `z_max`, `scale`, `location`, `parent_map`; `z_max` is the largest z among its own points, and a missing `width` or `height` is measured from the Map's own points and marked `extent: "points"`). Each Zone has one outline per Map it has Markers on, its Markers in `order` ascending (a Marker without a whole `order` after the ordered ones, then by id in code point order). Each Pin reports its point and its element (both keys or both null). What the reading left aside is counted by reason in `out` (`no_point`, `no_markers`, `dangling`, `no_map`). A number is whole by value (`2.0` is 2), and a row without a non-empty string `id` is no element. The fixture `profiles/map-space.fixture.json` (24 cases) is the test. Types: `OwMapSpace`, `OwMapSpaceMap`, `OwMapSpaceOutline`, `OwMapSpacePin`, `OwMapSpacePoint`, `OwMapSpaceOutReason`.
659
-
660
- ### Changed
661
-
662
- - **Schema repinned to schema-dist `v0.34.0-dist.105`** (from dist.104); the standard stays canonical **00.34.00** and no schema field moved. Serial 105 adds the map-space profile page and its fixture.
663
-
664
- ## [4.128.1] - 2026-11-02
665
-
666
- ### Fixed
667
-
668
- - **`draught(id)` without `compact` answers the fixed `OwDraught` shape again.** The server now serves the compact packet by default; the client sends `compact=false` when the caller leaves it out, so every section its type promises is there (since 4.47.0 a left-out `compact` was sent as nothing). `compact: true` still asks for the short form, typed `OwDraughtCompact`.
669
-
670
- ## [4.128.0] - 2026-11-02
671
-
672
- ### Added
673
-
674
- - **`worldGraph(world, reader, { readingAids: true })` brings the reading aids back into an unlensed graph** (schema-dist `profiles/lens.md`, serial 104). The reading aids are a world's charter, its guide and its domains, each recognised by its pair and the mark `x_charter_of` (the pair without the mark is lore and a node like any other). They are out of an unlensed graph by default and counted `reading_aid`; with the flag they are nodes, each marked `reading_aid` (last in `marks`), and a link to one is an edge. The flag is a boolean, false by default (null reads as false). Any signed reader may ask it; a public reader asking it is refused by name before anything is read, and of several refused flags `readingAids` is named last, after `hidden`, `proposed`, `rooms` and `mapSpace`. A flag never widens a lens: with a lens the flag adds nothing, and a lens that names the pair (`law/charter`) keeps the charter whatever the flag says. `OwGraphMark` gains `'reading_aid'`.
675
-
676
- ### Changed
677
-
678
- - **A reading aid a lens keeps carries the mark `reading_aid`**, as a room element a lens keeps carries `room`; before, it carried no mark.
679
- - **Schema repinned to schema-dist `v0.34.0-dist.104`** (from dist.103); the standard stays canonical **00.34.00** and no schema field moved. Serial 104 adds the graph's `reading_aids` flag to the lens profile, with five graph cases in the lens fixture (201 cases in all), and two readings to the landing profile, with three cases (33 in all): `orphan` and `near_orphan` are degree 0 and degree 1 by the landing page's own degree, so a node whose only link is to itself is a near-orphan and a list naming one target twice counts degree 2; and a `heads` too large for the reader reads as the largest it holds, so `heads: 1e30` is taken, never refused, and every linked Character is a head. `landingOf` answers all three as it stood.
680
-
681
- ## [4.127.0] - 2026-11-02
682
-
683
- ### Added
684
-
685
- - **`landingOf(graph, options)`: the landing profile** (schema-dist `profiles/landing.md`, serial 103). A world's first picture at a door, read from the graph `worldGraph` answers for the visitor, so it holds only what that reader may see. The answer is `{ mode, size, clusters, unclustered, gaps }`, all five keys always present: `mode` is `gaps` when the graph has fewer nodes than `size` (default 24), else `graph`; `clusters` are the most-linked Characters (at most `heads`, default 5), each with the nodes within `hops` steps of it (default 1), a node joining its nearest head; `unclustered` is every node no head reaches; `gaps` are the orphans, the near-orphans, a Character whose `location` names no Location in the graph (`no_place`) and a Location no Character's `location` names (`nobody_here`). Counted in integers, ids in code point order, so the same graph and options give the same bytes. An unknown option, or one that is not a whole number of 1 or more, is refused by name. The fixture `profiles/landing.fixture.json` (30 cases) is the test. Types: `OwLanding`, `OwLandingCluster`, `OwLandingMember`, `OwLandingGap`, `OwLandingGapKind`, `OwLandingMode`, `OwLandingOptions`.
686
- - **`LocationV2.vehicle`**, from canonical 00.34.00: a single link to the Object a place is the inside of (a ship, a train, a walking city), for a place that moves with the thing; a place inside another place still uses `parent_location`. It is in `FIELD_SCHEMA`, `SINGLE_LINK_FIELDS`, the Location's Setting section (after `parent_location`) and `SCHEMA.md`.
687
-
688
- ### Changed
689
-
690
- - **Schema repinned to schema-dist `v0.34.0-dist.103`** (from dist.102), canonical **00.34.00**, so `ONLYWORLDS_VERSION` is now `'00.34.00'`. Serial 103 adds `vehicle` to Location, the landing profile, the lens profile graph section's degree and out-degree with the lens fixture's `degree` cases (196 cases in all; `degreeOf` and `outDegreeOf` answer the five new ones as they stood), and seven readings in the record profile (its fixture from 32 cases to 40).
691
- - **`recordsOf` reads a run or a story as proposed exactly when `proposedMarkOf` does.** An empty mark (`{}`, `""` or `[]`) is settled, so a run or story carrying one is now read; before, any value but null left it out.
692
- - **`recordsOf` never reads an Event or a Narrative whose `id` is not a string**: no entry without an id, and no error. Before, a run with a numeric id was returned with that number as its id. The record profile's other new cases (`played_at`'s upper-case `T` and `Z` and ASCII digits, a name that is not a string, ties by code point) pass as the reader stood.
693
- - **`TOOLS` repinned at the tools list's `f10afdc`**: Tactical Tangle is a listed row (`tangle`). It has no open door, so `openersFor` offers it nowhere, and a record whose `tool` is `tangle` finds its row by id.
694
- - **A reading builds its reverse link index only when a lens hop asks for it.** `worldGraph` without a lens, and every other reading that walks no hop, no longer pays for it; no answer changes. On a 20,573-element world, `worldGraph` for the owner went from a median 434 ms to 324 ms, both builds interleaved in one process.
695
-
696
- ## [4.126.0] - 2026-11-01
697
-
698
- ### Added
699
-
700
- - **`openersFor(tools, kind, address, surface, { element })`: no row for a tool that cannot open the element.** The tools list's `open` now says what a tool's element door does not open, `element_skips`, in the graph reader's words: `room`, `hidden`, `proposed`, `map_space` (a Map, Pin or Marker; a Zone is content) and `reading_aid`. Given the element as served, `openersFor` leaves out a row whose door skips a mark the element carries, each mark read by the readers `worldGraph` uses (`isRoomElement`, the hidden mark, `handsOnAsProposed`, the type, the reading aids), so the menu and the graph give one answer; a test holds the two equal on every element of the lens fixture's worlds. Without `element` every row is there, as before. A world's menu takes no element. Refused by name: an unknown options key, an element for a world's menu or naming another id than the address, a skip word outside the list's, and skips on a row with no element door. The shared openers fixture gains 25 cases (90 in all), the same bytes as the Python package's.
701
- - **`mountWaymark` passes its `element` through**, so the Waymark's Open with rows follow the same rule with no change on the page.
702
- - **`OwOpenersOptions`, `OwElementMark`**, and `OwToolOpen.element_skips`.
703
-
704
- ### Changed
705
-
706
- - **`TOOLS` repinned** at the tools list's 6da4d55: each row carries `element_skips` (Little Lens skips room, hidden, proposed and map space; Atlas and Lantern none until their doors are walked).
707
-
708
- ## [4.125.0] - 2026-11-01
709
-
710
- ### Added
711
-
712
- - **`createWorldLink(worldId, { landsOn })` and `lands_on` on every link**: where a person link opens. `site` (the default) is onlyworlds.com's /join page; `dashboard` is the Dashboard's /join page, for someone invited to work in the Dashboard. Either page takes the same code; `dashboard` on an agent or player link is refused by the server. `OwWorldLink.lands_on` (and so `OwCreatedWorldLink`'s) is always present, `site` on an agent or player link.
713
- - **`OwWorldLens.skipped`**: with `resolve` and `readAs`, a lens the reader read as cannot see (deleted or hidden from it) is not read: `skipped` is `not_found` and `reading` is null, so `reading` is now typed `OwWorldLensReading | null`.
714
-
715
- ## [4.124.2] - 2026-11-01
716
-
717
- ### Fixed
718
-
719
- - **`recordsOf` orders tied records and stories by code point**, as the record profile says. It compared ids with JavaScript's `<`, which is UTF-16 order and differs from code point order for characters above U+FFFF (an id holding U+1F600 sorted before one holding U+FF00). Ids in ASCII, which every fixture case uses, order as before.
720
-
721
- ## [4.124.1] - 2026-11-01
722
-
723
- ### Changed
724
-
725
- - **`recordsOf` reads a world in one pass.** It finds the Character's stories by each Event they name once, then looks each record's stories up, where it scanned every element again for each record: on a world of 5,000 elements with 200 records, about 7 ms a call down to about 1 ms. Every answer is the same as 4.124.0's (compared on 20,000 random worlds, including Narratives that name an Event twice and `events` lists holding other values), and each record's stories are its own objects.
726
-
727
- ## [4.124.0] - 2026-11-01
728
-
729
- ### Added
730
-
731
- - **`recordsOf(characterId, elements)`, `recordMarkOf(element)` and `RECORD_FIELD`: the record profile** (schema-dist `profiles/record.md`, serial 102). A game's run is kept in the world its Characters live in, as an Event carrying the mark `x_record`: `tool` and `played_at` required; `version`, `seed`, `outcome` and `seats` optional. `recordsOf` reads one Character's records from the elements as the reader is served them, newest first, each with that Character's stories of the run (a Narrative naming them as `protagonist`, `narrator` or both). Who took part is the Event's `characters` alone. A run or a story still proposed is not read. `played_at` must be a real calendar instant (checked part by part, never by `Date.parse`, which rolls February 30 into March), and records order by the instant to the millisecond. The fixture `profiles/record.fixture.json` (32 cases) is the test; the Python package's `records_of` gives the same answers.
732
-
733
- ### Changed
734
-
735
- - **Schema repinned to schema-dist `v0.33.0-dist.102`** (from dist.101; the same canonical 00.33.00, so `ONLYWORLDS_VERSION` stays `'00.33.00'`). Serial 102 adds the record profile; no schema file moved.
736
-
737
- ## [4.123.5] - 2026-11-01
738
-
739
- ### Changed
740
-
741
- - **`TOOLS` carries Little Lens.** The tools list is repinned at `a1573ac` (from `86208c8`): Little Lens, a world as a web of its elements and links, read only, is a listed row, so `openersFor` offers it in the Open with menu for a world and for an element. No type or function changed.
742
-
743
- ## [4.123.4] - 2026-10-31
744
-
745
- ### Fixed
746
-
747
- - **The Waymark closes when focus comes back to the page somewhere else.** Tab off the last row with nothing after it on the page takes focus into the browser's toolbar; the fan-out stays open then, as it does for any window that loses focus. When focus came back into the page on another control (measured in Chromium and WebKit), the fan-out stayed open over the page with focus outside it. Focus arriving anywhere outside the fan-out now closes it, and stays where it arrived; coming back to a row keeps it open.
748
- - **Every row of the Waymark is reached by Tab in Safari.** WebKit leaves a link out of the Tab order unless it carries a `tabindex`, so Tab skipped the rows that are links (a tool that opens on the web) between the rows that are buttons. Link rows now carry `tabindex="0"`; Chromium and Firefox are unchanged. Read in Chromium, Firefox and WebKit at 390 and 1440 px wide.
749
-
750
- ## [4.123.3] - 2026-10-31
751
-
752
- ### Fixed
753
-
754
- - **A value in a request path that is not one segment is refused before anything is sent.** 4.122.1 escaped every path value, but the API decodes an escaped slash before it routes: `get('character', '<id>%2Fhidden')` and an id carrying `/` still reached another route (`/character/<id>/hidden`). Every type, id, link field and account-route id (world, seat, link, member, key, token, file, proposal) is now checked first: a type is a lower-case word, every other value letters, digits, `_`, `.` and `-`, not starting with a dot or a dash. Anything else is a `TypeError` naming the value, and nothing is sent; what passes is still escaped. Every id the API serves passes. A test reads every request path in the client's source and fails on one that does not go through the check.
755
-
756
- ## [4.123.2] - 2026-10-31
757
-
758
- ### Fixed
759
-
760
- - **`openersFor` refuses a link that opens no app.** A tools list row whose `open` template uses `javascript:`, `data:`, `vbscript:`, `file:`, `blob:` or `about:` (in any case), or does not begin with its scheme, is a `TypeError` naming the row, where it was drawn as a row a press would run. The tools list's own check refuses these too; this holds the reader to it for a list that skipped the check (the `tools` option, a served copy). Four cases in the shared openers fixture (61).
761
-
762
- ## [4.123.1] - 2026-10-31
763
-
764
- ### Changed
765
-
766
- - **The Waymark's mark is one step smaller**: its globe is `max(16px, 0.6em)`, where 4.123.0 drew `0.7em`, a little under the name's capitals. The 16 px floor reaches a name under 26.7 px. Its 44 px tap area is unchanged; a test now holds both sizes in the shipped stylesheet.
767
-
768
- ## [4.123.0] - 2026-10-31
769
-
770
- ### Added
771
-
772
- - **`tool: 'onlyworlds'` is the site itself.** `openersFor` (and `mountWaymark`'s `tool`) gives the site's own row "You are here" when the surface names itself `onlyworlds`, before "Needs a connection", as a tool's own row already did; the site passes it, so on onlyworlds.com its own row reads as where the reader is, not as a door. Three cases in the shared openers fixture (57 now).
773
-
774
- ### Changed
775
-
776
- - **The Waymark's web rows open in a new tab** (`target="_blank"`, `rel="noopener"`), so the page the reader came from stays open. A row whose link opens an app on the device (any scheme but `http:` and `https:`) is handed to the device as before.
777
- - **The surface's own row is drawn grey**, as where the reader already is (its own class, `ow-waymark__row--here`, beside the cannot-open grey). It stays a button that closes the fan-out.
778
- - **The mark is smaller**: its globe is `0.7em`, as tall as the name's capitals and centred on them, never under 16 px (`max(16px, 0.7em)`), where it was `1em`. Its 44 px tap area is unchanged.
779
- - **The schema pin moves to schema-dist serial 101** (`v0.33.0-dist.101`): a sentence on the strand page (a surface that only reads may draw the filled slots alone); no schema file moved, the strand fixture unchanged.
780
-
781
- ## [4.122.1] - 2026-10-31
782
-
783
- ### Fixed
784
-
785
- - **Every value a call puts in a request path is sent as one escaped segment.** `get`, `upsert`, `patch`, `delete`, `editLinks`, `adopt`, `hide`, `unhide`, `draught`, `list` and `create` put the element type, id, link field or Character id into the path as given, so an id carrying a `/`, `?` or `#` asked a different route (`get('character', 'abc/hidden')` asked `GET /character/abc/hidden/`). They now escape each one, as the file, review and proposal routes already escaped the id; the type is escaped on those routes too. On a real id (the API declares every element id a UUID) nothing on the wire changes. A test reads every request path in the client's source and fails on one that interpolates a value unescaped.
786
-
787
- ## [4.122.0] - 2026-10-31
788
-
789
- ### Added
790
-
791
- - **`listAccountTokens()` and `revokeAccountToken(tokenId, opts?)`**: the account's tokens (`GET /account/tokens`, each by name and last four, with `last_used_at`, `revoked`, `expires_at`, `keys_minted` and `audience`; typed `OwAccountToken`), and a revoke by id (`DELETE /account/tokens/{tokenId}`), this client's own token included, so a tool done with a sign-in can end it by the `tokenId` the sign-in answered. An id that is not one of the account's tokens is a 404 (`err.isNotFound`); `signal` and `keepalive` go to fetch as `revokeWorldKey`'s do. Account token only; a world key is a `TypeError` before sending.
792
-
793
- ## [4.121.1] - 2026-10-31
794
-
795
- ### Fixed
796
-
797
- - **`draught` reads with `world` set, as `list`, `get` and `getWorld` do.** The API serves the draught to an account token that names the world (the account's own access there) and to a keyless reader of a world open to the public; the client refused both before sending. It now sends `world=<id>` with the Bearer, or with no credential.
798
-
799
- ## [4.121.0] - 2026-10-31
800
-
801
- ### Added
802
-
803
- - **`reading_aids.strand` on `GET /me` and an agent's join answer** (schema-dist serial 100, the strand profile's rule 2). The server's front door serves the world's strand fourth in the reading aids, after the Charter, the Guide and the Domains: `strandOf`'s six slots as that key reads the world, every slot null when the world has no strand or its story is out of the key's view. `OwMe.reading_aids` and `OwAgentJoined.reading_aids` are typed `OwServedReadingAids`, `OwReadingAids` plus `strand: OwStrand`; `charterOf` still answers the three, and `strandOf(world, reader)` reads the same `strand` from a world snapshot.
804
-
805
- ### Changed
806
-
807
- - **The schema pin moves to schema-dist serial 100** (`v0.33.0-dist.100`). No schema file moved; the strand fixture is unchanged at 26 cases.
808
- - **`mintWorldKey`'s `name` follows the server's rule**: trimmed, at most 80 characters (a longer one is a 422), `null` or blank for a key with no name, the same rule as `createAccountWorld`'s `key_name`. The option now takes `null`; the doc said 255, the limit the server held before.
809
-
810
- ## [4.120.0] - 2026-10-31
811
-
812
- ### Changed
813
-
814
- - **A strand begins when its story is chosen** (schema-dist serial 99). `x_strand` needs only `narrative`: `event` and `location` may each be `null` or absent, read as unset, so a world whose story is chosen before its happening and place reads as a strand begun, the story, hero and now filled and the other slots empty. A place named with no event is an empty place. The value is still no mark when `narrative` is missing or not a non-empty string, or when `event` or `location` is anything but a non-empty string, `null` or absent (an empty string, a number, an object), so a broken writer never shows as a half-built strand. `strandOf` reads every case of the serial's 26-case fixture.
815
- - **The schema pin moves to schema-dist serial 99** (`v0.33.0-dist.99`). No schema file moved.
816
-
817
- ### Fixed
818
-
819
- - **No `X-OW-Agent` on hide, unhide, `setMyCharacter` and `mintWorldKey`.** An `agent: true` client sent the header on these four, where the API document declares none and the server reads none. It now sends it only where the document lists it: the element writes, the links route, `bulk`, `adopt` and the file ticket.
820
-
821
- ## [4.119.0] - 2026-10-30
822
-
823
- ### Added
824
-
825
- - **`strandOf(world, reader)`, the strand profile's reader** (schema-dist serial 98, `profiles/strand.md`). A strand is a world in miniature: one story, one of its happenings, where that happened, the story's hero and where the hero is now, read off four links the standard already has. The World's mark `x_strand` (exported as `STRAND_FIELD`) is an object of three ids, `{narrative, event, location}`; any other shape is no mark. `strandOf` answers `{narrative, event, location, character, now, same}` as `reader` sees the world, each filled slot `{id, name}` and an empty one `null`: `narrative` is the marked Narrative; `event` the marked Event when it is in the story's `events`; `location` the marked Location when it is in that event's `locations`; `character` the story's `protagonist`; `now` the hero's `location`; `same` true when `now` and `location` are one place (compared by id), false when two, else `null`. A slot holds only its own kind, no list order is read, and no slot is filled from another. The world and the reader are `resolveLens`'s (`OwStrandWorld`, `OwLensReader`), so every slot is read through that reader's view and the bodies it is served: a story the reader knows only as a stub fills its own slot and nothing past it, and a public reader reads nothing of a world that is not open. Types `OwStrand`, `OwStrandSlot`. All 17 of the profile's fixture cases are the test.
826
-
827
- ### Changed
828
-
829
- - **The schema pin moves to schema-dist serial 98** (`v0.33.0-dist.98`), which adds the strand profile and its fixture. No schema file moved: the generated output changes only its provenance lines.
830
-
831
- ## [4.118.0] - 2026-10-30
832
-
833
- ### Added
834
-
835
- - **`openersFor` takes `online` on the surface.** With `online: false`, every row whose href is a web address (its scheme `http:` or `https:`, read from the row's own filled href) says "Needs a connection", the site's own `onlyworlds` row included, which had no word before. The words in order, the first that holds: "You are here" for the surface's own tool, then "Needs a connection", then the account, browser and phone words, then "Opens the world"; so offline, a row that needed a sign-in says the connection. A row whose link opens an app on the device (any other scheme) keeps its own word, and every row keeps its href. `online` defaults to true, which gives the rows as before; anything but `true` or `false` is a TypeError naming it. `OwOpenerState` gains the word. The shared fixture gains nine cases (54; the same bytes as the Python package's).
836
- - **The Waymark draws the offline row.** At each open it reads `navigator.onLine` (no navigator reads as online) and draws the rows `openersFor` gives for it. Offline, a row that says "Needs a connection" is grey (`ow-waymark__row--cannot`, no new style) and is a link with no href and `aria-disabled="true"`, still a menu item in the arrow and Tab order; a click, Enter or Space on it does nothing, and the fan-out stays open. While the fan-out is open it listens to the window's `online` and `offline` events and redraws the rows in place: focus on a row stays on that row, and nothing else moves. The listeners go when it closes. Copy link is unchanged. No new option.
837
-
838
- ### Fixed
839
-
840
- - **The Waymark's sheet and its toast clear a phone's safe area.** On a page with `viewport-fit=cover`, an iPhone's home indicator covered the lower part of the sheet's last row (Copy link) and the "Link copied" toast. The sheet's bottom padding is now `calc(18px + env(safe-area-inset-bottom, 0px))` and the toast sits at `calc(24px + env(safe-area-inset-bottom, 0px))`, at 480 px and below and in the sheet a narrow box takes alike. On a page without `viewport-fit=cover` the inset is 0 and nothing moves.
841
-
842
- ## [4.117.2] - 2026-10-30
843
-
844
- ### Fixed
845
-
846
- - **The Waymark opens as the sheet when neither side fits.** Wider than 480 px, a fan-out whose start would cross the right edge of the box that clips it and whose end would cross its left edge (a mark in the middle of a narrow window, or of a narrow pane) took the side that crossed less and still left the box: in a 481 px window, a mark at 196 to 235 px opened its 300 px pop from 187 to 487, 6 px past the edge. The same holds for below and above. Now the open puts `ow-waymark--sheet` on the wrapper and the stylesheet draws the sheet from the bottom, the layout it already draws at 480 px and below (rows at least 44 px, the page behind dimmed), and no side class is set. A side is taken only where it fits, so `--end` and `--up` mean the pop lies inside the box. "Link copied", shown after the fan-out closes, keeps the sheet's place until it hides. The sheet's rules under the class are the media block's, declaration for declaration, held by a test. No node carries a style attribute.
847
-
848
- ## [4.117.1] - 2026-10-30
849
-
850
- ### Changed
851
-
852
- - **The schema pin moves to schema-dist serial 97** (`v0.33.0-dist.97`), which states in the address profile that a writer's empty string is never absent and cases it four times, once per part. `elementAddress` already answered that way, so no code changes: given `""` for the world, the type, the id or the lens it throws `OwAddressError` with that part's own reason (`bad_world`, `unknown_type`, `bad_id`, `bad_lens`); only `null` or a missing key counts as absent.
853
-
854
- ## [4.117.0] - 2026-10-30
855
-
856
- ### Changed
857
-
858
- - **The Waymark's Escape stops at the component.** Every Escape the fan-out handles (in the fan-out, on the mark or the "Open with" entry while it is open, and in the toast after a failed copy) calls `stopPropagation()` beside `preventDefault()`, so a page's own Escape listener (one on `document` that leaves the page, say) does not also run. An Escape while the fan-out is closed reaches the page as before.
859
- - **The fan-out picks its side when it opens.** At each open it measures itself against the box that clips it on each axis: the nearest ancestor of the name whose `overflow` on that axis is not `visible`, within the window, else the window. When the fan-out at the name's start would cross that box's right edge it opens at the name's end (`ow-waymark__pop--end`), and when below the name would cross its bottom it opens above (`ow-waymark__pop--up`), each only where that side crosses less. The toast shares the anchor and takes the same side (`ow-waymark__toast--end`, `ow-waymark__toast--up`); after a failed copy with the fan-out above the name, the toast sits below it. The fan-out's classes are removed when it closes, and the toast's when it hides. Where the fan-out is a sheet from the bottom, no side is chosen. The rules are in `waymark.css`, with no new colour, and still no node carries a style attribute.
860
- - **The sheet from the bottom starts at 480 px wide and below** (it was 390 px), so every portrait phone, which runs to about 430 px, gets the sheet with its rows at least 44 px tall rather than the popover; a desktop window 480 px wide or narrower gets it too, by width alone.
861
- - **The page's own row leads nowhere.** The row `openersFor` gives "You are here" (the `tool` option's own) is a button with `aria-current="page"` and no link, in its place in the arrow order and with the same look; pressing it closes the fan-out and returns focus to what opened it. Every other row is a link as before.
862
-
863
- ## [4.116.1] - 2026-10-30
864
-
865
- ### Changed
866
-
867
- - **The schema pin moves to schema-dist serial 96** (`v0.33.0-dist.96`), which states how the lens profile's graph reads a guest's own hidden creation and cases it twice. Both cases pass as `worldGraph` already reads them, so no reader changes: with no lens the creation is not in the guest's graph, with no count and no counted cut edge, as for any hidden element below the owner; in a lens's graph that keeps it, it is a node marked `hidden`.
868
-
869
- ## [4.116.0] - 2026-10-30
870
-
871
- ### Changed
872
-
873
- - **`members()` works on an account token.** On a client holding an `ow_a_` token with `world` set (`new OwV2Client({ apiKey, world })`), it sends `GET /members/?world=<id>` with the Bearer token and no PIN, and reads the roster with the token's grant in that world: the owner's token as the owner, a member's as its own member key. A world where the token has no grant answers 404. Without `world` set it is a TypeError before anything is sent, naming the option; before, every account token was refused. A world key's call is unchanged (no `world`, no PIN on the read), and a client with no key is still refused before sending: the roster has no public reader.
874
-
875
- ## [4.115.0] - 2026-10-30
876
-
877
- ### Added
878
-
879
- - **`dragField(a, b)`**: the link a drag between an element of type `a` and one of type `b` writes, in either order, from the schema distribution's new `pairs.json` (a drawing aid like the colour groups, not part of the standard), baked into the package at this release. It answers `{ on, field, holds }`: `on` the type that carries the link field `field`, and `holds` true when the drop puts one element inside the other (one square under the other), false when it only draws a line. Null for the same type twice, a pair the table leaves out (113 of the 231 pairs of two types are in it), or anything that is not one of the 22 type slugs as written; a pair answered null offers no drag. Its answer's type is `OwDragField`.
880
- - **`readMarkOf(relations, characterId, members, { keyless })`**: an optional fourth argument, `keyless` as `isToEveryone` takes it. With `{ keyless: true }`, for bodies read with no credential, no read mark counts and the answer is null: the public reader is served no `created_by` or `changed_by`, so a mark there carries no hand to check. Without it nothing changes, and an unknown key in it is a TypeError, as in every options object.
881
-
882
- ### Changed
883
-
884
- - **The lens, message and charter readers follow schema-dist serial 95** (`v0.33.0-dist.95`), whose five new cases replay in `resolveLens`, `readMarkOf` and `charterOf`. One reading changes for callers of `resolveLens` and `worldGraph`:
885
- - **Hidden beats `known_of`.** A hidden element a guest's `known_of` names, and that the guest did not make, is not in the guest's view at all: no stub, no row and no count, and a link to it in a body the guest is served is left out as for any element the guest cannot see. A guest's own hidden creations, listed in `whole`, stay in its view whole. Before, such an element was served to the guest as a known-of stub.
886
- - The serial's other cases replay unchanged: a guest's `full` row of its own hidden creation carries `hidden: true`, an owner's `full` row of a hidden Message carries it as any hidden element's row does, and a skill id a Charter names only in uppercase hex is kept, in lowercase.
887
-
888
- ## [4.114.0] - 2026-10-30
889
-
890
- ### Changed
891
-
892
- - **The lens readers and `charterOf` follow schema-dist serial 94** (`v0.33.0-dist.94`), whose twenty-four new cases replay in `resolveLens`, `worldGraph`, `charterOf` and `typeWord`. Five readings change for callers:
893
- - **An anchor the reader cannot see is counted before anything else, for the owner alone.** An anchor that names no element, an element of another type than its field's, or a hidden umbrella a child lens points back at (the lens's own id included) is counted as `anchor_not_visible` the first time it is met, ahead of any reason a hop or a child lens gives the same element, and each element is counted once. Any other reader reads the lens as if that anchor were not there, so when a child lens cuts the same element by `types` or the window, the count is under the child's reason, and a graph edge to it is cut under that reason too (`outside_lens` for an element the reader sees that no path kept).
894
- - **An empty string, or one of White_Space alone, in an anchor list is no anchor.** A lens whose anchor lists hold only such strings is a filter over the reader's view by its `types`, and nothing is counted for the strings; before, a string of White_Space alone was taken for an id that named nothing, so the lens kept nothing (an empty string was already no anchor).
895
- - **A note goes with its anchor.** A note captions a kept row only for an anchor the reader can see. A note on one it cannot see captions no row, even when another path keeps the element, and is not counted in `notesIgnored`.
896
- - **A `full` row is the body as served.** A row of an element under the hidden mark (the world's `hidden` list or the element's own `hidden`) carries `hidden: true`, as the API serves it: the owner's row of a hidden element, and a guest's row of a hidden element of its own. No other body carries a `hidden` key, so a stored `hidden: false` is dropped. A `name` or `first_line` row never carries it.
897
- - **A proposed element is in no Domain's hubs.** `charterOf` and `domainsOf` leave an element written as proposed out of a Domain's `hubs` for every reader, the owner included, as they already left it out of the Charter, the Guide, the Domains and `skill`.
898
- - The serial's other cases (a public reader's served list, a single link stored empty, an id one lens anchors under two fields met once under the first, a member reading the Charter whole from a stub, a skill id compared whatever its case, and the names map's six) replay unchanged.
899
- - **The waymark entry no longer loads the world-file and zip code.** `@onlyworlds/sdk/waymark` reached `readWorldFile`'s module, and with it fflate, through the case fold the room and names readers use; the fold now lives in its own module, so the chunk the entry loads imports no fflate (101 KB to 58 KB). `simpleCaseFold` is exported from the package root as before.
900
-
901
- ## [4.113.0] - 2026-10-30
902
-
903
- ### Added
904
-
905
- - **`@onlyworlds/sdk/waymark`**, with its stylesheet at **`@onlyworlds/sdk/waymark.css`**: `mountWaymark(target, options)` draws the OnlyWorlds mark beside a world's or an element's name and the fan-out it opens (the name line, the Open with rows, the lenses the page offers, Copy link), and returns a disposer that removes every node, listener and timer it added. Every word is the core's: the rows and their state words from `openersFor`, a member's "Last shaped in Atlas, 12 Oct" from `doorsOf` and `doorWhere`, the copied link from `elementAddress` (the site's address with the world's id; the toast says "Link copied"). The public face shows the name and type, only the rows that need nothing first, and Copy link. Options are refused by name when unknown, before anything is drawn. No node carries a `style` attribute, so it runs under `style-src 'self'`; every colour in the stylesheet is one of the Waymark's two grounds (`dark`, `paper`). The mark is a button and the fan-out a menu, keyboard first (Tab through the rows, arrows, Home, End, Escape back to the mark; focus moving out of it closes it and stays where it went, the end of the page included, while the window losing focus leaves it open); at 390 px wide and below it is a sheet from the bottom with rows at least 44 px tall, and the mark's tap area is at least 44 by 44 px at every width. `entry: 'mark' | 'menu' | 'both'` picks the way in: the mark beside the name, an "Open with" entry (`dispose.entry`) for the page's own "..." menu, or both; Escape returns focus to the one that opened it, and `returnFocus` (refused with `'mark'`) names the page's own element, its "..." button, to take focus back from a fan-out the entry opened. A failed copy (no clipboard, or the write refused) keeps the fan-out open, and the toast says "Could not copy the link." over the address in a read-only line to copy by hand, until the fan-out closes. The core (`.`) does not import it and stays free of the DOM.
906
- - **`npm run demo`**: builds the package and serves `demo/waymark.html`, one public element on a plain page with the fan-out beside its name (`?entry=` with `mark`, `menu` or `both`, `?reader=`, `?ground=`, `?signedIn=`, `?current=` for the page's own tool). Not in the package.
907
-
908
- ### Changed
909
-
910
- - **`sideEffects`** in `package.json` is `["*.css"]` (was `false`), so a bundler keeps an import of `waymark.css`; every JavaScript file is still free of side effects.
911
- - **The package has two entries.** `dist/index.js` and `dist/waymark.js` share one chunk of code, and each entry's declarations stand alone in its own `.d.ts`: `dist/index.d.ts` still holds every declaration of the core.
912
-
913
- ## [4.112.0] - 2026-10-30
914
-
915
- ### Changed
916
-
917
- - **The lens readers follow schema-dist serial 93** (`v0.33.0-dist.93`), whose nineteen new cases replay in `resolveLens` and `worldGraph`. Four readings change for every caller:
918
- - **A link reaches only an element of its field's target type.** An id naming an element of another type reaches nothing (a data fault, not a hop), draws no edge and takes no place in a reader's served list. A generic pair (a Pin's `element_type` and `element_id`) links only when its type key is exactly one of the 22 slugs as stored, nothing trimmed or folded, and in a graph it is placement, never an edge; the Pin's `map` link is the edge.
919
- - **An anchor the reader cannot see is counted.** An anchor id that names no element of the world, or an element of another type than its field's, is now counted for the owner as `anchor_not_visible`; it was skipped silently.
920
- - **A known-of stub keeps its base and public fields, every other field unset.** The stub is made from the body the reader is served: `change_seq` joins the base fields, a public link (a Character's `species`) is kept as trimmed for that reader, every other field of the type it holds is at its unset value (`''`, `null` or `[]`; the who and door fields `null`), and no extension key is kept.
921
- - **A graph node's name is the stored value**, `null` when absent, and an edge's `index` is the target's place in the list as served, which keeps a proposed id the reader's rung sees.
922
- - **`draughtText` names the public reader.** The API reads a public draught (keyless, or an account with no access in the world) as the public reader in either mode, and a public `voice` packet's text said "Read as the Character's own view." It now says "Read as the public reader." whenever the packet's `reader` is `public`; every other packet, and one without `reader`, reads by `mode` as before. The shared draughtText fixture gains the case (`public-reader`, 51 cases).
923
-
924
- ### Added
925
-
926
- - **`reader` on `OwDraught`**: the rung the packet was built through (`public`, `owner`, `co_builder`, `contributor`, `guest`, `seat` or `player`), which the API serves beside `mode` on every draught since 2026-10-30, kept under `compact`; `mode` stays what was asked.
927
-
928
- ## [4.111.0] - 2026-10-30
929
-
930
- ### Changed
931
-
932
- - **A Zone is a node in `worldGraph` by default.** schema-dist serial 92 (`v0.33.0-dist.92`) rules that a Zone is not map space: by its definition it is content that gains spatial presence through linked Markers, so the Markers are the geometry and the Zone is an element like any other, in a public reader's graph too. Without a lens, Zones are now nodes and their links edges; Maps, Pins and Markers stay out until `mapSpace` brings them back, and the `map_space` count no longer includes Zones. The fixture's eight graph cases over the shared world replay as the serial gives them. A graph read through a lens is unchanged: its nodes are still exactly what the lens kept.
933
-
934
- ## [4.110.0] - 2026-10-30
935
-
936
- ### Changed
937
-
938
- - **`openersFor` checks the whole address.** It handed the address writer only the world, type and id, in the site's form, so a malformed `lens` or `moment`, or a `form` that is neither `human` nor `text`, still gave rows. The whole address now goes to `elementAddress` in its own form (the site's when it has none), so it refuses with the writer's own `OwAddressError`, `bad_lens`, `bad_moment` or `bad_form`, as it already did for the world, type and id; an address without a world still refuses `no_world`, the text form inside a world included. The rows are unchanged: a lens and a moment still ride on no href. The shared fixture gains four cases (45; the same bytes as the Python package's): the three refusals, and a well-formed lens and moment giving exactly the rows of the same address without them.
939
- ## [4.109.0] - 2026-10-29
940
-
941
- ### Changed
942
-
943
- - **`writeWorldFile` leaves the author fields out of every element file by default.** `created_by`, `changed_by`, `changed_via`, `changed_client` and `doors` record which member and which tool wrote each element (a membership id, a door), which means nothing to someone the world is handed to. Each element file under `elements/<type>/` or `spatial/<type>/` holding any of the five is now written without them, in the folder's own form (two-space indent, a trailing newline, its other keys in their order); `created_at` and `updated_at` stay. A file holding none of them, or that is not a JSON object, travels byte for byte, and `world.json`, records, media and every other file are untouched; an `x_*` value that holds one of the names is the extension's and stays. A caller archiving their own world passes `{ includeAuthorFields: true }`, which writes the same bytes as 4.108.1.
944
-
945
- ### Added
946
-
947
- - **`includeAuthorFields`** on `writeWorldFile`, refused by name when it is not a boolean, as `includePrivateRecords` and `includeHidden` are; and **`authorFieldsLeftOut`** in its `report`, the element files written without their author fields (0 when the option is true). A shared fixture of five cases (`test/fixtures/author-fields.fixture.json`, the same bytes as the Python package's) holds both writers to the same entries.
948
-
949
- ## [4.108.1] - 2026-10-29
950
-
951
- ### Fixed
952
-
953
- - **A list filter with an empty value is sent.** `list` and `listAll` sent nothing for a key under `filter` whose value was `''`, so `filter: { characters: '' }` or `filter: { supertype: '' }` listed every element. On keel an empty value is a filter of its own: `supertype=` and `subtype=` list the elements with none, `characters=` the Messages addressed to everyone, and a single link's (`parent_location=`, since every single link became a list filter) the elements linking to nothing. Each key under `filter` is now sent as given, `''` included; an empty `name__icontains` still matches every name, and outside `filter` an empty value is still not sent. The `filter` doc names the single-link filters.
954
-
955
- ## [4.108.0] - 2026-10-29
956
-
957
- ### Added
958
-
959
- - **`doorWhere(door)`**: where one entry of `doorsOf` wrote the element, as the where-phrase a "Last shaped" line shows, and nothing else (no reader, no time: "by you" and when stay the caller's). It pairs with `doorsOf`: pass it an entry as `doorsOf` gives it, its tools-list match already made. The words are the ones the Open with menu will use:
960
- - a listed tool (`tool` not null): `in <tool name>`, such as `in Atlas`; this rule comes first;
961
- - `site`: `on onlyworlds.com`;
962
- - an agent's door (`isAgentDoor`: `seat`, `mcp`, `declared`): `with an AI`;
963
- - `server`: `by OnlyWorlds`;
964
- - anything else (`sync`, `api`, `v1`, a client the tools list does not hold, such as `sdk`): `in code`.
965
-
966
- An entry that is not an object, a `door` that is not a string, or a `tool` that is neither null nor an object with a string `name` is a TypeError. The shared fixture gives every case a `where` list, one phrase per expected entry, and a case with every rule's word (27 cases); the Python package's `door_where` answers it phrase for phrase.
967
-
968
- ## [4.107.1] - 2026-10-29
969
-
970
- ### Changed
971
-
972
- - **Schema repinned to schema-dist `v0.33.0-dist.91`** (from dist.90; the same canonical 00.33.00, so `ONLYWORLDS_VERSION` stays `'00.33.00'`). Serial 91, no change to what the package reads or writes: the dist's folder checker (`walk/lint_folder.py`) moves to folder format 0.3.21 and takes `doors` on an element file as a server-managed key beside `changed_client`, and its README says 0.3.21. No code changed.
973
-
974
- ## [4.107.0] - 2026-10-29
975
-
976
- ### Added
977
-
978
- - **`doorsOf(element, tools = TOOLS)`**: which tools wrote an element, newest first, from its `doors` field, each `{ door, at, tool, last, from }` (`OwDoor`). `door` is the key (a tool's client name, or a door word such as `site` or `mcp`), `at` its time exactly as served, `tool` the first `tools` row whose `client` is the door as `{ id, name }` (else null), `from` is `doors`. Times are compared as instants at their full precision, offsets and microseconds included, never as strings and never through `Date`; entries at the same instant go by key, by Unicode code point. A key that is empty, or a value that is not a time (`YYYY-MM-DDTHH:MM:SS`, an optional fraction, then `Z` or `±HH:MM`, with a real date and clock), is skipped. `last` is true on the first entry at the same instant as `updated_at`; when none is, the last write came through no door the field records (a restore or a scrub). An element without `doors`, or with no valid entry in it, gives one entry from its last write (`changed_client`, else `changed_via`, at `updated_at`, `from: 'last-write'`), or `[]` when it has neither door or no valid `updated_at`, as a public reader's element does. Data only, no words: a "Last shaped in Atlas" line reads the first entry. An element that is not an object, or `tools` that is not an array, is a TypeError. The Python package's `doors_of` answers the same shared fixture (26 cases) entry for entry.
979
-
980
- ## [4.106.0] - 2026-10-29
981
-
982
- ### Added
983
-
984
- - **`doors` on every element type**: each tool that wrote the element, with the time of its last write there, as the API serves it to a world's members since 2026-10-29. One key per tool (the name it sent in `X-OW-Client`, else the door word of that write, such as `site` or `mcp`), each value the `updated_at` that write left; at most 16 keys. Optional: absent for the public reader, on a known-of stub, in the change feed's bodies, in a world file, and on an element not written since the field arrived. Read-only; the server ignores it in a write.
985
-
986
- ### Changed
987
-
988
- - The doc comments on `changed_by`, `changed_via` and `changed_client` say they are served to a world's members only and absent for the public reader, as the API serves them since the same day.
989
- - The draught text fixture gains a `doors` case: the door history never prints as a field line.
990
-
991
- ## [4.105.0] - 2026-10-29
992
-
993
- ### Added
994
-
995
- - **`openersFor(tools, kind, address, surface)`**: the rows of an "Open with" menu for a world (`kind: 'world'`) or one element of it (`'element'`), in order, each `{ id, name, href, state }`. First the site's own row (`onlyworlds.com`, the world's or the element's page on www.onlyworlds.com, never a word), then one row per tools list row, in the list's order, that is `listed` and has an `open` link for this kind; a tool with only a world link is offered on an element as opening its world. `surface` is what the page knows, and nothing is guessed: `audience` (`member` or `public`, required; the public sees only rows whose `open.needs` is null), `signedIn` (true, false, or null when the page cannot know), `browser` (default true), `phone` (default false) and `tool`, the page's own client name. `state` is the first that holds: "You are here" (the row whose `client` is `tool`), "Opens in your browser" (needs an account, outside a browser), "Sign in to <name> first" (needs an account, signed out), "Needs the world on this phone" (needs the world on the device, on a phone), "Opens the world" (a world link on an element), else null. An address with no world, or with a slug for it, refuses with `OwAddressError` (`no_world`, `bad_world`); an element's rows from an address with no element, an unknown `kind`, a missing `audience`, and an unknown key on the address or the surface are a TypeError. Pure and DOM-free; the Python package's `openers_for` answers the same shared fixture (41 cases) row for row.
996
- - **`TOOLS`**: the OnlyWorlds tools list (github.com/OnlyWorlds/tools, `tools.json` format 1) as of this release, the keys `openersFor` reads (`id`, `name`, `url`, `client`, `rung`, `open`), with the types `OwTool`, `OwToolOpen`, `OwToolNeeds`, `OwOpener`, `OwOpenerKind`, `OwOpenerState` and `OwOpenerSurface`. It is generated by `codegen/tools.mjs` from a copy vendored byte for byte at a pinned commit (`codegen/tools-pin.json`, the commit and the file's sha256), and `npm run codegen:check` now also fails when the copy or the generated module is stale. A tool listed after a release is offered from the next one.
997
-
998
- ## [4.104.3] - 2026-10-29
999
-
1000
- ### Changed
1001
-
1002
- - **Schema repinned to schema-dist `v0.33.0-dist.90`** (from dist.89; the same canonical 00.33.00, so `ONLYWORLDS_VERSION` stays `'00.33.00'`). The calendar fixture gains one refusal, an intercalary day on a calendar with no months at `after_month: 0` (76 cases, 42 refusals), which `readCalendarConfig` has refused since 4.103.0; with that check removed, the case fails. The doc comments on `supertype` and `subtype` now say a classification within the element's type. No code changed.
1003
-
1004
- ## [4.104.2] - 2026-10-29
1005
-
1006
- ### Fixed
1007
-
1008
- - **`completeSignIn` keeps a sign-in pending when a return carries another state.** It cleared the pending record before checking the state, so any page that opened a native app's custom scheme (`obsidian://onlyworlds-signin?state=anything`) could cancel a sign-in in flight. Now a return whose state does not match leaves the record as it is, and the real return still signs in. Once a return carries the sign-in's own state, the record is cleared whatever follows: a refusal, a cancel on the consent page, or the wrong page.
1009
- - **`OwMember.sponsor_member`**, which the server now serves on agent rows of `members()`: who answers for the AI, as a membership id on the same list, `owner`, or null (a seat made from a link for an email address, or a sponsor with no live membership). Person and player rows carry none.
1010
-
1011
- ## [4.104.1] - 2026-10-29
1012
-
1013
- ### Fixed
1014
-
1015
- - `worldProposals` and `listProposals` say who reads the review queue as the server now answers: the owner's read key (or the write key, with no PIN) reads every row, and a demo key and the keyless public view get a 404. No code changed.
1016
-
1017
- ## [4.104.0] - 2026-10-29
1018
-
1019
- ### Added
1020
-
1021
- - **Sign in with OnlyWorlds from a native app** (RFC 8252). A tool registered as native lists a loopback return without a port (`http://` and the IPv4 or the IPv6 loopback address with a path) or its own scheme (`obsidian://onlyworlds-signin`). `signInWithOnlyWorlds` now takes, beside an `https:` address, the loopback address with the port the app bound for this sign-in (the same form with a port from 1 to 65535, written plainly), which the server matches against the portless one, and a custom scheme exactly as registered (lower-case, not `http`, `https`, `javascript`, `data` or `file`, something after `://`). It refuses `localhost`, a loopback without its port, and on any form userinfo, a query, a fragment and a dot segment (and an `https:` address with an upper-case host or no path, which it took before), with a TypeError naming the forms it takes: the server would refuse each of them at authorize. Whether the tool is registered as native is the server's to check; the SDK does not read the client list. A native app opens the authorize URL in the system browser through `navigate`, and passes the address its loopback server or scheme handler received to `completeSignIn` as `url`.
1022
-
1023
- ### Fixed
1024
-
1025
- - **`completeSignIn` compares the return address exactly**: its scheme, its host with the port, and its path. It compared `URL.origin`, which is `"null"` for every scheme that is not special, so a page at `evil://onlyworlds-signin` would have matched a sign-in begun for `obsidian://onlyworlds-signin` (out of reach until now, since only `https:` could be begun). A trailing slash is still tolerated on `https:`, where a static host may answer `/callback` by sending the browser to `/callback/`; on a loopback or custom-scheme address it is not.
1026
- - **`completeSignIn` cleans the address bar only when it read the address from `location`.** With `url` given, the window's history is left alone: a native app's callback address is not the address of the window it runs in.
1027
-
1028
- ## [4.103.0] - 2026-10-29
1029
-
1030
- ### Changed
1031
-
1032
- - **Schema repinned to schema-dist `v0.33.0-dist.89`** (from dist.88; the same canonical 00.33.00, so `ONLYWORLDS_VERSION` stays `'00.33.00'` and no type, field or constant moved). The calendar profile's fixture goes from 64 cases to 75 (209 steps, 41 values a reader must refuse) and the message profile's from 85 to 87; every case passes.
1033
- - **`readCalendarConfig` refuses more values, as the calendar profile now rules**, and `readCalendarValue`, `worldCalendars` and `resolveCalendar` follow (no calendar: show bare numbers):
1034
- - a `grain` with nothing left after trimming the room profile's White_Space set (`" "`, a tab, U+3000), where 4.102.5 refused only `""`. The trim is White_Space exactly, never JavaScript's `trim()`: a grain of U+FEFF still reads. `validateCalendarConfig` says `Grain is required` for the same values;
1035
- - an intercalary day whose `after_month` names a month the year does not have: past the last month (`after_month: 5` on a year of two months, which 4.102.5 read as a day in no month), below the first (`-1`), or any intercalary day on a calendar with no months. `after_month: 0` (before the first month) still reads, but only on a calendar that has months. With no months, `after_month` names a place between months and there are none, so even `after_month: 0` refuses there: this follows a ruling of the calendar profile that its fixture will carry later (serial 89's fixture refuses only an `after_month` above 0 there). An empty `intercalary` list still reads, and `validateCalendarConfig` already said `Festival days need months` for these values.
1036
- - **A `format` or `format_intercalary` that is not a string is ignored** and the default template prints. In 4.102.5 `formatStep` threw a TypeError on one.
1037
- - The calendar fixture's new case with a world's elements replays through `worldCalendars`: a calendar is found by the `atlas_calendar` field being present, never by type or supertype. The message fixture's two new cases, a `created_at` with an offset and a fraction of a second read as an instant, and the newest Note listed last, already passed: the message readers read times that way and keep the newest Note wherever it stands.
1038
-
1039
- ## [4.102.5] - 2026-10-29
1040
-
1041
- ### Fixed
1042
-
1043
- - **`readCalendarConfig` refuses an empty `grain`.** The calendar profile requires `grain` as a string, and an empty string is no grain: the reader took `grain: ""` as a calendar while `validateCalendarConfig` already refused it. It now returns `null` (show bare numbers), as the Python package's reader does from 0.84.3; schema-dist's next serial carries the case.
1044
-
1045
- ## [4.102.4] - 2026-10-29
1046
-
1047
- ### Fixed
1048
-
1049
- - **`OwMember.name` and `OwMe.name` say what the server serves.** A roster row's `name` is an agent's or a player's name to every reader, and on a person's row the person's username, served only to the owner's keys and co-builders and absent for everyone else. On `me()` it is your agent's or player's name. Only the doc comments changed; the types are the same.
1050
-
1051
- ## [4.102.3] - 2026-10-29
1052
-
1053
- ### Changed
1054
-
1055
- - **Schema repinned to schema-dist `v0.33.0-dist.88`** (from dist.87; the same canonical 00.33.00, so `ONLYWORLDS_VERSION` stays `'00.33.00'` and no type, field or constant moved). The lens fixture gains a graph case where one element's `x_review_proposed` is `{}` and another's is `""`: both are kept as settled nodes with no mark. `worldGraph` already read an empty mark this way since 4.102.1, so no code changed. The fixture's 153 cases (39 of them graph cases) pass.
1056
-
1057
- ## [4.102.2] - 2026-10-29
1058
-
1059
- ### Changed
1060
-
1061
- - **`draughtText` reads the review mark as the review profile's rule 1 does**, through `handsOnAsProposed`, as the Python package's `draught_text` does. In 4.102.1 it still read the key itself and wrote ` · proposed` on any value but absent or null. Now an empty mark (`""`, white space only, `[]`, `{}`) is settled and its line carries no ` · proposed`, while a doubt (`false`, `0`, a non-empty string) still does. The draught fixture's `proposed-odd-values` case now marks only its `false`.
1062
-
1063
- ## [4.102.1] - 2026-10-29
1064
-
1065
- ### Changed
1066
-
1067
- - **Schema repinned to schema-dist `v0.33.0-dist.87`** (from dist.86; the same canonical 00.33.00, so `ONLYWORLDS_VERSION` stays `'00.33.00'` and no type, field or constant moved).
1068
- - **An empty review mark is settled.** The review profile now reads `x_review_proposed` by presence, as the room and charter profiles read their marks, and for this mark an empty list or an empty object is absent too. Before, `handsOnAsProposed` answered true for any value but absent or null, so `""`, a string of white space only and `[]` were doubts and `{}` was a mark (`proposedMarkOf` gave `{ by: null, expires: null, lapsed: false }`). Now each of the four reads exactly as absent: `proposedMarkOf` gives `null`, `handsOnAsProposed` gives `false`, and every reader built on them follows, so `buildAskContext` sends such an element without the label and no longer lists it in `leftOut.proposed`, a lens or `worldGraph` keeps it as settled, and a Note carrying one counts as the newest own note. White space is the room profile's White_Space set, so a string of U+3000 is empty and a string of U+001C is a doubt. Any other present value is unchanged: `false`, `0`, a non-empty string and `[[]]` are doubts, and an object with any key, `{ x: null }` included, is a mark.
1069
- - The lens fixture's new graph case, one element of each of the 22 types with map space on, replays: each node carries its colour word and icon as `presentation.json` gives them. The fixture's 152 cases (38 of them graph cases) pass.
1070
-
1071
- ## [4.102.0] - 2026-10-29
1072
-
1073
- ### Added
1074
-
1075
- - **The colour groups, generated from the distribution's presentation defaults**: `ElementGroup` (`'blue' | 'green' | 'pink' | 'amber'`), `ELEMENT_GROUPS` (each type's group, from `types.<type>.family`), `GROUP_ORDER` (from `colors.order`) and `GROUP_COLORS` (each group's `{ light, dark }` hex, from `colors.families`), with `groupOf(type)`. The groups are named by their colour: a visual aid for drawing the 22 types, not part of the standard. The hexes are no longer hand-typed in the package; the generator refuses a group outside the order, a pair missing for a group, or a hex that is not `#rrggbb`, and the hexes themselves are unchanged. `elementColor(type, mode)` reads the groups, and an unknown type still throws its TypeError.
1076
-
1077
- ### Changed
1078
-
1079
- - **Schema repinned to schema-dist `v0.33.0-dist.86`** (from dist.85; the same canonical 00.33.00, so `ONLYWORLDS_VERSION` stays `'00.33.00'` and no field moved). `presentation.json` names each type's group by its colour word, and two types move: **Construct is pink** (it was with the green types) and **Ability is amber** (it was with the pink types), so `elementColor('construct')` and `elementColor('ability')` now answer those colours.
1080
- - **A `worldGraph` node's `family` is the type's colour word**, typed `ElementGroup`, as the lens profile now says: a key for drawing, never a classification or a label. The lens fixture asserts the words, and its 151 cases (37 of them graph cases) pass.
1081
- - `SCHEMA.md` heads each type with `colour:` and the colour word.
1082
-
1083
- ### Deprecated
1084
-
1085
- - **`ElementFamily`, `ELEMENT_FAMILIES`, `FAMILY_COLORS`, `FAMILY_ORDER` and `familyOf`** — use `ElementGroup`, `ELEMENT_GROUPS`, `GROUP_COLORS`, `GROUP_ORDER` and `groupOf`; the old names leave in 5.0. Until then they keep their old words (`agents`, `world`, `abstract`, `temporal`), so code that types one still builds, and they are derived from the colour groups (blue as `agents`, green as `world`, pink as `abstract`, amber as `temporal`): their membership follows the schema, so `familyOf('construct')` is now `'abstract'` and `familyOf('ability')` `'temporal'`.
1086
-
1087
- ## [4.101.1] - 2026-10-28
1088
-
1089
- ### Changed
1090
-
1091
- - **Schema repinned to schema-dist `v0.33.0-dist.85`** (from dist.84; the same canonical 00.33.00, so `ONLYWORLDS_VERSION` stays `'00.33.00'` and no type, field or constant moved). Four generated descriptions follow the standard: `supertype` and `subtype` are a classification within the element's category, and a Title's `grant_date` and `revoke_date` are the title's own dates, a holder's span being a Relation; `OwFrontDoor.charter` now says the Charter is the newest marked Charter Law, as the charter and checks profiles select it; no fixture case changed, and every reader replays as before.
1092
-
1093
- ## [4.101.0] - 2026-10-28
1094
-
1095
- ### Added
1096
-
1097
- - **`readingAidOf(element)`**: which reading aid an element's type and supertype make it, `'charter'`, `'guide'` or `'domain'` (from `CHARTER_PAIRS`, trimmed and compared without regard to case, the whole word), else `null`. It ignores the mark `x_charter_of`, so a tool can offer the right mark on an unmarked element; whether an element is a reading aid, the pair and the mark, is still `isCharterElement`.
1098
-
1099
- ### Changed
1100
-
1101
- - **Schema repinned to schema-dist `v0.33.0-dist.84`** (from dist.83; the same canonical 00.33.00, so `ONLYWORLDS_VERSION` stays `'00.33.00'` and no element field moved). Serial 84 adds four lens cases, a lens reached by a hop served whole to the owner and, to a contributor, a guest and the public reader, without its hidden anchor and every note keyed to a hidden element, and one charter case, a Guide held only as a stub; the lens fixture holds 151 cases (75 resolve, 37 graph), the charter fixture 23, every one passing with no reader change. The room, checks and spine profiles gain the folder format's re-key line, and the spine profile the line that an editor shows `x_spine_` keys read-only; nothing in this package restates either.
1102
-
1103
- ## [4.100.2] - 2026-10-28
1104
-
1105
- ### Fixed
1106
-
1107
- - **`x_served_trimmed` on `OwMe` and `OwMember`**: `me()` and `members()` answers carry it, typed `'lenses'[]`, when `lenses` was served with a lens left out that this key cannot see. The API has served it since the lens routes; the types now say so, as `OwMemberLenses` already did. Saving the list you were shown keeps that lens where it stands.
1108
-
1109
- ## [4.100.1] - 2026-10-28
1110
-
1111
- ### Fixed
1112
-
1113
- - **Lens rows below the public rung keep what the API serves about proposed elements, and no row carries a stored `x_served_trimmed`.** A proposed element (`x_review_proposed` anything but absent or null) is still in no reader's view: never a row, never hopped through. But a contributor or a guest that sees one is served the links to it, and a lens's note keyed to it, where `resolveLens` left both out; the public rung still drops them, unmarked. A body that carries `x_served_trimmed` as stored is served without it: the owner's rows lose it, and below the owner the mark names only what this reading trimmed (it was merged into the stored one). Eight more row cases hold these readings, 28 in all, the same file as the Python package's.
1114
-
1115
- ## [4.100.0] - 2026-10-28
1116
-
1117
- ### Added
1118
-
1119
- - **`charterOf(world, reader)` and `domainsOf(world, reader)`**: the charter profile's reader, the world's three reading aids as one reader sees a world snapshot. `charter` is the newest marked Charter it sees (newest by `created_at` compared as instants, none sorting as the oldest, then the higher id), `{ id, name, rules, skill }`: `rules` its `x_checks_rules` as stored (`[]` when none), `skill` the ids its description names by any `ow://` form, a site address or a bare id, lowercase, each once in the order written, kept when it is an Object of this world the reader sees, the first five of those kept, as the API's `charter_skill` reads it. `guide` is the newest marked Guide the same way, `{ id, name }`. `domains` is every marked Domain the reader sees, `{ id, name, sentence, hubs }`, in the order the Guide's `constructs` names them as the reader is served the Guide (a Domain named twice keeps its first place), then by folded name and id; `sentence` is the description to its first line break, trimmed, and `hubs` the distinct `{ type, id }` its links name that the reader sees, in the schema's field order, never itself, none for a Domain known only as a stub. The reader is a lens reader; the public rung reads nothing of a world that is not `open`. `domainsOf` is the list alone. Every case of the distribution's charter fixture is its own test.
1120
- - **`CHARTER_PAIRS`**: the charter profile's three pairs (`law`/`Charter`, `construct`/`Guide`, `construct`/`Domain`), each with the aid it is, generated from the distribution's charter fixture and checked against the page and the lens fixture's pairs.
1121
- - **`reading_aids` on `OwMe` and `OwAgentJoined`**: `me()` and an agent's `redeemJoin` answer carry the same packet beside `front_door`, typed `OwReadingAids` (`OwReadingAidsCharter`, `OwReadingAidsGuide`, `OwReadingAidsDomain`, `OwReadingAidsHub`), as the API's document declares it.
1122
-
1123
- ### Changed
1124
-
1125
- - **Schema repinned to schema-dist `v0.33.0-dist.83`** (from dist.82; the same canonical 00.33.00, so `ONLYWORLDS_VERSION` stays `'00.33.00'` and no element field moved). Serial 83 adds the charter profile and its fixture, writes the room and charter marks' presence rule, and adds fifteen lens cases; the lens fixture holds 147 cases (71 resolve, 37 graph), the room fixture 32, every one passing.
1126
- - **`isRoomElement` reads `x_room_of` by presence**: the mark counts when its value is not null, a string only when something is left after trimming White_Space, any other value (a number, an object) as present, the value never compared. It wanted a non-empty string before, so a number read as no mark and a mark of spaces as one. `isCharterElement` already read `x_charter_of` this way.
1127
- - **An umbrella lens's children keep their own budget and mail rule**: each child keeps at most its own `budget` (the rest counted `budget`) and its own mail rule takes its messages out of the child's kept, uncounted, before the union takes the umbrella's mail rule and budget. The rows are made once, from the union, by the umbrella's `show`, and a child's `notes` never reach a row. The union drops an element only by its id, so a sibling lens a child's hop reaches stays kept.
1128
- - **Nothing is hopped out of what the fence cuts**: an element cut `after_as_of`, an anchor among them, is not hopped out of. The second hop's `hops_not_served` count leaves out a room element or a reading aid whose pair `types` does not name.
1129
- - **Lens rows below the owner come as the API serves that reader.** A link to an element the reader cannot see was already left out, and still is. What moves: `x_served_trimmed` is no longer added for the public rung, and on a Message it names only `characters` and `narrator`, in that order (it named every trimmed link field, `objects` and `locations` too). A Pin aimed at an element the reader cannot see reads `element_type` `""` and `element_id` null (the type stayed), and so does one whose `element_type` is not its element's type; an id of the wrong type for its link field is left out. The owner reading as itself gets the bodies as stored (an id outside its view, a missing or proposed one, was dropped). A public reader has no mail part.
1130
-
1131
- ### Fixed
1132
-
1133
- - **4.99.0 handed hidden elements' ids to a reader below the owner, in the notes of a lens kept as an element.** A lens a hop reaches is kept as an element, and its row carried `x_lens.notes` whole, so a contributor, a guest or the public got the ids of hidden anchors and other hidden elements as note keys, with the notes' text. The row now keeps a note only for an anchor it is served or another element the reader sees; for a contributor or a guest, `notes` joins `x_served_trimmed` when an anchor's note went with its anchor. Twenty row cases hold these readings, the same file as the Python package's.
1134
-
1135
- ## [4.99.0] - 2026-10-28
1136
-
1137
- ### Added
1138
-
1139
- - **`setMyLenses(null)` and `setMemberLenses(world, member, null)`**: `null` puts a lens list back to never chosen, so a seat playing a Character starts with its Titles' lenses again, while an empty list now keeps a seat off them (the API tells the two apart from this release on). On the owner's own list `null` is the empty list. Anything that is neither a list nor `null` is still refused before sending.
1140
-
1141
- ## [4.98.0] - 2026-10-28
1142
-
1143
- ### Added
1144
-
1145
- - **The reading aids, and `isCharterElement(element)`**: the world's charter (a Law with supertype `Charter`), its guide and its domains (Constructs with supertype `Guide` or `Domain`) are reading aids when they carry the mark `x_charter_of`, the id of the world they belong to. `isCharterElement` is true for one: the pair, trimmed and case-blind as `isRoomElement` matches, AND the mark, present and not null (a string mark only when something is left after trimming). The pair without the mark is lore, a Law like any other. `resolveLens` keeps an aid a hop reaches only when `types` names its pair (`law/charter`, `construct/guide`, `construct/domain`, now legal `types` words), cuts it as `types` otherwise, and never hops on out of it unless the pair is named; an aid the lens names as an anchor is kept and followed like any anchor. `worldGraph` leaves the aids out of an unlensed graph, counted `reading_aid` in `cut` and `edgesCut`, and no flag brings them back; a lens that names the pair keeps them as nodes. The pairs and the mark are generated from the distribution's lens fixture.
1146
- - **`changed_within` in `x_lens`**: a whole number of days from 1. What a hop or a filter reaches is cut, counted `stale`, when it last changed longer ago than that many days before the reading's `clock` (an element exactly that many days old is kept); an anchor never is. The time read is `updated_at`, or for a folder element the newer of `server_updated_at` and `local_updated_at`; an element with no readable time is never cut. The cut's order is the fence, `types`, the window, `changed_within`, the budget. Any other value makes the lens unreadable as `changed_within`. `OwLensCutReason` gains `stale`, `OwGraphCutReason` gains `reading_aid`.
1147
- - **`OwWorldLens.x_lens.changed_within`**: the lens route's recipe echo carries the key, a whole number of days or null, as the API serves it since serial 82.
1148
- - **`outDegreeOf(graph)`**: each node's out-degree in a graph the graph reader answered, the number of distinct nodes its own link fields point at: a target listed in two fields counts once, a link to itself does not count, and a mirror pair counts 1 at each end. Every node is a key, 0 included. Beside `degreeOf` ("most linked"), it says how far an element reaches out. Held to a direct count on every graph case of the lens profile's fixture.
1149
-
1150
- ### Changed
1151
-
1152
- - **Schema repinned to schema-dist `v0.33.0-dist.82`** (from dist.81; the same canonical 00.33.00, so `ONLYWORLDS_VERSION` stays `'00.33.00'` and no element field moved). Serial 82 writes the reading aids and `changed_within` into the lens profile, and a lens served only as a stub resolves as not there, which the reader already did; the lens fixture holds 132 cases (58 resolve, 37 graph), every one passing.
1153
- - **The `client` option's words name every call that sends `X-OW-Client`**: the element writes, and also `propose`, `acceptProposal` and `reviewNewElement`, which sent it already; the words had named only the element writes.
1154
- - **`worldLenses`' words follow the server's wider `read_as`**: a co-builder, or the account that sponsors an agent's or a player's seat, may read a lens as that seat by its membership id, judged by the narrower view. The call already sent `readAs` as given; only the words moved.
1155
-
1156
- ### Fixed
1157
-
1158
- - **A lens's `notesIgnored` counts only notes on ids the reader sees.** It counted a note on any id that was not an anchor, so a guest's reading counted notes keyed to elements it cannot see, and to ids in no element, which told it something hidden exists. It now counts a non-anchor id only when the reader's view holds it, as the profile says and the API counts. A shared edge case holds it, the same file as the Python package's.
1159
-
1160
- ## [4.97.1] - 2026-10-28
1161
-
1162
- ### Fixed
1163
-
1164
- - **A lens's mail part no longer reads a known-of stub as a message to everyone.** A stub never serves `characters`, and the message profile reads a Message with no recipients as addressed to everyone, so a guest's mail part listed every Message it knew only as a stub, unread. The mail part now reads whole bodies only, as the API reads it. A shared edge case holds it, the same file as the Python package's.
1165
-
1166
- ## [4.97.0] - 2026-10-28
1167
-
1168
- ### Added
1169
-
1170
- - **Where a seat's lenses come from.** `OwMe` and `OwMember` carry `lenses_from`, `chosen` (the row's own list, empty for none) or `titles` (it chose none, and its Character's Titles bring these); `lenses` is then the lenses whose `titles` name a Title the Character holds. The owner's list is always its own choice. On `OwWorldLens`, `from_titles` says the caller starts with the lens through a Title (never with `chosen`), and the owner gets `title_seats`: each seat that starts with it through a Title, with those Titles, in the members' order. The API serves all three since 2026-10-28.
1171
-
1172
- ### Fixed
1173
-
1174
- - **An umbrella lens no longer enters a child lens the reader knows only as a stub.** Since 4.93.1 a lens named by id reads as nothing when the reader is served it only as a known-of stub; as an umbrella's child, the same stub was still entered and its stored anchors read, so a guest's reading kept elements reached through links it was never served. Such a child is now an anchor the reader cannot see (`anchor_not_visible`, counted for the owner only), as the API reads it. Two shared edge cases hold it, the same file as the Python package's.
1175
-
1176
- ## [4.96.1] - 2026-10-27
1177
-
1178
- ### Fixed
1179
-
1180
- - **The lens and graph readers read the world's now as the API serves it.** `GET /world` names the world's now `time_range_current`, and a world folder names it `time_current`; `resolveLens` and `worldGraph` read only `now` and `time_current`, so a World body passed straight from the API had no now: no fence, and a `window` that cut nothing. They now read `now`, else `time_current`, else `time_range_current`, and `OwLensWorld` types the third. A shared edge case holds it in both packages; the Python package's `resolve_lens` and `world_graph` read the same.
1181
-
1182
- ## [4.96.0] - 2026-10-27
1183
-
1184
- ### Added
1185
-
1186
- - **`DATE_FIELD`** and **`dateOf(element)`**: the lens profile's date table, exported (each dated type's date field by the type's lowercase name; the ten timeless types are not in it; frozen), and an element's date as that table reads it: the whole number in its type's date field, or null for a timeless type, a body with no known `type`, and a date field that is absent, null or not a whole number. It reads the body it is given, so a known-of stub reads null. A timeline or a horizon can now date elements exactly as the lens reader's fence does, from one table. The Python package's `DATE_FIELD` and `date_of` give the same answers.
1187
-
1188
- ## [4.95.0] - 2026-10-27
1189
-
1190
- ### Added
1191
-
1192
- - **`degreeOf(graph)`**: each node's degree in a graph the graph reader answered: its edges in or out, each counted once for every node it touches (a mirror pair counts 2 at each end, a link to itself once), every node a key with 0 included, in the graph's node order. It is "most linked" as a finder ranks it: the reader's own count, since what the reading left out has no edge. The Python package's `degree_of` gives the same answer on the same graph.
1193
-
1194
- ## [4.94.0] - 2026-10-27
1195
-
1196
- ### Added
1197
-
1198
- - **`setMemberLenses(worldId, memberId, lensIds)`**: `PATCH /account/worlds/{world_id}/members/{member_id}` with `{ lenses }` alone, the lenses a row starts with, lens ids in order (0 to 20, `[]` for none, which is a choice). An account token's call: a member sets its own row, a seat it answers for, or any seat as a co-builder; the owner any seat, and its own list with `'owner'` as `memberId`. Nothing but `lenses` is sent. A world key, or anything but a list, is a TypeError before sending. keel's codes are the 404 for a lens you or the row's holder cannot see, 422 `not_a_lens`, `lens_is_template` and `lens_hidden`, and 422 when the owner names a person's row.
1199
- - **`OwMemberLenses`**: what `setMemberLenses` returns, keel's `MemberLenses`: `id` (null for the owner's own list), `role`, `kind` (`owner` for the owner's own list), `lenses` as you are shown it, and `x_served_trimmed`, the served-only mark naming `lenses` when the list keeps a lens you cannot see.
1200
-
1201
- - **`expiresAt` on `completeSignIn`'s result**: when the account token stops working, as the exchange now answers it (`expires_at`), for a tool whose registration asks for a lifetime; null for one with none, whose token works until it is revoked, and from a server that does not send it.
1202
-
1203
- ### Changed
1204
-
1205
- - **`worldLenses` and `setMyLenses` take an account token**, as keel's document now declares. `worldLenses(worldId)` sends the Bearer token alone, the world in its path, no `world=`. `setMyLenses(lensIds)` sets your own list in the world the client's `world` names, sending `world=<id>` and the Bearer token; an account client without `world` set is a TypeError before sending. On a world key both send as before (`setMyLenses` with its PIN, never a `world`). `request()` takes the rule as an option, `accountToken: 'path' | 'world'`, beside `worldRead`.
1206
-
1207
- ## [4.93.1] - 2026-10-27
1208
-
1209
- ### Fixed
1210
-
1211
- - **A lens a guest knows only as a stub reads as nothing when named by id**, in `resolveLens` and so in `worldGraph`, as keel's own reader answers it. The reader took the stored body whenever the id was in the guest's view, a known-of stub included, so it followed anchors and a recipe the guest is never served: the kept rows stayed inside the guest's view, but which of them the lens kept said something about anchors the guest cannot see. A lens body passed in is read as given, as before.
1212
-
1213
- ## [4.93.0] - 2026-10-27
1214
-
1215
- ### Added
1216
-
1217
- - **The lens profile's graph reader**: `worldGraph(world, reader, options?)` reads a world as nodes and edges for one reader, over the same snapshot and `OwLensReader` as `resolveLens`, and returns `OwWorldGraph` (`nodes`, `edges`, `edgesCut`, `cut`, `unreadable`). Pure: nothing is stored or written. A node is `{ id, type, name, family, icon }`, `family` and `icon` from the distribution's presentation defaults, with `marks` (`room`, `hidden`, `proposed`) on an element of a kind that is out by default and `known_of: true` on a guest's known-of stub, which has no edges out. An edge is `{ source, field, target, index }`, one per element, field and target, from the schema's link fields and the two declared extension links (`x_lens_leads_to`, `x_spine_depends_on`); `index` is the target's place in the list as the reader is served it, and a link held on both ends is two edges. Nodes sort by id and edges by source, field and index, each by code point. Without `options.lens` the nodes are every element the reader sees less room elements, hidden ones, proposed ones and map space (Map, Pin, Marker, Zone); the flags `rooms`, `hidden`, `proposed` and `mapSpace` bring each back, and `cut` counts what stayed out. With a lens (a body, or the id of one the reader sees) the nodes are exactly the lens's `kept`, its `cut` rides along unchanged, and no flag adds to it; a lens whose `x_lens` cannot be read gives an empty graph with `unreadable` naming the key, as `resolveLens` does (`unreadable` is null otherwise, and always without a lens). A link to an element that is not a node is counted in `edgesCut` by that element's reason (a lens's cut reason; with a lens, `outside_lens` for one the reader sees that the lens neither kept nor cut, counted for every rung and never part of a lens's cut; without one, one of the four kinds; or `not_visible`), in one order: the lens's reasons, then `room`, `hidden`, `proposed`, `map_space`, `outside_lens`; a reader that is not the owner never carries `anchor_not_visible`, `not_visible`, `hidden` or `proposed`, and a link to an id not in the world is not counted. `hidden` or `proposed` from a reader that is not the owner, or any flag from a public reader, throws a TypeError naming the flag before anything is read; an owner reading `as` another reader is judged by that reader's rung. Unknown option keys are refused, as everywhere in the package. It answers the 34 cases of the fixture's `graph` kind (schema-dist serial 81), which the Python package runs too; the answers equal the Python package's `world_graph` on thousands of random worlds.
1218
-
1219
- ### Changed
1220
-
1221
- - **Schema repinned to schema-dist `v0.33.0-dist.81`** (from dist.80; the same canonical 00.33.00, so `ONLYWORLDS_VERSION` stays `'00.33.00'` and no element field moved). Serial 81 writes the lens profile's graph section, which `worldGraph` reads, and the fixture's `graph` kind: the graph tests now run its 34 cases and read its graph tables (the flags, the owner-only flags, the edge reasons, the declared links) back through the reader, in place of the package's own copy of the same cases.
1222
- - **`OwWorldMeta.counts` and `getWorld` say what keel serves since its PR 289**: every world read carries `counts`, as that reader may read, not only the keyless read. Only the doc comments moved.
1223
-
1224
- ## [4.92.0] - 2026-10-27
1225
-
1226
- ### Added
1227
-
1228
- - **The server's lens routes.** `setMyLenses(lensIds)`: `PUT /me/lenses`, the lenses this key's holder starts with, lens ids in order (0 to 20, `[]` for none, which is a choice); a world write key with its PIN, answering GET /me's body as `OwMe`; no `X-OW-Agent` or `X-OW-Client`, which the document declares on neither; an account token, or anything but a list, is a TypeError before sending. `worldLenses(worldId, { resolve, readAs })`: `GET /world/{world_id}/lenses`, every lens in the world this key can see with its recipe echoed, the Titles it is the default for and whether this key chose it, `resolve: true` adding each lens's `reading` and `readAs` (the owner only, with `resolve`) reading as another reader; a read, so no PIN; typed `OwWorldLenses` (`data: OwWorldLens[]`, `read_as`), its reading `OwWorldLensReading`; `resolve` is true or false, and an account token is a TypeError before sending.
1229
- - **`draught(characterId, { lens })`**: a lens id, sent as `lens`; the packet then carries `lenses`, one frame with that lens read for this key's own view (`OwDraught.lenses`, each frame an `OwDraughtLens`: `id`, and `name`, `rows`, `cut`, `mail`, `skipped`, `unreadable` as served). With no lens and no policy key but `compact`, keel adds one frame per lens chosen with `setMyLenses`. `lens` beside `depth` or `budget` is keel's 422. The options fence takes the new key; a misspelt one is still a TypeError before sending.
1230
- - **`OwMe.lenses` and `OwMember.lenses`**: the lenses the key's holder, or a roster row, chose, lens ids in order; GET /me and GET /members now serve them.
1231
-
1232
- ### Changed
1233
-
1234
- - **Schema repinned to schema-dist `v0.33.0-dist.80`** (from dist.79; the same canonical 00.33.00, so `ONLYWORLDS_VERSION` stays `'00.33.00'` and no element field moved). Serial 80 writes the lens profile's five rulings into its page and adds four lens cases, 88 in all; `resolveLens` already answers every one, so the reader is unchanged.
1235
- - **`x_served_trimmed` is documented on a lens too**: keel now serves the mark on a Construct, a lens whose hidden anchors were left out, naming those anchor fields, then `notes`. `OwElementBase`'s comment said it rode only a message; the generator's text now names both, and nothing else changes (the SDK never strips the mark from a body it is given, on any type).
1236
-
1237
- ### Not added
1238
-
1239
- - `PATCH /account/worlds/{world_id}/members/{member_id}` now takes `lenses`, but the SDK has no method for that route, so none is added here.
1240
-
1241
- ## [4.91.1] - 2026-10-27
1242
-
1243
- ### Changed
1244
-
1245
- - **Schema repinned to schema-dist `v0.33.0-dist.79`** (from dist.78; the same canonical 00.33.00, so `ONLYWORLDS_VERSION` stays `'00.33.00'` and no element field moved). Serial 79 adds four message cases (a Note is addressed to nobody, never in a thread and never unread, even when its `characters` names the reader; any hidden mark leaves a Note out, not only `true`), 85 in all; the message readers already answer every one, so no reader changed. `noteOf` still returns the Note's id.
1246
-
1247
- ## [4.91.0] - 2026-10-27
1248
-
1249
- ### Added
1250
-
1251
- - **The lens profile's reader** (schema-dist `v0.33.0-dist.78`, `profiles/lens.md`; the schema repinned from dist.77, the same canonical 00.33.00, so `ONLYWORLDS_VERSION` stays `'00.33.00'` and no element field moved). `isLens(element)`: a Construct with supertype `Lens`, trimmed and case-blind, the whole word; `isTemplate(element)`: a lens with subtype `Template`, which resolves to nothing. `resolveLens(world, lens, reader, { asOf, clock })` returns `OwLensReading` (`kept`, `rows`, `cut` by reason, `notesIgnored`, `mail`, `unreadable`): the profile's ten rules over a world snapshot, pure, for one reader (`OwLensReader`: `owner`, `contributor`, `guest` with `whole` and `known_of`, `public`, and an owner reading `as` another). A value outside the profile's table names the key in `unreadable` and resolves nothing. The reading's date is `asOf`, else the world's now (`now`, else the World's `time_current`); with none, nothing is fenced and the window cuts nothing. `clock` is the wall-clock time the mail rule's thirty days count back from, never the world's now. The mail rule reads the message profile's own readers. `lensesInUse(membership, xLensUses)`: the membership's choice wins over a Character's `x_lens_uses`. The fixture's 84 cases run as `test/lens.test.mjs`, beside ten shared edge cases (`test/fixtures/lens-edges.json`, the same bytes as the Python package's) for what the fixture leaves open; the answers equal the Python package's `resolve_lens` on thousands of random worlds.
1252
-
1253
- ## [4.90.0] - 2026-10-27
1254
-
1255
- ### Added
1256
-
1257
- - **`mintWorldKey(worldId, { keep: true })`**: the key is kept across a sign-out, so revoking the token that minted it leaves it live, for a tool that stores a key on the device; disconnecting the tool, or a replayed sign-in code, still takes it. `OwMintedKey.kept` says so. Without `keep`, revoking a token revokes the keys it minted unless the tool holds another live token of its own; a token made by hand, or one of a tool that asks each time, never takes its keys. A `keep` that is not true or false is a TypeError before anything is sent.
1258
-
1259
- ## [4.89.1] - 2026-10-27
1260
-
1261
- ### Fixed
1262
-
1263
- - **`noteOf` reads the hidden mark as every other reader in the package does**: any `hidden` but absent, null or false leaves a Note out (it read only `true`), so a doubt is left out, as the review and world-file readers already decide. The shared edge case moves with it, in both packages.
1264
-
1265
- ## [4.89.0] - 2026-10-27
1266
-
1267
- ### Added
1268
-
1269
- - **The address profile's readers (schema-dist serial 77): `elementAddress(address)` and `parseOwLink(text)`.** One element is named by one triple, the world's id, the type word and the element's id, in three spellings: the site's address `https://www.onlyworlds.com/w/<world id>/<type>/<id>`, `ow://<type>/<id>` inside a world and `ow://<world id>/<type>/<id>` across worlds; a world by itself is `https://www.onlyworlds.com/w/<world id>` or `ow://<world id>`. Each may carry a moment (`@<step>`, one whole step of the world's time line) and a lens (`?lens=<id>`). `elementAddress` writes the canonical string: lowercase, the world's id and never its slug, the moment before the lens. `parseOwLink` reads one string whole and gives `{ form, world, worldIsSlug, type, id, lens, moment, canonical }`; a slug in the site's address reads, and is never canonical. Both refuse with `OwAddressError`, whose `reason` is one of the profile's ten names (`OwAddressReason`), the first check that fails in the profile's order. The parser is written out, not the runtime's URL class: nothing is percent-decoded and case is ASCII case. An address object with a key it does not know is a TypeError, so a misspelt `world` cannot write an address inside a world. The fixture's 133 cases are its test, and the Python package's `element_address` and `parse_ow_link` give the same answer on every input (compared on 54,018 random and adversarial inputs, no disagreement). Schema repinned to schema-dist v0.33.0-dist.77 (from dist.76; same canonical 00.33.00, no generated type changed).
1270
-
1271
- ## [4.88.0] - 2026-10-27
1272
-
1273
- ### Added
1274
-
1275
- - **`world`, a constructor option: an account token, or no key, reads the world it names.** `new OwV2Client({ apiKey: 'ow_a_…', world: id })` sends `world=<id>` (URL-encoded, beside the call's own parameters, on every page of `listAll`) with the Bearer and no API-Key or PIN on `list`, `listAll`, `get` and `getWorld`, and keel answers as that account's own access in the world: the owner reads it whole, a member as its member key reads, any other account as the public reads (a world not open to the public is a 404). `new OwV2Client({ world: id })` with no key sends the same four reads with no credential at all, keel's public reader. The account calls (`/account` routes, sign-in, key and link management) work on the same client as before. Refused with a TypeError before anything is sent: `world` with a world key (`ow_w_`, `ow_r_`, a seat's key, a 10-digit key) or as anything but a non-empty string, at construction; on an account client, every call that needs a world key (element writes, hide, the change feed, the draught, files, images, proposals, `me`, `members`), which keel answered with a 401 before; on an account client without `world`, the four reads; and `world` inside a `list` or `listAll` filter, on any client. The rule sits in `request()`, read from the request: an /account route, or a GET marked `worldRead` (a new `request()` option, which adds `world=` on such a client). A client with no key and no `world` refuses as before, and its message now names the four reads.
1276
- - **`signal` on `list`, `listAll`, `get` and `getWorld`**: an AbortSignal passed to fetch as given; on `listAll` it covers every page of the walk. A time limit per request goes in the client's own `fetch`.
1277
- - **`counts` on `OwWorldMeta`**: a keyless world read's count of what the public reads, in all and per type (`{ total, types }`), absent with a key; `patchWorld` never sends it back.
1278
- - **`isRateLimited` and `isServedTrimmedNotWritable` on `OwApiError`**: a 429 `too_many_requests` (over the request budget; wait `retryAfter`), told from a daily allowance's `quota_exceeded`; and a 422 `served_trimmed_not_writable`, a write that carried the mark keel serves on a body it trimmed for its reader. The mark is left in an echoed body on purpose: the refusal is the signal.
1279
- - **The message profile's serial 76 readers** (schema-dist `v0.33.0-dist.76`). `hasUnseenRecipients(message, { keyless })`: true when the server left out recipients this reader may not see (`x_served_trimmed` names `characters`), never who or how many; `null` for a body read with no credential. `editedBy(message, members)` returns `OwEdited`, `{ edited, by }`: whether a later hand changed the message (`changed_by` present and not equal to `created_by`) and that membership's Character, or `null` when it plays none, is off the members list or there is no list. `NOTE_PAIR` (generated from the dist), `isNote(element)` and `noteOf(narratives, characterId, members)`: a Note is a member's own line, never a message, and `noteOf` gives the id of the Character's newest own-hand Note that is neither hidden (`hidden: true`) nor proposed (`x_review_proposed` an object), or `null`; on the wire only a verified sender's Note counts, in a folder its `narrator` is trusted.
1280
- - **`OwElementBase` declares `x_served_trimmed?: string[]`**, the mark the server sets on a message whose `characters` or `narrator` lost ids this reader may not see. It is served only, and a write that carries it is refused by name (422 `served_trimmed_not_writable`); the client leaves it in a body you hand back, unlike `known_of`, so that refusal is the signal and nothing is dropped silently.
1281
-
1282
- ### Changed
1283
-
1284
- - **Schema repinned to schema-dist `v0.33.0-dist.76`** (from dist.75; the same canonical 00.33.00, so `ONLYWORLDS_VERSION` stays `'00.33.00'` and no element field moved). Serial 76 changes only the message profile and its fixture (55 to 81 cases, all run as the message test).
1285
- - **An empty `characters` with `x_served_trimmed` naming `characters` is to someone the reader cannot see**: `isAddressedTo` is false for every Character and `isToEveryone` false; a list that still holds ids beside the mark addresses those ids. `isToEveryone(message, { keyless: true })` is never true, since a body served with no credential carries no mark. Passing anything but an options object (a bare `true`) or a `keyless` that is not a boolean is a TypeError, never read as false.
1286
- - **`senderStatus` answers `unseen`** for a message with no `narrator` whose `x_served_trimmed` names `narrator` (show "from someone you cannot see", never "from nobody"); without the mark it stays `unverifiable`.
1287
- - **A Note is addressed to nobody on every reader**: `recipientsOf` gives `[]`, `isAddressedTo`, `isToEveryone` and `hasUnseenRecipients` (keyless or not) are false, `unreadFor` never lists one, and `threadOf` never includes one, though a chain of `parent_narrative` walks through it.
1288
- - **Order by `created_at` reads one rule, the same as the Python package's**: an RFC 3339 date-time (`YYYY-MM-DDTHH:MM:SS`, an optional fraction, `Z` or an offset; `T` and `Z` in either case), or one with no zone, read as UTC; the date must exist, hours 00 to 23, minutes and seconds 00 to 59. Anything else (absent, `null`, `09:61`, a word, a date alone, a space for the `T`, surrounding spaces) comes after every readable time, in the list's order. Before, `Date.parse` decided: a time with no zone was read in the machine's local time, and a date alone or other forms it accepts sorted as times. `readMarkOf` reads `updated_at` by the same rule.
1289
-
1290
- ### Fixed
1291
-
1292
- - **`listAll` stops when a page hands back the cursor it was sent** with more to come, with an `OwNetworkError`, instead of reading the same page forever. What it yielded before stays with the caller.
1293
-
1294
- ## [4.87.1] - 2026-10-26
1295
-
1296
- ### Fixed
1297
-
1298
- - **`OwDraught`'s `recent` rows declare `door`** (`string | null`, the door the change came through), which the API's document requires on every row and keel has always served. The type left it out, so `row.door` compiled only through the row's index signature as `unknown`; `draughtText` already read it. Found by checking every nested object in the document against the SDK's types.
1299
-
1300
- ## [4.87.0] - 2026-10-26
1301
-
1302
- ### Changed
1303
-
1304
- - **`draughtText` names what the date horizon left out.** When the draught has an `as_of`, a line above the budget line says it: `Not listed: rows dated after Year 300.` when the read was the Character's own view (`mode: voice`, where the horizon fences every section), and `Not listed: Events, Knowledge and proposed Events or Narratives dated after Year 300.` when it was the caller's (`act`, where it fences only those). The packet carries no count of those rows, so the line gives none. Before, only the title said the horizon, and `Left out by the budget: nothing.` could be read as a complete record when the text was pasted alone.
1305
- - **A removed member reads `removed from the world`**, not `removed`, so the word cannot be read as part of the member's name. The shared fixture's expected texts are regenerated (45 of 47 cases change; the two with no `as_of` are unchanged), and the Python package's `draught_text` gives the same bytes from 0.71.0.
1306
-
1307
- ## [4.86.0] - 2026-10-26
1308
-
1309
- ### Added
1310
-
1311
- - **`draughtText(draught)`**: one Character's context packet (`draught(characterId)`, full or compact) as plain text to paste into a prompt or a note, the same bytes from both packages and held to them by one shared fixture (47 cases, the full and compact reads of each giving the same text). It reads the packet alone and calls nothing: a link the packet does not hold is counted, never looked up. The title claims only the read's horizon ("dated up to Year 300"), never what the Character knew; every row says how it was reached (`via`), its grade or date, and the marks hidden and proposed; the sections that hold nothing are named on one line, and the last line says only what the budget left out (`cut`). Nothing is trimmed or cut by length, so the two languages cannot disagree on where a string ends. One known difference between a full and a compact read: the server's compact form drops an empty `x_review_proposed` (`""`, `[]`, `{}`), which the full read counts as proposed. The Python package's `draught_text` (0.70.0) gives the same text.
1312
-
1313
- ## [4.85.0] - 2026-10-26
1314
-
1315
- ### Changed
1316
-
1317
- - **`resolveCalendar` is the new name of `resolveCalendarLens`**, with the same arguments and answers. "Lens" now names a cut of a world (what a reader is shown of it), so the function that picks a world's chosen calendar no longer uses the word. `resolveCalendarLens` stays as a deprecated function with the same answers until 5.0; its refusal of an unknown option names the function called.
1318
-
1319
- ## [4.84.1] - 2026-10-26
1320
-
1321
- ### Fixed
1322
-
1323
- - **`validateCalendarConfig` answers with its errors, never a TypeError, when `bundles.months` has no names list** (`0`, `''`, `false`, `{}`, or `names` that is not an array). It read `months?.names.length` and threw where it should have listed the problem; such months now count as none for the festival-day check, and their own error is listed as before. `readCalendarValue` was not affected (it answered `null`). Found by comparing this reader with the Python package's port of it, input by input.
1324
-
1325
- ## [4.84.0] - 2026-10-26
1326
-
1327
- ### Changed
1328
-
1329
- - **Schema repinned to schema-dist `v0.33.0-dist.75`** (from dist.74; the same canonical 00.33.00, so `ONLYWORLDS_VERSION` stays `'00.33.00'`, no schema file moved and no generated type changed). Serial 75 changes the message profile's rules and its fixture (43 to 55 cases), and text in the calendar, import-map and names pages.
1330
- - **`unreadFor(messages, characterId, readMark, members?)` takes the world's members list, and on the wire a message is the reader's own only when its sender is verified.** Before, a message whose `narrator` was the reader was never unread for them, so anyone could write a message claiming to be from you and it would never show as unread to you. Now a message that carries `created_by` (wire data) is the reader's own only when `senderStatus(message, members)` is `verified` with the reader as sender: a mismatch, an author off the members list or playing no Character, and every wire message when no members list is given, stays unread for the Character it claims. A message with no `created_by` key (a world folder, an export) has no hand to check, so its `narrator` is trusted as before. The decision is made per message. **Callers who pass wire data and no `members` will now see their own sent messages in the unread list**: pass `await client.members()` as the fourth argument.
1331
- - **`readMarkOf` counts a mark only when every write on it was its reader's own hand.** Besides the author check (its `created_by` on the members list plays the reader's Character), a mark that carries `changed_by` counts only when `changed_by` equals `created_by` (both null for the owner), so a mark someone else changed after it was made is ignored, the owner's hand included, and the reader's own older mark stands. A mark with `changed_by` but no `created_by` is ignored; a mark with `created_by` and no `changed_by` key gets the author check only; a mark with neither (a folder) counts as before.
1332
- - **The edge rulings now written in the doc comments**, which the readers already answered: a message with no `narrator` is sent by nobody, an empty id (`""`) is addressed by nothing and has no mark, and a thread walk stops at a `parent_narrative` that is not a string. The fixture's 12 new cases run as the message test; 3 of them failed on 4.83.0's readers (a mark changed by another hand, a mismatched sender, an author off the members list), and all 55 pass now.
1333
-
1334
- ## [4.83.0] - 2026-10-26
1335
-
1336
- ### Added
1337
-
1338
- - **The review profile's readers: `proposedMarkOf`, `handsOnAsProposed`, `PROPOSED_MARK` and `OwProposedMark`.** An element whose `x_review_proposed` is an object was written as proposed and is not canon until a keeper settles it. `proposedMarkOf(element, today?)` gives the mark as a reader that shows a world shows it, `{ by, expires, lapsed }`, or null for a settled element (absent, null, or a value that is not an object); `by` and `expires` are trimmed of Unicode White_Space as the room profile trims, `expires` is read only as a day (YYYY-MM-DD, ASCII digits), and `lapsed` is true from that day on, in UTC as keel reads it (`today` defaults to today in UTC; a `today` that is not a day is a TypeError). `handsOnAsProposed(element)` is what a reader that hands a world on leaves out: `x_review_proposed` with any value but absent or null, so a malformed value (a doubt) is left out too (the profile's rule 4). `buildAskContext` now reads the same function. The profile has no fixture yet; the tests hold 17 cases, the Python package's `tests/test_review.py` the same ones, including one per language default that would read differently (a byte order mark, U+001C, Arabic-Indic digits).
1339
-
1340
- ## [4.82.1] - 2026-10-26
1341
-
1342
- ### Changed
1343
-
1344
- - **The draught's words follow keel's 2026-10-26 change.** Under `compact: true` every element body in the packet now leaves out each field whose value is `null`, `""`, `[]` or `{}` (`id`, `type` and `name` always stay; `0` and `false` are values and stay), so a compact body is the list route's body less its empty fields, not byte for byte the same. And a Message without the room mark is placed as any Narrative unless `messages` is named, when it is in `messages` and no other section; a room-marked Message is in `messages` or nowhere. The doc comments on `draught()`, `OwDraughtItem.element` and `OwDraughtCompact` said the body was the list route's exactly; they now say both rules. Types only in words: `OwElement` already lets every field but `id`, `type` and `name` be absent, so no type changed.
1345
-
1346
- ## [4.82.0] - 2026-10-26
1347
-
1348
- ### Added
1349
-
1350
- - **`front_door.my_seats`, and `OwMySeat`** (keel's front door, live 2026-10-26). `me()`, `setMyCharacter()` and an agent's `redeemJoin()` answer now list the live AI and player seats the key's holder answers for, agent seats first: each row is `{ id, kind, name, character }`, with `id` the seat's membership id as `members()` lists it, `kind` `'agent'` or `'player'`, `name` null when the seat has none, and `character` the Character it plays (null when it plays none the key can read). The list is empty for an AI or a player seat's own key and for anyone who answers for none; a seat made from a link for an email address is never listed. Types only: the client already passed the body through. Required on `OwFrontDoor`, as keel's document declares it; a hand-built `OwMe` or `OwAgentJoined` in a consumer's test needs the field.
1351
- - Tests: `me()` reads a front door with seats and one with an empty list through unchanged; the published `OwFrontDoor` declares its six keys in the document's order and `OwMySeat` its four.
1352
- - Test: `readMarkOf` takes the newest read mark wherever it stands in the list. The message profile's fixture lists the newer mark first, so a reader that kept the first mark that counts passed every case; the new test lists it last (and replays the fixture's case reversed), and fails on that reader. No reader changed.
1353
-
1354
- ## [4.81.1] - 2026-10-26
1355
-
1356
- ### Changed
1357
-
1358
- - **Schema repinned to schema-dist `v0.33.0-dist.74`** (from dist.73; the same canonical 00.33.00, so `ONLYWORLDS_VERSION` stays `'00.33.00'`, no schema file moved and no generated type changed). Serial 74 is text only: the dist's README ends on a maintainer line, and six profile pages (checks, names, packet, review, room, spine) no longer name people or point inside the project's own tracker. No reader and no fixture changed.
1359
-
1360
- ## [4.81.0] - 2026-10-26
1361
-
1362
- ### Added
1363
-
1364
- - **`isToEveryone(message)`**: true when a message is addressed to every member (its `characters` absent, null or `[]`). A list that names no valid id, or a value that is not a list, is addressed to nobody and false here, though `recipientsOf` gives `[]` for both; this is the one call that tells them apart.
1365
-
1366
- ## [4.80.0] - 2026-10-26
1367
-
1368
- ### Added
1369
-
1370
- - **The message profile** (schema-dist `profiles/message.md`, serial 72): a message between a world's members is a Narrative with supertype `Message`, from its `narrator` to the Characters in its `characters`. `MESSAGE_PAIR` is that pair, generated from the fixture's `pair` and checked at codegen against the page and against `ROOM_PAIRS` (a parlor's line is a message too). The readers:
1371
- - `isMessage(element)`: the pair, the type and supertype each trimmed and compared without regard to case by the room profile's own trim and fold (`Messages` is not a message).
1372
- - `recipientsOf(message)` and `isAddressedTo(message, characterId)`: an empty `characters` (absent, null or `[]`) is addressed to every member. Nothing else addresses a message: not `narrator`, not a name in the text, not `protagonist` or any other link. A list that names no valid id, or a value that is not a list, addresses nobody, never everyone.
1373
- - `isSentBy(message, characterId)`: the `narrator` is the Character, the claimed sender, proved or not.
1374
- - `senderStatus(message, members)`: `{ status, claimed, writtenBy }`. `verified` when the Character of the membership in `created_by` (on `client.members()`, the owner's row matching a `created_by` of null; a removed member still proves it) is the `narrator`; `mismatch` when it is another, named in `writtenBy`; `unverifiable` with no `narrator`, no `created_by` key (a folder, an export), no members list (pass null), an author not on it, or a row naming no Character. Show a claim that is not verified as a claim, never as a proved sender. `roster` is in the profile's signature and no rule reads it yet.
1375
- - `readMarkOf(relations, characterId, members)`: the reader's read mark, a Relation with supertype `Read` whose `actor` is the reader, counted only in the reader's own hand (`created_by` on the members list; a mark without `created_by` at all, in a folder or an export, counts), the newest `updated_at` of those. A newer mark in another hand is ignored.
1376
- - `unreadFor(messages, characterId, readMark)`: the messages addressed to the reader, not sent by them and not on the mark, in `created_at` order. A message one sent is never unread for them.
1377
- - `threadOf(messages, rootId)`: every message whose chain of `parent_narrative` reaches the root, the root included; a loop in a chain ends the walk. Not named on the profile's page; its thread rule and fixture cases are.
1378
- - Both lists order by `created_at` (microseconds counted), and equal times keep the order of the list given, the server's, so every reader agrees. Never by id.
1379
- - Types `OwSender`, `OwSenderStatus`, `OwMessageMember`. A convention on the 22 types: no type, field or wire change.
1380
- - Tests: the profile's 43-case fixture, read from the vendored distribution, as its own test, with the count of each kind; the documented calls compiled against the built types.
1381
-
1382
- ### Changed
1383
-
1384
- - **Schema repinned to schema-dist v0.33.0-dist.73** (from dist.71; same canonical 00.33.00). Serial 72 adds `profiles/message.md` and `profiles/message.fixture.json`, and serial 73 settles the page's rule 1: a Message Narrative without the room mark stays lore in every lore view, and a message view is a second reading of it; `VERSION`, `README.md` and `MANIFEST.json` carry the new serial; no schema file moved. Generated output moves its provenance lines and gains `MESSAGE_PAIR` and the read mark's pair and fields.
1385
-
1386
- ## [4.79.0] - 2026-10-26
1387
-
1388
- ### Added
1389
-
1390
- - **`revokeWorldKey(worldId, keyId, { signal, keepalive })`**: both go to `fetch` as given. With `keepalive: true` a browser finishes the revoke after the tab that sent it closes, so a page can revoke its key on Sign out and close at once; a `signal` (for example `AbortSignal.timeout(5000)`) puts a time limit on it, and an abort rejects with an `OwNetworkError` whose `cause2` is the abort. A dead account token is a 401 (`err.isAuthError`), and then the key was not revoked. An unknown option is a TypeError before anything is sent.
1391
- - **`request()` takes `signal` and `keepalive`**, passed to `fetch` as given and sent only when given.
1392
-
1393
- ## [4.78.0] - 2026-10-26
1394
-
1395
- ### Added
1396
-
1397
- - **`revokeWorldKey(worldId, keyId)`**: revoke a world key by the `id` `mintWorldKey` returned (`DELETE /account/worlds/{id}/keys/{key_id}`, account token only). The key then answers 401 `key_revoked`. The owner revokes any of the world's keys, a member only their own; any other id is a 404. Revoking a revoked key answers the same 204, so a tool that mints a key when a person signs in can revoke it when they sign out without tracking whether it already did.
1398
-
1399
- ## [4.77.0] - 2026-10-26
1400
-
1401
- ### Added
1402
-
1403
- - **`isCharacterPlayed`** on `OwApiError`: a delete (or a review's decline) of a Character someone plays on the roster, the world's `owner_character` or a live member's, an AI seat's or a player's, answers 422 `character_played` (`param` id) and deletes nothing. The predicate tells it from a malformed id's 422, which carries the same `param`. The way out is to re-point that member to another Character, or remove the member, then delete.
1404
-
1405
- ## [4.76.2] - 2026-10-25
1406
-
1407
- ### Changed
1408
-
1409
- - **Schema repinned to schema-dist v0.33.0-dist.71** (from dist.69; same canonical 00.33.00). The room profile now says what trimmed means: Unicode White_Space removed at both ends, a byte order mark kept. Its fixture gains three cases, 28 in all: a tab and an ideographic space around the name still match, a leading U+FEFF and a trailing U+001C do not. `isRoomElement` already read all 28, so no reader changed; the room test now asserts exactly 28 cases. No generated type changed. Serial 70 carried the same content with a manifest that did not match its files and is pinned in no release.
1410
-
1411
- ## [4.76.1] - 2026-10-25
1412
-
1413
- ### Changed
1414
-
1415
- - **Schema repinned to schema-dist v0.33.0-dist.69** (from dist.68; same canonical 00.33.00). The calendar fixture's `v2-except-chain-gregorian` case gains 12 steps, the days around Leap Day in 1900, 2000 and 2100 (no Leap Day in 1900 and 2100, one in 2000), each checked against LegendKeeper's own reading; a reader whose days ignore `except` while its year count keeps it misreads only these. The test replays all 64 cases, now 198 steps. The script that generates the amendment's cases (`scripts/calendar-fixture-cases.mjs`) adds those century days and keeps the case's earlier sampled steps as published; its 14 cases equal the published ones. No generated type changed.
1416
-
1417
- ## [4.76.0] - 2026-10-25
1418
-
1419
- ### Added
1420
-
1421
- - **Calendar version 2: a leap rule with exceptions and an offset** (the calendar profile's amendment, schema-dist v0.33.0-dist.68). An intercalary entry with `every_years` may carry `except`, divisors that flip the rule in order (`every_years: 4, except: [100, 400]`: every 4, not every 100, yet every 400), and `offset`, added to the year index before the rule (1 for a calendar with no year 0). Gregorian is `every_years: 4, except: [100, 400], offset: 1`. A value that uses either key is `version: 2`; `readCalendarConfig` reads versions 1 and 2, reads the two keys only under version 2, and refuses an `except` that is not strictly ascending with each entry a multiple of the one before (the first of `every_years`), or an `offset` that is not a whole number. `validateCalendarConfig` names each of these, and the keys under version 1. Year arithmetic stays closed-form. `intercalaryInYear(cfg, entry, year)` says whether an intercalary day falls in a year.
1422
-
1423
- ### Changed
1424
-
1425
- - **An era with no `end` runs to the next up era's start**, skipping down eras, whatever the list order. An up era listed before a down era with the same pivot no longer ends where it begins: with eras `After` (up, from 0) and `Before` (down, pivot 0), step 5 reads `6 AF`, where it read `5` with no era.
1426
- - **Schema repinned to schema-dist v0.33.0-dist.68** (from dist.66; same canonical 00.33.00). The calendar fixture's 64 cases are replayed by a test, which no test did before; the import map profile (serial 67) is vendored with the distribution and not read by this package. No generated type changed.
1427
-
1428
- ## [4.75.0] - 2026-10-25
1429
-
1430
- ### Changed
1431
-
1432
- - **Schema repinned to schema-dist v0.33.0-dist.66** (from dist.63; same canonical 00.33.00). `ROOM_PAIRS` gains the room profile's ninth pair, `relation`/`Take` (one member's take on a work item; it also carries `x_room_item`, the item's id), so `isRoomElement` counts a marked Take as a room element. The room fixture's 25 cases are replayed. No other generated type changed.
1433
-
1434
- ## [4.74.0] - 2026-10-25
1435
-
1436
- ### Added
1437
-
1438
- - **A hide's and an unhide's row on `/changes`**: `OwChangeUpsert` names `changed_at`, `changed_by` and `changed_via`, which keel sends only on the upsert row a hide or an unhide gives the readers who still see the element (the owner's keys, a co-builder, the member who made it). The body carries `hidden: true` after a hide and no `hidden` after an unhide; the row's `change_seq` is the act's, the element's own stays in `element`. Every other upsert row carries none of the three.
1439
-
1440
- ## [4.73.0] - 2026-10-25
1441
-
1442
- ### Added
1443
-
1444
- - **`OwSignInResult.tokenId`**: the account token's id from Sign in with OnlyWorlds' exchange answer (`token_id`, live on onlyworlds.com 2026-10-25). A tool done with the token revokes it with `DELETE /account/tokens/{tokenId}`, sending the token itself as the bearer, instead of finding it in the token list by its last four characters. `null` from a server that does not send it yet, never a made-up id.
1445
-
1446
- ## [4.72.4] - 2026-10-25
1447
-
1448
- ### Changed
1449
-
1450
- - **Schema repinned to schema-dist `v0.33.0-dist.63`** (from dist.61; the same canonical 00.33.00, so `ONLYWORLDS_VERSION` stays `'00.33.00'`, no schema file moved and no generated type changed). Serials 62 and 63 change only the dist's folder checker, `walk/lint_folder.py`: it imports the schema walk only when run as a command. This package does not ship it; the repin keeps both client packages on one serial.
1451
-
1452
- ## [4.72.3] - 2026-10-24
1453
-
1454
- ### Changed
1455
-
1456
- - **Schema repinned to schema-dist `v0.33.0-dist.61`** (from dist.59; the same canonical 00.33.00, so `ONLYWORLDS_VERSION` stays `'00.33.00'`, no schema file moved and no generated type changed). Serial 60 rules eight spine readings and adds `x_spine_retires`, the profile's thirteenth key; serial 61 gives `schema.json` the advisory minimum and maximum of the 15 integer fields that have them (this package's types and `FIELD_SCHEMA` are generated from the field list, which does not carry them), rules two more spine readings with four fixture cases, and has the folder lint name the format's version. The SDK reads no spine key; the profile pages it replays (names, room, review) moved only their ruled-when lines.
1457
-
1458
- ## [4.72.2] - 2026-10-24
1459
-
1460
- ### Fixed
1461
-
1462
- - **`OwElementBase.hidden`'s doc comment says what a create does with it.** It said a write ignores the mark. Since 2026-10-23, a create (a POST, a PUT of a new id, a `/bulk` item that creates) carrying `hidden: true` creates the element hidden, from the owner's keys or a co-builder's; any other key gets 403 `hidden_not_allowed` and nothing is created. Every other write still ignores it. The SDK already sent the mark as the element had it; only the words were wrong, in the types an editor shows (`dist/index.d.ts`). (Corrected after release: the entry first said they reach the docs' TypeScript reference, which lists the methods and not this field.)
1463
- - Tests: `create`, `upsert` and `bulk` send `hidden` as given (it is not on the stripped list); a 403 `hidden_not_allowed` surfaces as `OwApiError.code`.
1464
-
1465
- ## [4.72.1] - 2026-10-24
1466
-
1467
- ### Fixed
1468
-
1469
- - **`readWorldFile` refuses a deflate stream that ends before its entry's declared compressed size** (folder spec §4.7, OW-14, its size bullet; ruled 2026-10-24, published at schema-dist serial 58). The size bullet speaks of every size an entry declares, the compressed one too. fflate stopped at the stream's end and ignored the bytes after it, so an entry whose stream ended four bytes early, those bytes junk and its sizes and CRC honest, was read (`fail-OW-14-deflate-early-end`); it is now `not_openable`. fflate never says where a stream ended, so each deflated entry is inflated twice: once whole, as before, then again short of its last byte (in 64 KiB chunks past that size, the output dropped), where a stream that ends at its declared size runs out and fflate refuses it, and one that ends sooner still inflates. A stream that never ends (`fail-OW-14-deflate-no-end`) was already refused, as `not_openable`. A world file whose streams end where their entries say reads as before.
1470
- - Tests: the whole `.owworld` set, 96 cases, through the reader, the count floor 96; built streams with 1 and 4 junk bytes after their end, short and past 64 KiB, refused, and the same streams with nothing after them read. The three zips the fix was tested on before the serial was vendored (`test/fixtures/owworld-next/`) are gone: the set carries them.
1471
-
1472
- ### Changed
1473
-
1474
- - **Schema repinned to schema-dist `v0.33.0-dist.59`** (commit a40a5f3; same canonical 00.33.00; no schema file moved, `schema.json` and `presentation.json` unchanged, no generated type changed, `ONLYWORLDS_VERSION` still `'00.33.00'`; generated output moves only its source header and SCHEMA.md's line). Serials 58 and 59 together bring what follows; 59's own change is text only (the set's generator and verdicts regenerated with the same 96 zips byte for byte, and the profile pages, the checks fixture's note and the folder lint say when a rule was ruled). The `.owworld` set at 96 zips: the two deflate cases above, and `pass-OW-18-nfc-unassigned`, U+11383 beside U+11382 U+113C9, two names at Unicode 15.1, which 4.71.0's `nfc` already keeps apart; none of the earlier 93 zips moved. The spine profile's serial-58 sentences: `x_spine_owner` the keeping seat's name, the projection's World mark `x_spine_projection`, the audit's one Seat and folded-name uniqueness, the Retirement Event's typed link. The dist's README, LICENSE and SECURITY.md, written by its publisher.
1475
-
1476
- ## [4.72.0] - 2026-10-24
1477
-
1478
- ### Changed
1479
-
1480
- - **`buildAskContext` now leaves proposed, hidden and room elements out of the context unless asked.** A caller that passes such elements gets fewer items in `context` and `kept` than before: an element whose `x_review_proposed` is anything but absent or null (the review profile, rules 1 and 4: a value that is not a proper mark is a doubt and counts as proposed), one whose `hidden` is anything but absent, null or false (folder format §4.8, the same test as `writeWorldFile`'s `includeHidden`), and a room element (`isRoomElement`, the room profile, rules 1 and 4) are not sent. Their ids come back in the new `leftOut: { proposed, hidden, room }`, in the order given, so a tool can say how many it held back. An element carrying more than one mark is listed once, in the first of hidden, proposed, room whose option is off: a hidden proposed element is under `hidden`, so `includeProposed` alone never sends it. A sent element's link to an element a mark leaves out, anywhere in the index, reads as "not present in the world data", never by its name. `dropped` keeps its meaning (the 8 items and the 12,000 characters), counted over what remains. Until now the SDK relied on the caller (Lantern) to filter first; the rules bind the exported function itself.
1481
-
1482
- ### Added
1483
-
1484
- - **`buildAskContext(items, index, { includeProposed, includeHidden, includeRoom })`**, each default false: true sends that kind again. A proposed element sent carries the review profile's label on its first line, `[proposed, not yet reviewed] Name (type)`, where the AI reading the text sees it. An option that is not a boolean is a TypeError, and an unknown key is refused by name (the options fence, `test/options.test.mjs`). Types `OwAskContextOptions` and `OwAskLeftOut`.
1485
- - **`REVIEW_PROPOSED_LABEL`** (`'proposed, not yet reviewed'`), generated into `types.generated.ts` from rule 1's own sentence in the vendored `profiles/review.md` ("labels each one it keeps **"…"**"); codegen exits 1 if the sentence is missing or occurs twice.
1486
- - **`minted_by: string | null` on `OwWorldLink` and `OwCreatedWorldLink`**: the membership (its id in GET /api/v2/members) of the co-builder who minted the link over POST /api/v2/links; null for a link the owner made. Served by `createWorldLink` and `listWorldLinks`. No client for `/api/v2/links` itself yet.
1487
- - Tests: each mark left out by default and listed; `hidden` as §4.8 reads it (`"true"` and 1 out, false and null in); a doubt (`"yes"`, `{}`) treated as proposed and labelled when asked, a cleared `null` settled; each `include*` bringing back its own kind only; the label on the first line; a hidden proposed element counted once, under hidden; a kept element's links to a hidden element and from a room element read without their names; the limits over what remains; the fence and the boolean check. Ten mutants, each red on its own test: no proposed filter, no hidden filter, no room filter, no label, a doubt not proposed, proposed checked before hidden, left-out elements counted toward the total, no link view, no option fence, no boolean check.
1488
-
1489
- ## [4.71.0] - 2026-10-24
1490
-
1491
- ### Added
1492
-
1493
- - **`nfc(text)`**: Normalization Form C as Unicode 15.1 defines it, built from Unicode's own `UnicodeData-15.1.0.txt` and `CompositionExclusions-15.1.0.txt` (schema-dist serial 57), never the runtime's. It is the NFC half of OW-18's fold, beside `simpleCaseFold`, exported so a tool that compares names the way a world file does gets the same answer. This export is why the release is a minor, not 4.70.1.
1494
-
1495
- ### Fixed
1496
-
1497
- - **The world file's name check uses Unicode 15.1's NFC, not the runtime's** (folder spec §4.7, OW-18, ruled at serial 57). `readWorldFile` and `writeWorldFile` folded names with `String.prototype.normalize('NFC')`, which follows the runtime's ICU: Node 22's is Unicode 16.0, which composes 20 code points 15.1 leaves unassigned (Todhri, Tulu-Tigalari, Gurung Khema, Kirat Rai), so a name with U+11383 and one with U+11382 U+113C9 were refused as one name where the spec keeps two, and an older runtime could miss what 15.1 added. Both now fold with `nfc`. Where 15.1 and the runtime both assign a code point, nothing changes (Unicode's normalization stability): the set's 93 cases read the same either way. The tables are `src/nfc-tables.generated.ts`, made by `codegen/nfc_tables.mjs` from the vendored files.
1498
-
1499
- ### Changed
1500
-
1501
- - **Schema repinned to schema-dist `v0.33.0-dist.57`** (commit e21b867; same canonical 00.33.00; no schema file moved, `schema.json` and `presentation.json` unchanged, no generated type changed, `ONLYWORLDS_VERSION` still `'00.33.00'`; generated output moves only its source header and SCHEMA.md's line). Serials 50 to 57: the export fixture's world block is the World whole, with `origin` and `x_names` (50, 51); the review profile's reader rule, a reader that hands a world on leaves proposed elements out unless asked (52, 53); the folder lint takes `hidden` as a folder key (53); the checks profile's list case and reader rules 6 to 8 (54); the spine profile and its fixture (55, 56); the `.owworld` set at 93 zips, eight OW-18 NFC pairs with `UnicodeData-15.1.0.txt` and `CompositionExclusions-15.1.0.txt` beside `CaseFolding-15.1.0.txt` (57). None of the earlier 85 zips moved.
1502
- - Tests: the NFC against the two files for every code point (each composite composed back unless excluded, its mapping to the same NFC, each Hangul syllable from its jamo, each combining class by where ordering puts the mark), against the runtime wherever both assign a code point, and every code point 15.1 leaves unassigned left alone; U+11383 against U+11382 U+113C9 and U+16D68 against U+16D67 U+16D67 accepted as two names. The fixture count is the published 93. Four mutants (the runtime's `normalize` back in the fold, one exclusion dropped from the table, an unstable ordering, no Hangul LV+T), each red on its own line. Checked outside the repo against Unicode's `NormalizationTest-15.1.0.txt`: 19,074 lines, 0 disagreements.
1503
-
1504
- ## [4.70.0] - 2026-10-23
1505
-
1506
- ### Added
1507
-
1508
- - **`writeWorldFile(files, topDir, { includeHidden: false })`** (folder spec §4.8, Skeld, 2026-10-23): a world file is a copy, so an element file carrying the hidden mark travels byte for byte by default, as before. A tool making a file to hand to someone else passes `includeHidden: false`, and every `.json` file directly in a type folder of `elements/` or the legacy `spatial/` whose JSON has a top-level `hidden` (anything but absent, `null` or `false`, so `"true"` and `1` count, as the Python package's `is_hidden`) is left out whole. A file there that is not UTF-8 JSON, or not an object, is kept. Nothing else is rewritten: a visible element's link to a hidden one keeps its id. The two defaults differ on purpose: private records out unless asked, hidden kept unless the write is a share.
1509
- - **`report: (r: OwWriteWorldFileReport) => void`**, a `writeWorldFile` option: called once with `{ hiddenLeftOut }` after the archive is made and before it is returned, never on a refusal. The return stays the bytes. An `includeHidden` that is not a boolean, or a `report` that is not a function, is a `TypeError` before anything is written; both keys join the options fence.
1510
-
1511
- ### Fixed
1512
-
1513
- - **`draught`'s doc comment names `compact`** (4.69.0), so the docs' generated reference row carries it.
1514
- - Tests: a hidden element file byte-identical by default and with `includeHidden: true`, the report 0; left out as a share, the report 1, every other file byte-identical; the legacy layout, `.JSON`, and a doubtful mark left out; not JSON, not UTF-8, not an object, a null or nested mark, a deeper file, a non-JSON file and the other parts kept; the same bytes for the same files either way; the report not called on a refusal; the fence. Watched failing with the leave-out turned into a keep (3 tests red), restored. The 85-case `.owworld` set unchanged.
1515
-
1516
- ## [4.69.0] - 2026-10-23
1517
-
1518
- ### Added
1519
-
1520
- - **`draught(id, { compact: true })`**: the draught without its empty sections. Every empty array or object section and an empty `cut` are left out. The frame (`world`, `calendar`, `as_of`, `now`, `mode`, `policy`) and every null stay, so a null is still an answer: `character` null means she is hidden from this reader, `style` null that she has no style element, `recent_cursor` null that the read had no `since`. An absent section means nothing in it, never that it was not read (`policy.components` says what was). The answer is typed `OwDraughtCompact` (`cut` optional); without `compact`, or with `compact: false`, it is `OwDraught` as before, the fixed shape. A `compact` that is not a boolean is a TypeError before anything is sent. MCP's `get_draught` defaults to compact; this route defaults to the fixed shape.
1521
-
1522
- ## [4.68.0] - 2026-10-23
1523
-
1524
- ### Fixed
1525
-
1526
- - **`patchWorld` strips `counts`**, which a keyless `GET /world` carries (the public reader, on since 2026-10-23) and `PATCH` answers as an unknown key, so the body of a keyless read goes back too. Quillon's port found it (2026-10-23); the Python package's live wire test went red on the same key the same hour.
1527
- - **`reviewNewElement` refuses a reason with an accept** before sending (keel takes `{ decision: 'accept' }` alone; a reason was its 422), and **checks its answer**: an accept that answers no `element`, or a decline no `deleted`, is a `TypeError` naming the route. Both as the Python package does (Quillon, 2026-10-23).
1528
-
1529
- ## [4.67.0] - 2026-10-23
1530
-
1531
- ### Added
1532
-
1533
- - **The draught's `room` and `proposed` components**: `OwDraughtComponent` is now keel's thirteen, in its order. `room` is the room profile's other elements that reach her (a board entry, a spec, a heart, a vote, a sitting, a member card, a work item that is not open or taken), each there or nowhere; the type missed it since keel shipped it. `proposed` (live 2026-10-23) is every element carrying `x_review_proposed` that reaches her, there or nowhere, whatever section would have held it; `cut.proposed` counts the ones left out, the component unnamed or its budget spent, so a tool handing a draught on can say so. Both are off unless named, as `messages` is: `draught(id, { components: [..., 'room', 'proposed'] })`. `OwDraught.room` and `OwDraught.proposed` are typed.
1534
-
1535
- ### Fixed
1536
-
1537
- - **`OwDraughtItem` nests the element**: keel has served each draught row as `{ via, circles?, her_links, grade, dated?, element }` since the route shipped, and the type spread the element's fields onto the row instead. Read `row.element.name`, not `row.name` (which was `undefined` at runtime all along); TypeScript code that read the spread form now fails to compile where it was already wrong. In the same way `OwDraught.character` is `{ self, grade, element }`, and optional (absent unless `self` is served), `work` rows carry `packet` (the work item's `x_packet_elements`), `messages` rows carry `author` (`membership`, `kind`, `role`), and rows may carry `circles`. Read from keel's `elements/draught.py` and its OpenAPI `Draught` schema; nothing at runtime changed.
1538
- - Tests: `components: ['room', 'proposed']` is sent as `components=room,proposed`; the published `OwDraughtComponent` lists keel's COMPONENTS in order; the nested rows, `cut.proposed`, `character.element`, `packet`, `author` and `circles` compile against the built `.d.ts`, and a spread read and an unknown component do not. Watched failing with `'proposed'` cut from the type.
1539
-
1540
- ## [4.66.0] - 2026-10-23
1541
-
1542
- ### Added
1543
-
1544
- - **`createAccountWorld({ name, pin?, keyName?, source?, producer? })`** (account token): `POST /account/worlds` makes a world on the account with its one write key. Sent in keel's spelling, only the keys given (`keyName` as `key_name`: 1 to 80 characters, trimmed, `null` for none; `source` and `producer`: the folder format's provenance lines, each an object or `null`, kept on the World and served by `getWorld()` as `origin`). `pin` only while the account has no PIN (it becomes the account PIN); an account with one must leave it out. Either mistake is a 422 with `err.param === 'pin'` and nothing made, so a tool that cannot tell which account it holds sends without `pin` first and again with it on that 422, as Atlas's Make online does. Returns `OwCreatedAccountWorld`: the world's row as `listAccountWorlds` lists it, plus the full write key once in `key`. No Idempotency-Key (the document declares none), and neither `X-OW-Agent` nor `X-OW-Client` (it declares neither here). keel's document gives the 201 no schema, so the type is read from keel's handler. A world key is a `TypeError`, as is an unknown option or a missing `name`, before anything is sent. Atlas can retire its own `AccountClient.createWorld` for it.
1545
- - **The world's queue lists new elements written as proposed** (live 2026-10-23): `worldProposals()` now returns `OwWorldProposalList`, whose rows are `OwWorldProposalRow`, told apart by `kind`: `OwProposal` (`kind: 'proposal'`) or `OwProposedNewElement` (`kind: 'new_element'`: an element carrying `x_review_proposed`, its `by`, `evidence` and `expires` the mark's words, `state` `open` or `lapsed`). One order, newest first by `created_at`; the cursor is opaque and goes back as it came (an old bare proposal id still reads). `listProposals` keeps `OwProposalList`: one element's proposals only.
1546
- - **`reviewNewElement(type, id, { decision, reason? })`**: `POST /{type}/{id}/review`, a keeper (the owner or a co-builder, with the PIN) settles a new element written as proposed. `accept` clears the mark and returns `{ element }`; `decline` deletes the element and returns `{ deleted }` (its id); the reason is answered back and kept nowhere. Sends `X-OW-Client`, as the document lists, and no `X-OW-Agent`. A `decision` other than `accept` or `decline`, or an unknown key, is a `TypeError` before sending. Type `OwReviewed`, options `OwReviewInput`.
1547
- - **`err.isNotProposed`**: 422 `not_proposed`, the element carries no mark to settle. `isProposedOnly` (403) also answers the seat the mark names.
1548
- - **`OwProposal.kind`** (`'proposal'`) and **`OwProposal.omitted`**: a row's `fields` now holds only what differs from the element when proposed, and `omitted` names the fields an upsert left out that were not empty (each in `fields` at its empty value, so an accept empties it; `[]` for a patch, a link edit or `propose`). A propose with nothing new is a 422 with `param: 'fields'`; a seat's write that changes nothing under the world's switch answers 200 with the element and makes no row.
1549
-
1550
- ### Changed
1551
-
1552
- - `worldProposals()`'s answer type is `OwWorldProposalList` instead of `OwProposalList`: TypeScript code that reads a proposal's own keys (`fields`, `base`) off a queue row now narrows on `row.kind === 'proposal'` first. The wire changed under it on keel's side: the 4.65.0 type described rows keel no longer sends alone.
1553
- - Tests: the body sent for each `createAccountWorld` option alone and together, `null` included, an `undefined` option left out; the PIN dance's 422 surfacing with `err.param` `'pin'`, then the retry with it; the headers on the create (no agent, client or idempotency header); the queue's two kinds and its cursor; both review decisions, their headers and their refusals; the published types. Watched failing on 30 mutants, each restored byte for byte.
1554
-
1555
- ## [4.65.0] - 2026-10-23
1556
-
1557
- ### Fixed
1558
-
1559
- - **`patchWorld` takes a `getWorld()` body back again** (live 2026-10-23). keel's `GET /world` now serves `origin`, and a PATCH naming it is a 422 (an unknown key), so a world body read and sent back failed. `patchWorld` now leaves out `origin`, and `id` too: keel answers `id` with the same 422 (it is not a key PATCH takes), so the round trip 4.57.1's entry promised never held for a whole body. `owner_character` is still sent.
1560
-
1561
- ### Added
1562
-
1563
- - **`OwWorldMeta.origin?: OwWorldOrigin`**, read-only: where the world first came from, `{ source, producer }`, both keys always present, each the folder format's provenance line (spec 4.2) as the tool that made the world from a folder gave it, or `null`. Types `OwWorldOrigin`, `OwWorldSource` (`server`, `world`, `at`) and `OwWorldProducer` (`tool`, `version`, `input`), each open to keys keel keeps as given. The origin is written as the plain keys `source` and `producer`: `patchWorld({ source, producer })` sends them as given (an object, or `null` to clear).
1564
- - **The World's extension keys on `OwWorldMeta`**: `x_*`, `atlas_*` and `shadow_*`, typed `unknown` by template-literal index signatures (the first is `x_names`, the names profile `typeWord` reads). `patchWorld` sends every one as given, `null` included (keel removes the key); over keel's 64 KB cap on the whole store the answer is a 422 with `param: extensions`.
1565
- - Tests: a full `GET /world` body sent back sends neither `origin` nor `id` and keeps `x_names`; `source`, `producer` and a key of each extension prefix pass untouched, `null` included; the published types declare `origin` and its two keys, and the documented calls (`patchWorld(await getWorld())`, `patchWorld({ source, producer, x_names })`, `typeWord(world, …)`) compile against the built `.d.ts`. Watched failing on mutants: `origin` and `id` each out of the strip list, `origin` out of `OwWorldMeta`, and a narrowed `patchWorld` parameter.
1566
-
1567
- ## [4.64.0] - 2026-10-23
1568
-
1569
- ### Changed
1570
-
1571
- - **Every options object in the package now refuses an unknown key**, not only the client's typed methods (4.51.0 to 4.53.0): `readWorldFile`, `parseDate`, `resolveCalendarLens`, `lookupWorld`, `renderElement`, `planPhoto`, `preparePhoto`, the client's `request()` escape hatch, `listAll` (with its own keys: a `cursor` it was handed was overwritten unread), the deprecated `tokens.consume` and `new OwSignInError`'s third argument throw a `TypeError` naming the function, the key and the keys it takes, before anything is read, decoded, parsed or sent. A dropped key was never harmless: `readWorldFile(bytes, { maxbytes: 1e6 })` read with the 2 GiB budget, `parseDate(cfg, input, { Now: 12 })` parsed as if the world had no now, and W11's `list('character', { name__icontains })` read every Character and a cleanup script deleted 99 of them (2026-10-22). A plain-JavaScript caller passing a misspelt key will now see a `TypeError` where the call used to run with the default; TypeScript callers were already refused at compile time. `writeWorldFile` and `typeWord` keep their refusals on the same shared check; `typeWord`'s message now reads like the others (`typeWord: unknown option 'plural'. The keys are supertype, many.`).
1572
- - Tests: one test calls every options parameter of the exported surface with `{ __unknown__: 1 }` and asserts a `TypeError` naming the key, nothing sent; a second reads `src/index.ts` with the TypeScript checker and fails on any exported object parameter that is neither in that list nor named as data (a calendar config, a world index, an element), so a new options object cannot ship unfenced. Watched failing on mutants of the `readWorldFile`, `parseDate`, `preparePhoto` and `request()` fences, and on an exported options function added without one.
1573
-
1574
- ## [4.63.0] - 2026-10-23
1575
-
1576
- ### Added
1577
-
1578
- - **`writeWorldFile(files, topDir, { includePrivateRecords })`**, OW-8's private clause (folder format spec v0.3.15, §4.5 rule 11 and §4.7). By default the writer now leaves out, whole, every `records/<producer>/` directory whose entry in `records/RECORDS.json` says `"private": true`, so a tool's own notes do not travel the first time a world is shared. `RECORDS.json` itself travels byte-exact, its private entries kept, so a reader can see a producer was left out. Only the boolean `true` marks a directory private (`"true"`, a string, does not). A producer's name matches its directory by the archive's one fold (NFC, then simple case folding), so a manifest's `Fold` leaves out `records/fold/` on a disk that ignores case. With `includePrivateRecords: true` every record travels as before and the manifest is not read. No manifest, nothing is private. Type `OwWriteWorldFileOptions`.
1579
- - **A manifest the writer cannot read is refused, not guessed at**: a `RECORDS.json` that is not JSON, not an array, holds an entry that is not an object, or marks an entry private without a producer name throws a `TypeError` naming the file and the option, having written nothing, unless `includePrivateRecords: true` is passed. A world whose manifest is in an older shape now needs the option to be written. The options object refuses an unknown key, and an `includePrivateRecords` that is not a boolean, with a `TypeError`, before anything else.
1580
- - Tests: the conformance set's four private cases, written from its `sources.private` world (each source file checked by its sha256): the default side equals `pass-OW-8-private-left-out` and the asked side `pass-OW-8-private-asked`, entry for entry, and neither is `fail-OW-8-private-partial` or `fail-OW-8-private-manifest-rewritten`. Also the string `"true"`, a producer whose name is a prefix of another's, no manifest, each unreadable manifest refused and carried when asked, and the options. Each behaviour watched failing on its own mutant.
1581
-
1582
- ### Changed
1583
-
1584
- - **Schema repinned to schema-dist `v0.33.0-dist.49`** (same canonical 00.33.00; no schema file moved, no generated type changed, `ONLYWORLDS_VERSION` still `'00.33.00'`). The `.owworld` conformance set grows to 85 zips with the four OW-8 private-records writer cases on `sources.private`; none of the earlier 81 moved.
1585
-
1586
- ## [4.62.0] - 2026-10-22
1587
-
1588
- ### Added
1589
-
1590
- - **`typeWord(world, type, { supertype, many })`**, the names profile's reader (schema-dist `profiles/names.md`): a world's own word for a type, from `x_names` on the World (a `getWorld()` body, keel serving `x_` keys on the World, or a folder's `world.json`). The supertype's entry first, matched trimmed and without regard to case (the room profile's one definition, shared with `isRoomElement`), then the type's entry, else `null`: print the standard's name. `many: true` gives the plural, `one` when the pair has none. A value that is not a names map is `null` everywhere; a bad entry or an unknown slug is ignored. The options object refuses an unknown key with a `TypeError`. Types `OwNamesMap` and `OwWordPair`. Display only: the API, the folder, the schema and every id keep the standard's names.
1591
- - Tests: the profile's 23-case fixture, read from the vendored distribution, as its own test; no world, no map and an unknown option.
1592
-
1593
- ### Changed
1594
-
1595
- - **Schema repinned to schema-dist `v0.33.0-dist.48`** (same canonical 00.33.00; no schema file moved, no generated type changed, `ONLYWORLDS_VERSION` still `'00.33.00'`). The distribution gains `profiles/names.md` and `profiles/names.fixture.json`, the names profile, and `profiles/checks.md` names the Charter by keel's D152: the newest Law with supertype `Charter` made by the owner or a co-builder, by `created_at` then id.
1596
-
1597
- ## [4.61.0] - 2026-10-22
1598
-
1599
- ### Added
1600
-
1601
- - **`delete(type, id, { ifMatch })`** (live 2026-10-22 21:3x): `DELETE` reads `If-Match` exactly as `PATCH` does. With the element's `change_seq` as you read it, a delete of an element someone changed since is refused, `OwPreconditionFailedError` (412) with `current`, nothing deleted; on an element already gone it is the idempotent 204. Under a world's proposals switch a delete is never a proposal. The options object refuses an unknown key like every other.
1602
- - Tests: the header sent and left out, the unknown option, the 412; each watched failing with the header dropped. The wire gate now reads from keel's document which Character operations take `If-Match` (it held a hand list of two and missed this one) and fails on 4.60.0.
1603
-
1604
- ## [4.60.0] - 2026-10-22
1605
-
1606
- ### Added
1607
-
1608
- - **`OwFrontDoor`, and `front_door` on `OwMe` and `OwAgentJoined`** (live 2026-10-22 20:1x). `me()`, `setMyCharacter()` and an agent's `redeemJoin()` answer carry the world's front door for the key: `sponsor` (`{id, kind, name}` of the person who answers for a seat, or null), `sponsor_character`, `my_character`, `charter` (the world's Charter, every member's to read: read it first) and `charter_skill` (up to five Objects the Charter names). Every id is one the key may read; anything hidden is null or left out. Types only: the client already passed the body through. Required on both types, as keel's document declares it; a hand-built `OwMe` in a consumer's test needs the field.
1609
-
1610
- ## [4.59.0] - 2026-10-22
1611
-
1612
- ### Changed
1613
-
1614
- - **`API-Pin` is sent only where keel checks it** (Skeld's ruling, 2026-10-22). keel verifies a PIN wherever one is sent (argon2, 64 MiB each), and on 2026-10-22 scripts reading with the PIN OOM-killed staging on a cold cache after a restart. The client used to send a configured `apiPin` on every request. It now sends it on every request whose method is not GET (every write: elements, links, bulk, adopt, hide and unhide, media and file tickets, seats, world links, proposals, `setMyCharacter`, `mintWorldKey`, `patchWorld`, and any route added later), on `GET /me` (`me()`, the connection test: keel checks a PIN there when sent), and on every request of a 10-digit legacy key (v1 parity: such a key needs the PIN on reads of a world with a PIN). Any other read from an `ow_w_`, `ow_r_` or member key goes without it; an account token never sent it and still does not. The rule is derived in `request()` from the method and path, so a raw `request()` call follows it too. No API changed.
1615
- - Tests: `test/api-pin.test.mjs`, five, on a recorded fetch: eight reads, `me()` and ten writes across POST, PATCH, PUT and DELETE, for an `ow_w_` key with a PIN, a legacy key with a PIN, an `ow_r_` key with none and an account token with one configured; and raw `request()` GETs to other paths, `/me` and other methods. Two existing tests that expected the PIN on a GET now expect none. Mutants: the PIN on every request again fails four tests; without the `/me` clause, three; without the legacy clause, one.
1616
-
1617
- ## [4.58.0] - 2026-10-22
1618
-
1619
- ### Added
1620
-
1621
- - **The proposed change** (live 2026-10-22 16:25). Six calls: `propose(type, id, { fields, evidence, expires? })` (any member key that reads the element; `fields` cleaned as a PATCH body is; `Idempotency-Key` only when given), `listProposals(type, id, { limit?, cursor? })`, `worldProposals({ state?, since?, element?, limit?, cursor? })` (the queue, newest first; keepers see every row, others their own), `acceptProposal(type, id, pid)` (`{ proposal, element }`: the keeper's own write, one feed row), `declineProposal(type, id, pid, { reason? })` and `withdrawProposal(type, id, pid)` (204). Types `OwProposal`, `OwProposalList`, `OwProposed`, `OwProposalState`, `OwProposalQuery`, `OwWorldProposalQuery`, `OwProposalAccepted`. Every options object refuses an unknown key with a `TypeError` before anything is sent. `X-OW-Client` goes on `propose` and accept, where the document lists it; `X-OW-Agent` on none of them.
1622
- - **A write held as a proposal is an error, never a body.** Under the world's **AI seats propose, a person accepts** switch an agent seat's `patch`, `upsert` or `editLinks` answers `202`: the client throws `OwProposedError` (`status` 202, `proposal`, `isProposed`) rather than returning something that is not the element; a `202` with no proposal in its body is a plain `OwApiError` 202. A `/bulk` slot with `status: 202` is left in the result as the document shapes it, typed. `create`, `adopt` and `delete` are not read for a 202: the document gives them none.
1623
- - **Five codes**: `isStaleProposal` (`OwStaleProposalError` with `conflicts: [{ field, base, current }]` and `current`, parsed like the 412), `isOwnProposal`, `isProposedOnly`, `isTooManyOpen`, `isFieldNotVisible`.
1624
- - **`OwWorldMeta.agent_writes_proposed?: boolean`**, read-only; `patchWorld` strips it with `slug`, `public_page`, `public_read` and `agent_narrator_rule`.
1625
- - Tests: 20 on a recorded fetch, 18 red first; two mutants caught. Live on production with the demo key, reads only: `getWorld()` carries `agent_writes_proposed: false`; the queue answers `404` to a read key with no member, as the document says. Not exercised live yet: `propose`, accept, decline, withdraw and the `409` (staging does not serve the proposed change at release time; a round-trip script waits for it). Not wired yet: the queue's `ETag`/`If-None-Match` (a conditional `worldProposals`, on the `changesHead` pattern, when a poller asks).
1626
-
1627
- ## [4.57.2] - 2026-10-22
1628
-
1629
- ### Changed
1630
-
1631
- - **Schema repinned to schema-dist `v0.33.0-dist.47`** (same canonical 00.33.00; no schema file moved, no generated type changed). The distribution gains `profiles/checks.md` and `profiles/checks.fixture.json`, the checks profile: a work world's Charter carries its house rules as `x_checks_rules`, beside the sentences a person reads. `profiles/room.md` now defines "trimmed and without regard to case" once, for every profile that folds like the room's pairs: trimmed is leading and trailing whitespace removed, nothing inside touched; without regard to case is the trimmed value after simple case folding.
1632
-
1633
- ### Fixed
1634
-
1635
- - **`isRoomElement` trims Unicode White_Space exactly**, the 25 code points of the property, at both ends of the type and the supertype. It used JavaScript's `trim()`, which also strips U+FEFF, a byte order mark. Skeld's ruling (2026-10-22) is that a byte order mark stays, so a supertype `"\uFEFFMessage"` is no longer a room element; `" Message\u3000"` still is, and a leading U+200B (zero width space, not White_Space) still is not. The room fixture's 23 cases answer as before.
1636
-
1637
- ## [4.57.1] - 2026-10-22
1638
-
1639
- ### Fixed
1640
-
1641
- - **`patchWorld` strips `public_read` and `agent_narrator_rule`** beside `slug` and `public_page`: keel answers both as unknown fields (a 422 naming the key; `agent_narrator_rule` measured on staging, `public_read` by keel's own rule), so a world body read with `getWorld` and sent back now round-trips. `owner_character` is accepted by the server and is still sent.
1642
-
1643
- ## [4.57.0] - 2026-10-22
1644
-
1645
- ### Added
1646
-
1647
- - **The conditional head poll** (live 2026-10-22): `changesHead()` keeps the last head read's `ETag` per client and sends it back as `If-None-Match` on the next one, exactly as received (keel's 200 sends a weak `W/"821:-"`; nothing is quoted, parsed or compared). keel's tag is `"<head>:<epoch>"`, so a view change moves it as a new change does. When nothing moved, keel answers 304 Not Modified with no body, and `changesHead()` returns the `head`, `cursor` and `epoch` of the last 200 with `unchanged: true`; every 200 says `unchanged: false`. keel refunds a 304 against the request budget, so a poll that mostly finds nothing costs little. A 409 `resync_required` from `changesHead()`, `changes()` or `changesAll()` clears the held tag, so the next head read is a full 200 with `epoch_reason`. A 304 while no tag is held is an `OwApiError` with status 304, never a made-up head. The client still never polls or sleeps on its own: the interval is the caller's.
1648
- - **The public page on `OwWorldMeta`** (live 2026-10-22), both read-only: `slug` (`string | null`), the page's address, onlyworlds.com/w/<slug>, null while the page is off, and also a keyless `?world=` value; `public_page` (the new `OwWorldPublicPage`), the page as its owner set it: `on`, `palette` (midnight, daylight, parchment, brass, dusk, or null for the house's), `hero` (one plain line), `sections` (the types shown in order, empty for all), `pin` (an element id or null) and `index` (always false for now), every key present with the page off. Both are set by the owner on the world page; `patchWorld()` strips them, so a body read from `getWorld()` stays writable.
1649
- - **`request()` gains two options**: `ifNoneMatch` (a tag sent as `If-None-Match` as given) and `onEtag(etag, status)` (called with each response's `ETag` and status before the body is read; with it given, a 304 returns null instead of throwing). Every other path is unchanged.
1650
-
1651
- ### Changed
1652
-
1653
- - **`changesHead()`'s return type gains `unchanged: boolean`**, beside `head`, `cursor`, `epoch` and the optional `epoch_reason`.
1654
-
1655
- ## [4.56.0] - 2026-10-22
1656
-
1657
- ### Added
1658
-
1659
- - **The client's name** (live 2026-10-22): a constructor option `client`, default `'sdk'`, sent as `X-OW-Client` on every element write (`create`, `upsert`, `patch`, `delete`, `editLinks`, `bulk`, `adopt`) and on no read. keel stores it on the element as `changed_client`, so a reader can tell which tool made the last change (`new OwV2Client({ ..., client: 'atlas' })`). The rule is keel's, exact here: lower case letters, digits and hyphens, 1 to 32 characters (`[a-z0-9-]{1,32}`). Anything else, a capital, a space, an empty string, 33 characters or a value that is not a string, is a TypeError at construction naming the option and the rule, and nothing is trimmed; keel's own fence for a stranger's header is to store null, never a 422. The name is the client's claim about itself, never proof; it changes no door and no permission, and `agent` still decides `X-OW-Agent`.
1660
- - **`changed_client`** on `OwElementBase` (`string | null`, server-managed, read-only, beside `changed_by` and `changed_via`): the name the last write gave, null when it gave none or one outside the rule, for keel's own writes, and for changes from before names were recorded. Sent back in a write it is ignored. **`deleted_client`** on a `/changes` delete (`OwChangeDelete`, beside `deleted_by` and `deleted_via`), the same for the delete; null on a `hidden` row.
1661
-
1662
- ### Changed
1663
-
1664
- - **The constructor refuses an unknown key by name**, as every options object has since 4.52.0: `new OwV2Client({ clinet: 'atlas' })` is a `TypeError` naming the key and the eight the constructor knows, where it fell back to the defaults silently before. Every known caller (Atlas, Lantern, the toolkit) passes known keys only.
1665
- - **Schema repinned to schema-dist v0.33.0-dist.46** (same canonical 00.33.00). The dist's example export (`example/world-export.json`) gains `changed_client: null` on its bodies, `walk/lint_folder.py`'s server-key set names `changed_client`, and the README's sentence on server keys names it; `VERSION` and `MANIFEST.json` carry the new serial; no schema file moved. Generated output moves its provenance lines and gains `changed_client`.
1666
-
1667
- ## [4.55.0] - 2026-10-22
1668
-
1669
- ### Added
1670
-
1671
- - **The room profile** (schema-dist `profiles/room.md`, serial 45; Skeld's rulings, 2026-10-22): `ROOM_PAIRS`, the eight type-and-supertype pairs a parlor puts in a world (narrative/Message, narrative/Board, relation/Work item, construct/Spec, relation/Heart, relation/Vote, event/Sitting, character/Member), and `isRoomElement(el)`, true when an element's type and supertype, each trimmed and compared without regard to case, match one of them **and** it carries the mark `x_room_of` (the id of the world or sitting the room belongs to) as a non-empty string. A pair without the mark is lore (a novel's letters are Narratives with supertype Message); a mark without a pair is not a room element. The pair match is the whole word, a missing or empty supertype is never a room element, and a Law with supertype `Charter` stays lore. A lore view leaves room elements out unless the person asks; the room's own views show them whole. A convention on the 22 types: no type, field or wire change. `ROOM_PAIRS` is generated from the fixture's `pairs` and checked at codegen against the page's table (they must agree, or codegen exits 1); the mark's name is generated from the fixture's `mark.field` and checked against the page. Case is folded by Unicode simple case folding (`simpleCaseFold`), the same fold as the Python package's `is_room_element`. A new test replays the fixture's 23 cases from the vendored dist.
1672
-
1673
- ### Changed
1674
-
1675
- - **Schema repinned to schema-dist v0.33.0-dist.45** (same canonical 00.33.00). The dist adds `profiles/room.md` and `profiles/room.fixture.json`; `VERSION`, `README.md` and `MANIFEST.json` carry the new serial; no schema file moved. Generated output moves its provenance lines and gains `ROOM_PAIRS` and the mark's name. Serial 44, the pairs without the mark, was superseded within the hour and is skipped.
1676
-
1677
- ## [4.54.0] - 2026-10-22
1678
-
1679
- ### Added
1680
-
1681
- - **`Zone.shape`** (canonical 00.33.00, a minor step of the standard): a string in Zone's Scope section beside `role`, saying how the zone's markers are joined. `"polygon"` is a closed area (the last marker joins the first), `"line"` an open path drawn in marker order; lowercase. Empty, absent or any other value means not stated, and a reader draws a polygon. Generated as `shape?: string` on `ZoneV2`, `FIELD_SCHEMA.zone.shape` (`{ type: 'text' }`), and in Zone's `Scope` entry of `ELEMENT_SECTIONS`. `ONLYWORLDS_VERSION` is `'00.33.00'`.
1682
-
1683
- ### Changed
1684
-
1685
- - **Schema repinned to schema-dist v0.33.0-dist.43** (canonical 00.33.00). Only `schema/zone.yaml` and `schema.json` moved, by the one field; `VERSION`, `README.md`, `example/world-export.json` and `MANIFEST.json` carry the new version; nothing else.
1686
-
1687
- ## [4.53.3] - 2026-10-22
1688
-
1689
- ### Changed
1690
-
1691
- - **Schema repinned to schema-dist v0.32.2-dist.42** (same canonical 00.32.02; fixtures/owworld gains `fail-OW-6-marker-bom`, a `mimetype` marker that starts with a byte order mark, 81 cases; nothing else; no schema file moved). It is the case 4.53.2's fix refuses: the reader rejects it as `marker`, and the set runs 81 of 81. Generated output moves its provenance lines only.
1692
-
1693
- ## [4.53.2] - 2026-10-22
1694
-
1695
- ### Fixed
1696
-
1697
- - **A `mimetype` marker with a byte order mark is refused** (`marker`, spec §4.7 OW-6). The reader compared the entry as decoded text, and `TextDecoder`'s default strips a leading BOM, so a marker of BOM plus the media type passed; the spec says bytes exactly. Found by the Python package's port of the reader (`onlyworlds` 0.17.0), which refused it. No fixture carries the case yet (proposed for the conformance set).
1698
-
1699
- ## [4.53.1] - 2026-10-22
1700
-
1701
- ### Changed
1702
-
1703
- - **Schema repinned to schema-dist v0.32.2-dist.41** (same canonical 00.32.02; the dist now carries the `.owworld` conformance set at `fixtures/owworld/`, 80 cases, with `verdicts.json` and `CaseFolding-15.1.0.txt`, and its folder lint names a `world.json` with an `api` block that git tracks; no schema file moved). The set's two newest cases are `fail-OW-14-crc`, an entry whose bytes fail the CRC its headers declare, and `fail-OW-14-stored-short`, a stored entry holding fewer bytes than it declares. The reader already refused both as `not_openable`, naming the entry, and needed no change. The world-file conformance test reads the set from the vendored dist, so it runs in CI too, where it used to skip; it also checks that every zip on disk has a verdict. The fold table's test and generator read the dist's `CaseFolding-15.1.0.txt`, the same file the SDK held, and the table is unchanged. Folder spec §4.7 is RULED (v0.3.13). Nothing in the published package changes but its version and these documents: `codegen/` does not ship.
1704
-
1705
- ## [4.53.0] - 2026-10-22
1706
-
1707
- ### Changed
1708
-
1709
- - **The last seven options objects refuse an unknown key** too: `patch` and `editLinks` (`ifMatch`; the edit itself is `{add, remove}`), `uploadImage`, `createFileTicket`, `uploadFile`, `createWorldLink` and `createSeat`. With 4.51.0 and 4.52.0 every method of the client that takes an object now names a stray key before anything is sent; a client that drops a key it does not know has no place left.
1710
-
1711
- ## [4.52.0] - 2026-10-22
1712
-
1713
- ### Changed
1714
-
1715
- - **Every options object refuses an unknown key** the way `list()` did in 4.51.0: `draught`, `get`, `create`, `bulk`, `changes`, `changesAll`, `removeImage`, `redeemJoin`, `signInWithPassword`, `mintWorldKey`, `signInWithOnlyWorlds`, `completeSignIn`, `ask` and `sort` throw a TypeError naming the key and the known keys before anything is sent. A dropped key was never harmless: `atomc: true` ran a batch non-atomically, `idempotency_key` let a retry create twice, `afetr` read the whole feed, `mode` misspelt widened a draught to the caller's view. TypeScript callers were already refused at compile time; this is for plain JavaScript.
1716
-
1717
- ## [4.51.0] - 2026-10-22
1718
-
1719
- ### Changed
1720
-
1721
- - **`list()` and `listAll()` refuse an unknown top-level key** with a TypeError before anything is sent, naming the key and where a wire filter goes (`filter: { name__icontains: ... }`). Until now an unknown key was dropped without a word, so a plain-JavaScript caller who wrote `list('character', { name__icontains: 'probe' })` got every Character, unfiltered, and TypeScript's compile-time check was the only fence. A cleanup script written that way deleted 99 Characters on a test world on 2026-10-22; this is the SDK's half of the lesson. A caller who passed a stray key by accident now hears about it; the known keys are `limit`, `cursor`, `ids`, `createdBy`, `expand`, `fields`, `filter`, `viewAs`.
1722
-
1723
- ## [4.50.0] - 2026-10-22
1724
-
1725
- ### Added
1726
-
1727
- - `setMyCharacter(characterId | null)`: `PUT /me/character`, a person member's pick of the roster Character that is them in this world; answers the `/me` body. `err.isCharacterHidden` for the 422 `character_hidden` it and the member PATCH answer.
1728
- - `hidden?: true` on every element body (`OwElementBase`): present only when the owner or a co-builder hid the element and this reader still sees it; never on other readers' bodies, ignored in a write.
1729
- - `site` joins `OwWriteDoor`: a write or delete the owner made signed in on onlyworlds.com.
1730
-
1731
- ## [4.49.0] - 2026-10-22
1732
-
1733
- ### Changed
1734
-
1735
- - **The world file follows Folder Format §4.7, the `.owworld` archive**, and gives the spec's verdict and cause on every one of the 74 fixtures in its conformance set (Quillon's, `owworld-fixtures`), plus its three on-demand floor cases.
1736
- - The reader parses the central directory itself (zip64 included) and runs every check on names, kinds, methods and sizes before it inflates anything.
1737
- - Six new refusal causes join `OwWorldFileReason`: `duplicate` (OW-17), `collision` (OW-18: NFC, then simple case folding), `link` (OW-20), `method` (OW-21), `name` (OW-22) and `marker` (OW-6). A `Record<OwWorldFileReason, …>` in a caller needs the six new keys.
1738
- - The OW-18 fold is the C and S rows of Unicode's CaseFolding.txt 15.1, built in as a table, never the runtime's casing, so a fold a newer Unicode adds stays out until the pin moves (Skeld's ruling, 2026-10-22). `simpleCaseFold(text)` exports it.
1739
- - Files beside the world's folder are ignored, no longer `more_than_one` (OW-4). A `world.json` under `records/` is a producer's data, not a second world (OW-5). A spanned archive is `not_openable`. An entry that inflates past the size it declares is `too_big` at any budget (OW-14).
1740
- - The default budget is the spec's floor: `WORLD_FILE_MAX_BYTES` 2 GiB (was 1 GiB), `WORLD_FILE_MAX_ENTRIES` 500,000 (was 200,000) (OW-15).
1741
- - A `mimetype` marker is never required. One with other bytes than `WORLD_FILE_MEDIA_TYPE` (`application/vnd.onlyworlds.world+zip`, newly exported) is refused, and the marker is never returned as a file.
1742
- - `files` come back in code point order (the UTF-8 bytes of the name), not UTF-16 order.
1743
- - The writer gives one set of bytes for one input (OW-12, OW-13). It places the `mimetype` marker first and stored, orders entries by the UTF-8 bytes of their names, and stamps every entry 1980-01-01 00:00 (was the clock).
1744
- - The writer writes only `world.json`, `elements/`, `spatial/`, `media/`, `records/` and a tool's dot-folder it is handed (OW-7). It leaves out any other file, and a dot-name or `.meta` inside a content folder. It refuses a `name` or a `collision` as the reader would.
1745
-
1746
- ## [4.48.1] - 2026-10-22
1747
-
1748
- ### Changed
1749
-
1750
- - Schema repinned to schema-dist `v0.32.2-dist.37` (canonical 00.32.02 unchanged, so no type, field or constant moved): the dist added `profiles/` (identity, calendar with its fixture, packet, review), its folder lint reads known-of stubs and never `records/`, and its export example carries `source` and `exported_as`. The generated headers, `SCHEMA.md` and `AGENTS.md` name the new pin.
1751
-
1752
- ## [4.48.0] - 2026-10-22
1753
-
1754
- ### Added
1755
-
1756
- - `listHidden()`: `GET /world/hidden`, the world's hidden elements newest first as `OwHiddenRow[]` (`type`, `id`, `set_at`, `set_by`). Any key of the owner's or a co-builder's; served with `Cache-Control: no-store`, so read it each time. For the parlor, whose host room marks what it hid (Temper, 2026-10-22).
1757
-
1758
- ## [4.47.0] - 2026-10-22
1759
-
1760
- ### Added
1761
- - `draught(characterId, policy)`: `GET /character/{id}/context` (live 2026-10-22 02:05), one Character's context in one read, typed as `OwDraught` with the policy's seven keys (`components`, `mode`, `asOf`, `undated`, `since`, `budget`, `depth`) as `OwDraughtPolicy`.
1762
- - `list` takes `ids` (1 to 200, the list's own order; one the key cannot see is absent) and `createdBy` (a membership id, or `'owner'`), the two list filters of the same PR; both were already reachable through `filter`.
1763
-
1764
- ## [4.46.0] - 2026-10-22
1765
-
1766
- ### Added
1767
- - **Sign in with OnlyWorlds** (live 2026-10-22 01:50): `signInWithOnlyWorlds({ clientId, redirectUri, resume, storage, navigate })` mints `state` and a PKCE S256 verifier, keeps them under the client id in `storage` (sessionStorage by default; required in Node), refuses a `redirectUri` that is not an `https:` string before sending, and sends the browser to `/signin/authorize`. `completeSignIn({ clientId, storage, url })` on the callback page checks `state` and the page, takes `code` and `state` out of the address bar, trades the code once at `POST /api/v2/auth/token` with no credential, and returns `OwSignInResult` (`token`, `account`, `resume`); the record is cleared either way and the SDK keeps no token. `OwSignInError` gains `invalid_grant` and `state_mismatch`. `pkceChallenge` and `base64url` exported.
1768
-
1769
- ## [4.45.0] - 2026-10-22
1770
-
1771
- ### Added
1772
- - `hide(type, id)` and `unhide(type, id)`: the hidden mark's writes, `PUT` and `DELETE /{type}/{id}/hidden`, for the parlor (Temper, 2026-10-22). 204; the element's body and `change_seq` do not move.
1773
- - `listAccountWorlds()` (`GET /account/worlds`, `OwAccountWorld[]`: owned and shared worlds with their keys by name and last four) and `mintWorldKey(worldId, { scope, name })` (`POST /account/worlds/{id}/keys`, `OwMintedKey` with the full key once), the two account routes Atlas and the parlor hand-rolled. Account token only.
1774
-
1775
- ## [4.44.0] - 2026-10-22
1776
-
1777
- ### Added
1778
- - `changes({ after })` and `changesAll(since, { after })` (live 2026-10-22 00:55): `GET /changes?after=<ISO time>` returns only the changes made at or after that time, a filter and not a position: order, `cursor`, `has_more` and `head` are unchanged, and the window is read by passing the cursor back as `since` with the same `after`. A bad value is a 422 naming `after`.
1779
-
1780
- ## [4.43.0] - 2026-10-22
1781
-
1782
- ### Added
1783
- - `adopt(type, id, member)`: change who made an element, `POST /{type}/{id}/adopt` with `{member}` (live 2026-10-22 00:43). `member` is a membership id of the world or `'owner'` (`created_by` null). The owner's key or a co-builder's with its PIN; a contributor's is 403 `not_author` (`isNotAuthor`). Returns the updated element; the write takes a new `change_seq` and the feed carries the element with its new `created_by`.
1784
- - `createWorldLink`'s `maxUses` is documented for person links too (live the same hour): 2 to 50 lets in that many accounts, each joining once with the link's role. The client sent it already; the comment said player links only.
1785
-
1786
- ## [4.42.4] - 2026-10-21
1787
-
1788
- ### Fixed
1789
- - **`parseDate` refuses a date the typed era does not hold**, rather than reading its number as an absolute year in another era. With a down era Before Gate turning on 6 Frost, "10 Thaw, 1 BG" read as step 39 ("10 Thaw, 2 AG") and "3 Frost, 0 BG" as "3 Frost, 1 BG"; both now answer `not a date in Before Gate`. The absolute reading stays for a step in no era, where `formatStep` prints the absolute year. No date `formatStep` prints changes how it parses: 20,826 round trips over the profile fixture's calendars read the same before and after. Found by Temper in Atlas 1.67.1 (2026-10-21).
1790
-
1791
- ## [4.42.3] - 2026-10-21
1792
-
1793
- ### Fixed
1794
- - **`parseDate` reads a down era's dates back to the step they were written from, with a mid-year pivot too.** 4.42.2 moved the display count to the year holding the step before the pivot, and the parse kept the old count, so with a mid-year pivot every date before it parsed back one year early ("6 Frost, 4 BG" read 20 steps early on a 20-step year). Year-boundary pivots were never affected. Found by Quillon porting 4.42.2 to the Unity SDK and confirmed on this kernel by Temper (2026-10-20); a round-trip test now holds every step on both sides of the pivot.
1795
- - `redeemJoin`'s doc names a full world's refusal as 422 `world_full` (`isWorldFull`), as keel answers since the party link; it said `invalid_request` (Quillon, 2026-10-20).
1796
-
1797
- ## [4.42.2] - 2026-10-20
1798
-
1799
- Two calendar reading fixes, found writing the `atlas_calendar` profile's conformance fixture (2026-10-20).
1800
-
1801
- ### Fixed
1802
- - **A down era has no year zero, even when its pivot falls mid-year.** Its year is now counted from the year holding the step just before the pivot, which reads `year_base` (1 by default). With the pivot mid-year, the earlier steps of the pivot's own year read year 0 ("0 BG") and every year before it one lower; with the pivot on a year's first step, nothing changes. The mirror of an up era, whose first partial year is year 1.
1803
- - **Equal-start eras resolve by a stated rule:** an up era over a down one, then the later era in the list. The comparator returned `-1` for any pair led by an up era, so two eras of one direction came out in the sort's order rather than the list's.
1804
-
1805
- ## [4.42.1] - 2026-10-20
1806
-
1807
- The calendar reader refuses what the `atlas_calendar` profile refuses (schema-dist `profiles/calendar.md`; Quillon's Unity read, 2026-10-20).
1808
-
1809
- ### Fixed
1810
- - **`readCalendarValue` / `readCalendarConfig` no longer throw on a falsy `eras` or `intercalary`** (`false`, `0`, `""`): it threw a `TypeError` where the profile says a reader refuses and shows bare numbers. It now returns `null`.
1811
- - **A step is an integer, so the reader now refuses** an era whose `start` or `end` is not a whole number, a `year_base` other than `0` or `1` (a string `"1"` printed as `"01"` before), and month or weekday names that are not strings. Atlas writes none of these; the Unity SDK refuses them the same way.
1812
-
1813
- ## [4.42.0] - 2026-10-20
1814
-
1815
- Schema 00.32.02 (Skeld, 2026-10-20): repinned to schema-dist v0.32.2-dist.29.
1816
-
1817
- ### Changed
1818
- - **A Narrative's `start_date`** reads "Date when the narrative is first told or recorded, measured in world TIME units; the span it recounts lives on its events" (it read "Date when the narrative begins"), and **its `end_date`** reads "Date when the narrative stops being told, measured in world TIME units; empty while it is still told" (it read "Date when the narrative ends"), in the generated doc comments and `SCHEMA.md`. No field, type or link changed.
1819
- - **`ONLYWORLDS_VERSION`** is `'00.32.02'`. It is an `as const` export, so its literal type moves too, which is why this is a minor and not a patch.
1820
-
1821
- ## [4.41.0] - 2026-10-20
1822
-
1823
- A known-of stub's public fields, live in production, and schema-dist v0.32.1-dist.28, which publishes them.
1824
-
1825
- ### Added
1826
- - **`PUBLIC_FIELDS`** (generated from the distribution's `presentation.json`, `types.<type>.public`): per element type,
1827
- the fields onlyworlds.com shows on a known-of stub besides the base fields, with their values, as a readonly
1828
- `as const` array; all 22 types present. A Character's are `physicality`, `height`, `weight`, `species` and
1829
- `reputation`; every other type's list is empty for now. Platform behaviour, not a default to override. Codegen
1830
- exits 1 on a tree without the sidecar, a type without the key, or a name that is not a field of its type.
1831
- - **`err.isKnownOfNotWritable`**: keel's named refusal of a write carrying `known_of`, true or false (422
1832
- `known_of_not_writable`, `param` known_of). This client never sends `known_of` (4.40.0's strip and refusal are
1833
- unchanged), so it reaches you only through `request()`; `/bulk` puts it in the item's slot.
1834
-
1835
- ### Changed
1836
- - Repinned to schema-dist v0.32.1-dist.28: `presentation.json` gains `types.<type>.public` and `_meta.public`; no
1837
- schema file moved, so the generated interfaces change only their source lines.
1838
- - **A stub's description** on `known_of` (the generated doc comment, `SCHEMA.md` and `AGENTS.md`) says a stub
1839
- keeps its type's `PUBLIC_FIELDS` with their values too; it said every other field reads unset.
1840
-
1841
- ## [4.40.0] - 2026-10-20
1842
-
1843
- The "known of" stub's client half, live in production since 2026-10-20 13:47.
1844
-
1845
- ### Added
1846
- - **`known_of?: true`** on `OwElementBase` (generated; `SCHEMA.md` says it too): present, and true, only on a
1847
- stub, an element a guest or player reaches only through its own Character's sheet, one step out. A stub
1848
- keeps `id`, `type`, `name`, `description`, `supertype`, `subtype`, `image_url`, `created_at`, `updated_at`
1849
- and `change_seq`; every other field reads unset (`""`, `null`, `[]`), `created_by`, `changed_by` and
1850
- `changed_via` are null, it has no extension fields, and its files are 404. For Lantern and Atlas, which
1851
- render stubs (Temper).
1852
- - **`isKnownOf(el)`**: true only for `known_of: true`; absent or anything else is a whole element.
1853
-
1854
- ### Changed
1855
- - **A stub is never written**: `upsert`, `patch`, `create` and `bulk` throw a `TypeError` before sending a
1856
- body with `known_of: true` (`bulk` names the item, and sends none of the batch). Stripping the marker would
1857
- turn keel's 422 into a write that blanks the element for any key that may write it (an owner pushing a
1858
- player's copy). `known_of` joins the strip list, so any other value of it is never sent.
1859
-
1860
- ## [4.39.0] - 2026-10-20
1861
-
1862
- The party link, live in production since 2026-10-20 13:15.
1863
-
1864
- ### Added
1865
- - The party link (for Lantern's pick screen, Temper): one player link for a
1866
- table of players. `createWorldLink(worldId, { reader: 'player', maxUses, characters })`, sent as `max_uses`
1867
- and `characters`: 2 to 50 uses, each player picking one Character from the owner's list, in its order; no
1868
- list, each player names a fresh Character after themselves. A table takes no `character`. A link for one
1869
- sends what it sent before. `OwWorldLink` and `OwCreatedWorldLink` gain a table's `max_uses` and
1870
- `characters` (`{ id, name, taken }`); `uses` counts a table's seats. `OwJoinPreview` gains a table's
1871
- `max_uses`, `uses_left`, `name_required` and `characters` (`{ id, name, image_url, taken }`).
1872
- `redeemJoin(code, { character })` sends the pick; with `name_required`, `name` alone. A pick a live seat
1873
- already plays is 422 `character_taken` (`err.isCharacterTaken`); a pick off the list, hidden or deleted
1874
- since the link was made, or missing, is 422 `invalid_request` with param `character`
1875
- (`err.isCharacterInvalid`, true for that param only); a world with its 50 players is 422 `world_full` (`err.isWorldFull`, its own code at my ask, Skeld, 2026-10-20); the last place
1876
- lost to another player is the one 404 (`err.isNotFound`).
1877
-
1878
- ### Changed
1879
- - Repinned to schema-dist v0.32.1-dist.27: only `example/world-export.json` moved (its `schema_version` now 00.32.01, Skeld, 2026-10-20); no schema file moved, so the generated files change only their source lines.
1880
-
1881
- ## [4.38.0] - 2026-10-20
1882
-
1883
- Schema 00.32.01 (Titus's word, 2026-10-20): repinned to schema-dist v0.32.1-dist.26.
1884
-
1885
- ### Changed
1886
- - **An Event's `characters`** reads "Characters who know about the event: they were present, aware of it, or involved in it" (it read "Key characters relevant to the event"), in the generated doc comment and `SCHEMA.md`. No field, type or link changed.
1887
- - **`ONLYWORLDS_VERSION`** is `'00.32.01'`. It is an `as const` export, so its literal type moves too, which is why this is a minor and not a patch.
1888
-
1889
- ## [4.37.0] - 2026-10-20
1890
-
1891
- The hidden mark, live in production since 2026-10-20 09:24; walked on staging with the mark on, 25 of 25.
1892
-
1893
- ### Added
1894
- - A `/changes` delete row's `reason` (`OwDeleteReason`: `deleted` | `hidden` | `out_of_view`; keel's hidden
1895
- mark): an element hidden from this reader leaves its copy as a deleted one does, and a GET of it answers 404,
1896
- not 410. `isResyncRequired`'s comment names every case keel now sends it for, not only a guest's.
1897
- - `viewAs` on `get`, `list` and `listAll`, sent as `as` (view as): the owner's key reads as a
1898
- member (a membership id), as a Character's guest (a Character id) or as the public (`'public'`). A member key
1899
- gets 403 `view_as_owner_only` (`err.isViewAsOwnerOnly`); an element hidden from that reader is 404, never 410;
1900
- an unknown value is 422. Until keel's hidden mark is on, keel answers 422 (an unknown parameter).
1901
- - `err.resyncReason` (`'role'` | `'view'` | `undefined`) and `OwResyncReason`: why `/changes` answered 409
1902
- `resync_required`, from the envelope's `reason` (only while keel's hidden mark is on).
1903
- `epoch_reason` beside `epoch` on `OwChangesPage` and on `changesHead()`'s answer: why the epoch last moved.
1904
-
1905
- ### Changed
1906
- - The resync instruction adds the hidden mark's reasons to 4.36.1's keep (Titus, 2026-10-20): after a
1907
- 409 `resync_required`, walk again from the start; reason `role` (the key's own role or membership moved):
1908
- keep what the local copy holds that the walk no longer names; reason `view`, or a delete row with reason
1909
- `hidden`: drop it; no reason (keel's hidden mark off): keep it, a reset never empties a copy. A tool never
1910
- empties a copy on its own. `isResyncRequired`'s comment and the `changes`,
1911
- `changesHead` and `OwChangesPage` comments say so.
1912
-
1913
- ## [4.36.2] - 2026-10-20
1914
-
1915
- ### Changed
1916
-
1917
- - `pinExpectation`'s doc comment: a legacy key that can write sends its PIN on every read of a world that has one, whether or not the world is open to read (the demo split). The `'required-for-private-reads'` literal keeps its name until 5.0, since renaming a literal in a published union is a type break. A demo-range key is kind `legacy` too and never needs a PIN. No code change.
1918
-
1919
- ## [4.36.1] - 2026-10-20
1920
-
1921
- ### Changed
1922
-
1923
- - The reset rule in `isResyncRequired`'s and `changes`' doc comments: on a 409 `resync_required`, walk from zero and keep what the local copy holds that the new walk does not name; a reset never empties a copy. It said replace. Keel says keep (Titus's ruling, 2026-10-20: sharing is forward-only). No code change.
1924
-
1925
- ## [4.36.0] - 2026-10-20
1926
-
1927
- ### Changed
1928
-
1929
- - **`fieldLabel`** keeps a word written all in capitals as written (`STR` reads "STR", `plan_B` "Plan B"), so a custom stat block reads the same in every tool (Temper, 2026-10-20; Lantern and Atlas both call it). Schema field names are all lowercase, so no schema label changes.
1930
-
1931
- ## [4.35.0] - 2026-10-20
1932
-
1933
- ### Added
1934
-
1935
- - **`createWorldLink(worldId, { preview })`** and `preview` on `OwWorldLink` (live 2026-10-20): a person or player link with `preview: true` is `/join/<world_id>#<code>`, so a chat that unfurls it shows the world's name and picture; the code stays in the fragment. Absent means off, as keel rules for the API; `true` on an agent link is 422, param `preview`. The wire gate failed on production's document until this release.
1936
-
1937
- ## [4.34.0] - 2026-10-20
1938
-
1939
- ### Added
1940
-
1941
- - **`fieldLabel(name)`** and **`FIELD_ACRONYMS`**: a field name as the words a tool shows, by the house rule (Titus, 2026-10-20): sentence case, acronyms kept wherever they fall (`birth_date` → "Birth date", `image_url` → "Image URL"). Seeded from Atlas's `humanizeFieldName` as of 1.59.1, so Atlas, Lantern and Council read one copy. The name is taken as given: a tool that hides an extension prefix drops it first.
1942
-
1943
- ## [4.33.1] - 2026-10-19
1944
-
1945
- ### Fixed
1946
- - An `agent: true` client's file uploads now declare it: `createFileTicket` (and so `uploadFile`) sends
1947
- `X-OW-Agent`, the one files route keel's document lists it on, so the file reads
1948
- `added_by.agent: true`. The other files routes and the upload host get no such header.
1949
-
1950
- ## [4.33.0] - 2026-10-19
1951
-
1952
- ### Added
1953
- - `FILE_EXTENSIONS` (every extension a file on an element may end in, to its kind: keel's list, `.yml` and
1954
- `.markdown` included) and `fileKindOf(name)` (the kind, or `undefined`), so a file picker's accept list reads
1955
- the SDK's instead of a copy (Atlas, 2026-10-19).
1956
-
1957
- ## [4.32.0] - 2026-10-19
1958
-
1959
- ### Added
1960
- - **Files on elements**, for Atlas's file list on an element:
1961
- - `listFiles(type, id)`: the element's files, newest first (`OwFileRow`: `id`, `name`, `kind`,
1962
- `size`, `label`, `added_by` with `name`, `membership` and `agent`, `added_at`), as keel sends them.
1963
- - `uploadFile(type, id, file, { name?, kind?, label?, ticket?, onProgress? })`: a ticket from the API,
1964
- then the bytes to the upload host, as `uploadImage` does. Takes a File, Blob, ArrayBuffer or
1965
- Uint8Array; the name is a File's own unless given; the kind comes from the name's extension
1966
- (`.yml` is yaml, `.markdown` is md; anything else not taken is a TypeError before any request).
1967
- `onProgress(loaded, total)` reports the bytes sent through XMLHttpRequest where there is one (a
1968
- browser); elsewhere it is called once, at the end. Returns the host's 201 (`OwUploadedFile`).
1969
- - `createFileTicket(type, id, { name, kind?, size, label? })`: the first step alone. No
1970
- Idempotency-Key, by keel's rule.
1971
- - `fileDownloadLink(type, id, fileId)`: the link (`url`) and when it stops working (`exp`, 5 minutes
1972
- on); the link is the credential, so fetch it when the file is opened and don't keep it.
1973
- - `removeFile(type, id, fileId)`: the world's owner or whoever added the file.
1974
- - `worldFiles()`: every file the key may see, with `used_bytes` and `cap_bytes` for "41 MB of 250 MB".
1975
- - `err.isFileTooLarge` (413 `file_too_large`, over 25 MB) and `err.isFilesFull` (403 `files_full`, the
1976
- world's 250 MB has no room). The world's 200 file tickets a day use up as `err.isQuotaExceeded` (429,
1977
- with `retryAfter`), as image tickets do. The upload host's refusals (409 `exists`, 415
1978
- `unsupported_type`, 401 `ticket_*`) arrive as `OwApiError` with their codes, as the image edge's do.
1979
- - Types: `OwFileKind`, `OwFileAddedBy`, `OwFileRow`, `OwWorldFileRow`, `OwWorldFiles`, `OwFileTicket`,
1980
- `OwUploadedFile`, `OwFileDownload`.
1981
- - An `agent: true` client sends no `X-OW-Agent` on the files routes: keel's document declares it
1982
- on the element writes only.
1983
-
1984
- ## [4.31.0] — 2026-10-18
1985
-
1986
- ### Added
1987
- - **Signing in with a password**: `signInWithPassword({ email | username, password, tokenName })`
1988
- turns an email address (or username) and password into a new account token (`ow_a_`) for this device,
1989
- named `tokenName`, the name the account's tokens page lists it under. It signs in at keel's sign-in service
1990
- (`{baseUrl}/auth/app/v1`), mints the token with that session at `POST /account/tokens`, then ends the
1991
- session, whether the mint worked or not. Only the token comes back (`OwSignedIn`: `token`, `id`, `name`,
1992
- `last4`); the password is sent once and held nowhere. Works on a client with no key, and sends no key
1993
- whatever the client holds. Exactly one of `email` and `username`, else a TypeError before sending.
1994
- - `OwSignInError`, with a `reason`: `invalid_credentials`, `email_unverified`, `rate_limited` (with
1995
- `retryAfter` when keel sends it), `offline`, `mint_failed` or `failed`, plus keel's `status` and `code`.
1996
- The password is never in its message or properties. For Atlas's sign-in (atlas #387) and Lantern, which
1997
- carried its own copy. The session's end is cut off after five seconds, so a hung network never holds a
1998
- sign-in whose token is already made. Checked live on production (a sign-in, the token reading the
1999
- account's worlds, revoked and read back as `key_revoked`).
2000
-
2001
- ### Changed
2002
- - `me()`'s doc names the player kind.
2003
-
2004
- ## [4.30.1] — 2026-10-16
2005
-
2006
- ### Fixed
2007
- - **`changes()` and `changesAll()` hand back every cursor as a string**, as `OwChangesPage.cursor` has
2008
- always said. keel sends the cursor of a page at the head of the feed (an empty page, for any key but
2009
- a guest's) as a number, and it passed through as one: a walk of an up-to-date world returned
2010
- `{ cursor: 818 }`, so a cursor stored and compared as a string could miss. `changesHead()` already
2011
- did this. Found on production with a demo key.
2012
-
2013
- ## [4.30.0] — 2026-10-16
2014
-
2015
- ### Added
2016
- - **A friend with only a link**:
2017
- - `redeemJoin(code, { name?, agentName? })`: redeems a join code that needs no account, at `POST /join`,
2018
- sending no credential whatever key the client holds. The answer's `kind` tells the two apart:
2019
- `OwPlayerJoined` (a read key, `ow_r_`, no PIN, and the Character the player reads as) or
2020
- `OwAgentJoined` (a seat's write key, secret and `.env` lines). Shown once; the code is spent, so a
2021
- lost answer means a new link. Every bad code is one 404.
2022
- - `createWorldLink(worldId, { reader: 'player', character? })`: a player link, guest only; `character`
2023
- is a Character's id or `new` (the default). `OwWorldLink.character` names it on the list.
2024
- - `player` joins the `kind` of `OwMe` and `OwMember`.
2025
- - **A client with no key**: `apiKey` is optional. `new OwV2Client()` (or an empty key) has `keyKind: 'none'`
2026
- and can call `health`, `previewJoin` and `redeemJoin`, which is what a friend with only a link has; any
2027
- other call is a TypeError before anything is sent. Until now a key was required even for the routes
2028
- that send none.
2029
-
2030
- ## [4.29.0] — 2026-10-16
2031
-
2032
- ### Added
2033
- - `rekeySeat(worldId, seatId)`: a new key and secret for a seat whose AI lost
2034
- its own, by its sponsor or the world's owner. The old key and secret stop at once; the answer has
2035
- `createSeat`'s shape (`OwSeatCreated`: `key`, `pin`, `env`), shown once. No Idempotency-Key, by keel's
2036
- rule: if the answer is lost too, ask for another.
2037
-
2038
- ### Changed
2039
- - `createSeat`'s doc names `rekeySeat` as the way back from a lost answer.
2040
- - `OwApiError.param`'s doc: `API-Pin` on a 401 PIN refusal, not only a 422/400's field.
2041
-
2042
- ## [4.28.0] — 2026-10-16
2043
-
2044
- ### Added
2045
- - **Seats for your own AI**, on an account token (`ow_a_`):
2046
- - `createSeat(worldId, { name, role?, character? })`: a seat for your own AI in a world you own or
2047
- belong to, with its own write key and `ow_s_` secret and no join link. You answer for it. `role` is at
2048
- most your own; `character` is `new` (default), `none`, or `mine` (it plays the Character you play).
2049
- The answer (`OwSeatCreated`) carries `key`, `pin` and the `.env` lines in `env`, shown once.
2050
- - `listSeats(worldId)`: the seats you can see (yours, or every seat for the world's owner), as `OwSeat`
2051
- rows with their `sponsor`, no keys.
2052
- - `removeSeat(worldId, seatId)`: its sponsor or the world's owner removes it; its key and secret stop.
2053
- - No Idempotency-Key on `createSeat`, by keel's rule (a replay would keep the key). A lost answer:
2054
- `listSeats` shows the seat, `removeSeat` it and make another.
2055
-
2056
- ## [4.27.2] — 2026-10-16
2057
-
2058
- ### Fixed
2059
- - **`bulk` mints an id for each element that has none, as `create` does.** It sent id-less elements
2060
- bare and keel minted them. With an `idempotencyKey`, each id is now derived from the key and the
2061
- element's position in `items` (the ids the Python package mints), so the same key and items send the
2062
- same body from either client, and a resend after keel's 24-hour replay window rewrites the same
2063
- elements instead of creating new ones. Without a key, a fresh UUIDv7 each. Ids you pass are kept, and
2064
- your items are not changed. Found by Quillon's parity audit of the Unity SDK (2026-10-16).
2065
-
2066
- ## [4.27.1] — 2026-10-16
2067
-
2068
- ### Fixed
2069
- - **A retried `create` with the same `idempotencyKey` now replays instead of failing.** `create` minted a
2070
- fresh id on every call, so a retry of an id-less element sent a different body and keel answered 409
2071
- `idempotency_error`. With a key and no id, the id is now derived from the key (a UUIDv5, the same id the
2072
- Python package mints), so the retry sends the same body and gets the first answer back. Without a key,
2073
- a fresh UUIDv7 as before; an id you pass is kept.
2074
- - `create`'s doc said a keyed retry "rewrites the same element"; keel's replay writes nothing. Found by
2075
- Quillon's parity audit of the Unity SDK (2026-10-16).
2076
-
2077
- ## [4.27.0] — 2026-10-16
2078
-
2079
- ### Added
2080
- - `OwApiError.isIdempotencyInFlight`: a request with the same Idempotency-Key is still
2081
- running (409 `idempotency_in_flight`, `retryAfter` 1). Nothing was written by this one; send again with
2082
- the same key and body after `retryAfter` and it replays the first answer. `isIdempotencyConflict` stays
2083
- the different-body case only. The client still does not retry for you.
2084
-
2085
- ## [4.26.0] — 2026-10-15
2086
-
2087
- ### Added
2088
- - **The door of every write**. An opt-in
2089
- `agent: true` on `OwClientConfig` sends `X-OW-Agent` on every element write (create, upsert,
2090
- patch, delete, editLinks, bulk: the operations keel's document declares it on; never on reads,
2091
- account routes or the image edge); keel records such writes as `declared`, agent-made, and stores only that the header was
2092
- there. Off by default: it describes the caller, so a tool people use never sets it.
2093
- - `OwElementBase` gains the read-only `changed_by` (the membership of the last write) and
2094
- `changed_via` (its door: `seat`, `mcp`, `declared`, `api`, `v1`, `sync`, `server`); a `/changes`
2095
- delete op gains `deleted_by` and `deleted_via`. Keel drops them from a write body, like `created_by`.
2096
- - `OwWriteDoor`, `AGENT_DOORS` and `isAgentDoor(door)`: seat, mcp and declared are agent-made.
2097
- - The two kinds of `/changes` item are named types, `OwChangeUpsert` and `OwChangeDelete` (`OwChange` is their union, as before), and both declare `change_seq`, which keel always sends and the type never named.
2098
-
2099
- ## [4.25.0] — 2026-10-15
2100
-
2101
- ### Changed
2102
- - **`OwWorldMeta` types every field keel's `World` declares** (it named four): the description and
2103
- picture, the calendar (`time_format_names`, `time_format_equivalents`, `time_basic_unit`, the three
2104
- `time_range_*`), the three unit labels, `public_read`, `owner_character`, `created_at`, and new
2105
- **`agent_narrator_rule`** (whether a Narrative written through an agent seat is
2106
- narrated by that seat's own Character; read-only, set on the world page). Past `id` and `name`
2107
- they stay optional so code that builds a world meta keeps compiling; keel always sends all of them.
2108
- `updated_at` and `created_at` read `string | null`, as the document says. The wire gate now
2109
- holds this type to `World` (it had held only the PATCH body). Council and Lantern type-check on it.
2110
-
2111
- ## [4.24.0] — 2026-10-15
2112
-
2113
- ### Added
2114
- - **A world's invite links** (keel C1, 48c10d4), on the owner's account token (`ow_a_`):
2115
- `createWorldLink(worldId, { reader, role?, days?, inviterName?, withCharacter? })` answers the
2116
- new link with its `code`, `link` and the one `line` to send (the code is in this answer only:
2117
- keel keeps a hash); `listWorldLinks(worldId)` the links without codes, each with its `status`
2118
- and `uses`; `revokeWorldLink(worldId, linkId)`. Types `OwWorldLink`, `OwCreatedWorldLink`. A
2119
- non-owner, and a world that doesn't exist, get 404; a world key is a TypeError before sending.
2120
- For Lantern's Invite someone (Temper, 2026-10-15).
2121
-
2122
- ### Changed
2123
- - `acceptJoin`'s doc comment and the README: accepting needs no account PIN for any role,
2124
- so its refusals are 404 and 409 `already_member`; `pin_required` now belongs to
2125
- minting a write key for the membership. `isPinRequired` stays.
2126
- - Doc comments say "token rating" for the account's daily budget (`ask`, `sort`, `allowance`,
2127
- `OwAllowance`, the answers' `token_rating`), as the README did in 4.23.0; the docs site's
2128
- generated reference reads them. No code change.
2129
-
2130
- ## [4.23.0] — 2026-10-14
2131
-
2132
- ### Deprecated
2133
- - **`TokenResource.consume()`, `getAccessKey()`, `revokeSession()`, `revokeAllSessions()`**: keel
2134
- serves none of these routes (each answers 404 `not_found` on production; `accounts/token_api.py`
2135
- mounts only `status/` and `encryption-info/`). Spending is server-side, in the route that does the
2136
- work (the Council's acts through keel's council proxy, Skeld's ruling, 2026-10-14), and
2137
- `OwV2Client.allowance()` reads the token rating. They stay callable and leave in 5.0. A 404 from
2138
- any token route now says what keel serves instead of "an older server". Found by the wire
2139
- gate, which until now never drove this class (its routes are not in keel's document).
2140
-
2141
- ### Changed
2142
- - README and docs: "token rating" is the one word for the account's daily budget (Titus,
2143
- 2026-10-14). Names in the API (`allowance()`, `/account/allowance`) are unchanged.
2144
-
2145
- ## [4.22.0] — 2026-10-14
2146
-
2147
- ### Added
2148
- - **`client.me()` and `client.members()`**: GET /me (who this world key is: world, scope, kind,
2149
- role, membership id, Character) and GET /members (the roster's rows, unwrapped from `{ data }`).
2150
- Lantern reads a player's role from /me by hand today. Types `OwMe`, `OwMember`, typed from
2151
- keel's declared schemas and held to them by the wire gate. An account token is a
2152
- TypeError before sending; the tests fail with that guard removed, or with the rows left
2153
- wrapped.
2154
-
2155
- ### Changed
2156
- - The calendar kernel is the SDK's own: Atlas 245aad3 deleted its `core/calendar.ts` and imports
2157
- this one (Temper, 2026-10-14), so the copy record (`src/calendar/SOURCES.json`) and the re-copy
2158
- script (`codegen/atlas-calendar.mjs`) are gone. A kernel change is an SDK release now. No code
2159
- change.
2160
-
2161
- ## [4.21.0] — 2026-10-14
2162
-
2163
- ### Added
2164
- - **`client.allowance()`**: GET /account/allowance (keel 872431f, Skeld, 2026-10-14), the
2165
- account's daily AI allowance read without spending: `token_rating`, `tokens_used_today`,
2166
- `tokens_remaining`, `resets_at` (the next midnight UTC). Account token only, like ask and
2167
- sort. Type `OwAllowance`, held by the wire gate to the route's declared schema, as
2168
- `OwAskAnswer`, `OwSortAnswer` and `OwJoinPreview` now are to theirs.
2169
-
2170
- ### Fixed
2171
- - README and AGENTS.md no longer say a sort counts as one ask: since keel a6244db both spend
2172
- tokens from the allowance.
2173
-
2174
- ## [4.20.0] — 2026-10-14
2175
-
2176
- ### Added
2177
- - **World calendars**, Atlas's calendar code in the SDK so Lantern can name dates (Temper,
2178
- 2026-10-14): `formatStep`, `formatYearLabel`, `breakdownStep`, `parseDate` and the rest of
2179
- Atlas's `src/core/calendar.ts` (Atlas 02d6a6d), plus `worldCalendars`,
2180
- `resolveCalendarLens` and `readCalendarValue` from its `calendar-lens.ts`, on any element
2181
- shape. `CALENDAR_FIELD` is `atlas_calendar`: Atlas owns the field, its meaning and its
2182
- writes, and other tools read it (Skeld, 2026-10-14).
2183
- - `src/calendar/calendar.ts` is Atlas's file byte for byte, recorded in
2184
- `src/calendar/SOURCES.json`; `codegen/atlas-calendar.mjs` re-copies it
2185
- and ports Atlas's 48 kernel tests (TypeScript's transpiler, a five-matcher `test/expect.mjs`
2186
- over node:test). Watched failing: a planted `floorMod` edit failed two tests and the record
2187
- (EDITED); a lens edit failed its test. An app that imports none of it bundles none of it.
2188
-
2189
- ## [4.19.0] — 2026-10-14
2190
-
2191
- ### Added
2192
- - **`tokens_remaining` and `token_rating` on `OwAskAnswer` and `OwSortAnswer`**: since keel
2193
- a6244db (Skeld, 2026-10-14) ask and sort spend from the account's daily token rating, one
2194
- meter with the Council. `remaining` stays a count of answers: how many more calls of this
2195
- size fit in what is left today (keel 504437b), never tokens. Typed from keel's code
2196
- (`accounts/ask.py`), since its OpenAPI document gives these answers no schema.
2197
-
2198
- ### Changed
2199
- - **Custom fields (`x_*`) are in the lookup and on the card.** `buildWorldIndex` finds an
2200
- element by its custom text and numbers, and `renderElement` (so `buildAskContext`) shows
2201
- each one by its own name: `x_eye_colour` reads `eye colour: grey`; an id in one is a link,
2202
- by name; a long value is cut like a description. `atlas_*` machinery stays out, and so do
2203
- object values (a tool's state). Lantern shows and edits custom fields now, so Ask can answer
2204
- what find already could (Temper, 2026-10-13). The trial's lookup
2205
- still skips them; parity holds on its ten questions, whose fixture has no custom text.
2206
- - The lookup's stopword set is an array literal, so bundlers mark it pure and drop it from
2207
- apps that never call the lookup (451 bytes; Temper, 2026-10-14).
2208
-
2209
- ## [4.18.1] — 2026-10-12
2210
-
2211
- ### Changed
2212
- - **Schema repinned to schema-dist v0.32.0-dist.23** (same canonical 00.32.00). The one
2213
- generated change is the `id` doc comment: a UUID, new ids minted as uuidv7, earlier uuidv4
2214
- ids still valid. No type changes. The vendored dist gains `walk/lint_folder.py`.
2215
- - README and AGENTS.md describe `sort()`.
2216
-
2217
- ## [4.18.0] — 2026-10-12
2218
-
2219
- ### Added
2220
- - **`client.sort({ worldId, notes, slice })`**: POST /account/sort, ask's sibling (Temper for
2221
- Lantern, 2026-10-12; Skeld's route). Notes as words or
2222
- `{ text, image_url }`; the slice is the app's own lookup. Rows of four kinds come back to
2223
- review (`make`, `add_to`, `link`, `ask` with an optional `then`) and `not_proposed` with
2224
- reasons; nothing is written. One allowance with `ask`, and the same refusals. Types
2225
- `OwSortRef`, `OwSortNote`, `OwSortRow` and its four kinds, `OwSortAnswer`. Built against
2226
- keel PR #4 (90961fa, Skeld's final for the types), released once it was on keel
2227
- main and in production: verify-sdk-wire PASS 1,111 against production's document.
2228
-
2229
- ## [4.17.0] — 2026-10-12
2230
-
2231
- ### Added
2232
- - **`changesAll` returns `epoch`** beside `cursor` and `head`: the last page's, so a full walk
2233
- says which view epoch it ran under (a guest's integer; `null` for other keys or a keel
2234
- without the field) and the first poll after it needs no extra walk to learn it. Live on
2235
- Moppetopia: a full walk of 465 changes returns `{ cursor, head: 818, epoch: null }`. The first
2236
- tests of `changesAll`'s return value (Temper, 2026-10-12).
2237
-
2238
- ## [4.16.0] — 2026-10-12
2239
-
2240
- ### Added
2241
- - **`epoch` on the feed** (keel 979eb0c). `OwChangesPage.epoch` and `changesHead()`'s
2242
- `epoch`: a guest's view epoch (an integer), `null` for every other key. It moves when what a
2243
- guest may see changes, even with `head` unchanged, so a poll walks when either `head` or
2244
- `epoch` moved; nothing needs to read inside the cursor. `changesHead()` answers
2245
- `epoch: null` from a keel without the field. Checked live on Moppetopia (the head read and a
2246
- page both carry `epoch: null`). The wire gate now holds `OwChangesPage` and `OwBulkResponse`
2247
- to keel's `ChangesPage` and `BulkResult`, and failed on 4.15.0 for the missing
2248
- `epoch` before this release.
2249
- Published from tag `v4.16.0`; the workflow's read-back gave up before npm showed the
2250
- version, so the tarball (shasum `d08de22b`) was read back by hand against a build of the tag.
2251
-
2252
- ## [4.15.0] — 2026-10-12
2253
-
2254
- ### Added
2255
- - **`changesHead()`**: `GET /changes?head=true`, the feed's position now without a page
2256
- (keel reads no element). Returns `{ head, cursor }`: `head` is the world's `change_seq`,
2257
- `cursor` a position to pass as `since` that re-sends nothing (a guest's carries its view's
2258
- epoch). The cheap "anything new?" for a poll: compare `head` with the head you last applied
2259
- and walk `changesAll` only when it moved. keel sends the owner's cursor as a number; it comes
2260
- back as a string, like every cursor the SDK hands out. Checked live on Moppetopia: a walk
2261
- from the returned cursor brings nothing. Asked for by Lantern's 20-second poll (Temper,
2262
- Skeld's S15).
2263
-
2264
- ## [4.14.0] — 2026-10-12
2265
-
2266
- ### Fixed
2267
- - **The co-builder role is `co_builder` on the wire.** `OwJoinPreview.role` and
2268
- `OwJoinedMembership.role` listed `'co-builder'`, which is only the label keel shows (keel
2269
- `MemberRole.CO_BUILDER = "co_builder"`). The types were open, so nothing broke at runtime, but
2270
- a comparison written from them would never match. Caught by Temper from Lantern's fakes.
2271
-
2272
- ### Added
2273
- - `OwMemberRole` (`'co_builder' | 'contributor' | 'guest'`, open), the type both now use.
2274
-
2275
- ## [4.13.0] — 2026-10-12
2276
-
2277
- ### Added
2278
- - **Ask your world.** `client.ask({ worldId, question, context })` sends a
2279
- question to keel's AI proxy with the account's `ow_a_` token (a world key is a `TypeError`
2280
- before sending) and returns `OwAskAnswer` (`answer`, `model`, `stop_reason`, `usage`,
2281
- `remaining`). The answer comes from `context` only, or is "The world doesn't say.".
2282
- `buildAskContext(items, index)` turns the elements the user sees (lookup hits or plain
2283
- elements, in the caller's order, never re-looked-up) into that `context` within keel's
2284
- limits (`ASK_LIMITS`: 8 items, 2,000 characters each, 12,000 in all, counted in characters
2285
- as keel counts them) and says what it did: `kept`, `dropped` (past the 8th, or from the first
2286
- one the total cannot hold, so what is sent is always the top of the list) and `truncated`.
2287
- New getters: `isQuotaExceeded` (429; `retryAfter` runs to midnight UTC), `isAskUnavailable`
2288
- (503), `isAskFailed` (502). keel's document gives no request schema for `/account/ask`, so
2289
- the body (`world_id`, `question`, `context`) is read from keel's code (`accounts/ask.py`).
2290
- Not yet run live: no account token was at hand to run it with.
2291
-
2292
- ## [4.12.0] — 2026-10-12
2293
-
2294
- ### Added
2295
- - **Joining a world by link.** `previewJoin(code)` reads what a join code holds
2296
- without using it (`OwJoinPreview`: `reader`, `world`, `invited_by`, `owner_character`,
2297
- `role`, `with_character`, `expires_at`) and sends no credential, whatever key the client
2298
- holds. `acceptJoin(code)` accepts an invite link as the account behind the client's `ow_a_`
2299
- token and returns the new membership (`OwJoinedMembership`, `roster_character` included);
2300
- a world key is a `TypeError` before anything is sent. New getters on `OwApiError`:
2301
- `isNotFound` (every bad code is one 404), `isAlreadyMember` and `isPinRequired` (the two
2302
- 409s, told apart by code). Shapes read from keel's code (`accounts/join_api.py`,
2303
- `accounts/api.py`), since the OpenAPI document gives no response bodies for these routes.
2304
- Checked live: a bad code on production answers `isNotFound`. A real accept has not been
2305
- run from here (no account token); Lantern's door is its first live use.
2306
-
2307
- ## [4.11.0] — 2026-10-12
2308
-
2309
- ### Added
2310
- - **World lookup.** `buildWorldIndex(elements)`, `lookupWorld(index, question, { limit })` and
2311
- `renderElement(element, index)` find the elements a plain-language question is about, in a
2312
- world held in memory: no model, no network, a few milliseconds. Ranking: names and short
2313
- forms first, then name words, supertype and subtype, one link step in both directions, then
2314
- other text and numbers (always below any name hit). Each hit says why it matched.
2315
- `renderElement` gives a compact text card with links by name, which is what a model reads
2316
- when one phrases the answer. Measured on two real worlds (a Lantern Ask trial):
2317
- the answer was in the first three results for 60 of 60 questions that name their subject
2318
- and 17 of 24 held-out questions that only describe it; this code returns the same top 8
2319
- as the trial's on all 100 questions. Names in any script are matched (Chinese and Japanese
2320
- only where spaces or punctuation separate the words: there is no word splitting). The question words
2321
- that steer the link step ("where", "rules", "allies") are English.
2322
- - **Photo preparation.** `preparePhoto(blob, { maxSide, quality, maxBytes })` shrinks and
2323
- converts a picture in the browser before `uploadImage`: longest side 2048 by default, webp at
2324
- quality 0.82, from the upload page's converter. A browser that cannot encode webp returns a
2325
- PNG when asked; the result's real type is checked, and a photo then falls back to JPEG while a
2326
- PNG stays PNG. When re-encoding would only make a file bigger, the original is kept.
2327
- `planPhoto` is the pure part (sizes and type order). Walked in Chromium, WebKit and Firefox
2328
- (a 4032×3024 photo became a 2048-wide webp of about 0.5 MB) and with webp encoding forced off.
2329
- Needs a browser; in Node it throws a `TypeError` that says so. `PHOTO_TYPES`,
2330
- `PHOTO_MAX_BYTES` are keel's accepted types and its 15 MB limit.
2331
-
2332
- ## [4.10.0] — 2026-10-11
2333
-
2334
- ### Added
2335
- - **World files.** `readWorldFile(bytes, { maxBytes, maxEntries, skip })` opens a `.zip` that
2336
- holds one world folder (Folder Format §4.1, a portable copy) and returns
2337
- `{ worldId, name, world, files }`; `writeWorldFile(files, topDir)` makes one. Bytes in,
2338
- bytes out: no DOM or file system, so it runs in the browser and in Node. Absorbed from
2339
- Atlas's reader (Temper's), so Atlas, Lantern and scripts open one file the same way.
2340
- The reader refuses before returning anything, with **`OwWorldFileError`** and a `reason`:
2341
- `unsafe_path` (climbs out, absolute, a drive, a backslash, an empty part, a path that is
2342
- both a file and a folder), `too_big` / `too_many` (1 GiB and 200,000 files uncompressed,
2343
- `WORLD_FILE_MAX_BYTES` / `WORLD_FILE_MAX_ENTRIES`; the inflated bytes are counted again,
2344
- since a declared size can lie), `not_a_world` (no `world.json`, unreadable, no string
2345
- `id`), `more_than_one` (two worlds, loose files beside the folder, a nested world.json),
2346
- `not_openable` (not a zip). Archive junk (`__MACOSX`, `.DS_Store`, `Thumbs.db`) is
2347
- dropped. `world.json`'s `api` (the link and its key) is removed on read and on write; a
2348
- `world.json` without one keeps its exact bytes. The writer refuses a `world.json` that will
2349
- not parse rather than zip it raw. Every world in a real Atlas folder (15, up to 3,832
2350
- files) round-trips byte for byte.
2351
- - **First runtime dependency**: `fflate` (zip, no dependencies of its own, the one Atlas
2352
- uses). Tree-shaken away for code that never calls the world-file functions.
2353
-
2354
- ## [4.9.0] — 2026-10-11
2355
-
2356
- ### Added
2357
- - **Only if unchanged.** `patch(type, id, partial, { ifMatch })` and
2358
- `editLinks(type, id, field, edit, { ifMatch })` send `If-Match: "<change_seq>"`: the write
2359
- lands only if the element is still at that `change_seq`. When it has moved, nothing is
2360
- written and the call throws **`OwPreconditionFailedError`** (412 `precondition_failed`), an
2361
- `OwApiError` whose `current` is the element as it is now, so the client's re-read is free.
2362
- `OwApiError.isPreconditionFailed` names the case. The server never merges; the client
2363
- compares the copy it edited from with `current`. An `ifMatch` that is not a whole number
2364
- throws `TypeError` before anything is sent. Without `ifMatch`, nothing changes.
2365
- Checked live on keel-staging: a matching token lands, a stale one is refused with
2366
- `current` at the live `change_seq`, the refused write changes nothing, and the links
2367
- route behaves the same.
2368
-
2369
- ## [4.8.1] — 2026-10-11
2370
-
2371
- ### Changed
2372
- - **Schema repinned `v0.31.0-dist.19` → `v0.32.0-dist.20`** (canonical 00.31.00 → 00.32.00).
2373
- The standard's World gains `public_read`, the walk's first boolean field (`scalar_bool`).
2374
- `OwWorldMeta` already carried `public_read` from `GET /world`; it stays read-only, since
2375
- `PATCH /world` refuses it (flipping a world public is an owner-level act on
2376
- onlyworlds.com). `ONLYWORLDS_VERSION` is now `'00.32.00'`. No element type, field or
2377
- `FIELD_SCHEMA` entry changed.
2378
-
2379
- ## [4.8.0] — 2026-10-10
2380
-
2381
- ### Added
2382
- - **`removeImage(keyOrUrl, { ticket? })`**: removes one uploaded image. A removal ticket from
2383
- keel (`POST /media/remove-ticket`, keel D91), then the removal at the ticket's `remove_url`
2384
- with the ticket only, never the API key. Takes the object key or the whole `image_url`. The
2385
- image's uploader or the world's owner key may remove it; the bytes go back to the uploader's
2386
- storage. Elements keep their `image_url`: clear it yourself.
2387
- - **`createRemovalTicket(key)`** and the types **`OwRemovalTicket`** (`ticket`, `remove_url`,
2388
- `exp` in unix seconds, `key`, `referenced`: how many elements still show the image) and
2389
- **`OwRemovedImage`** (`removed`, `bytes`, `cache`). Refusals arrive as `OwApiError`: keel's
2390
- `permission_error`, `not_found`, `removal_in_flight` and `upload_pending` (both 409 with
2391
- Retry-After); the edge's `ticket_expired`, `ticket_used`, `ticket_invalid`, `not_removable`,
2392
- `remove_failed`.
2393
-
2394
- ## [4.7.3] — 2026-10-10
2395
-
2396
- ### Changed
2397
- - **Schema repinned `v0.31.0-dist.18` → `v0.31.0-dist.19`** (canonical unchanged, 00.31.00).
2398
- Two generated values move: `ELEMENT_ICONS.zone` `architecture` → `pentagon`,
2399
- `ELEMENT_ICONS.marker` `location_on` → `polyline` (Material Symbols names; Marker no longer
2400
- shares the pin glyph). No field, type or `FIELD_SCHEMA` entry changed. The vendored tree grows
2401
- from 31 to 69 files (dist.19 adds `schema.json` and an example world); 69/69 hashed against
2402
- the pinned manifest.
2403
- - **`AGENTS.md` re-read against the new pin** (its gate fired on the repin, as designed); its
2404
- wire facts hold, its "Current as of" line moves.
2405
-
2406
- ## [4.7.2] — 2026-10-09
2407
-
2408
- ### Changed
2409
- - **README first run** finds Admiral Fluffington by name (the demo world's default list order moves with every edit, so "the first three" could not be promised); the npm page shows the corrected output.
2410
- - **Schema repinned `v0.30.2-dist.16` → `v0.31.0-dist.18`** (canonical **00.31.00**).
2411
- `ONLYWORLDS_VERSION` is now `'00.31.00'`. The World gains three optional free-text settings,
2412
- `length_unit`, `mass_unit` and `distance_unit`; the nine element fields that promised "world
2413
- LENGTH / MASS / DISTANCE units" now name them in their descriptions (Ability `range`, Character
2414
- and Creature `height` and `weight`, Creature `speed`, Location `elevation`, Object and Species
2415
- `weight`). No element field shape and no `FIELD_SCHEMA` entry changed. The world body is not
2416
- typed per field (`OwWorldMeta` is open), so `getWorld` returns the three keys and `patchWorld`
2417
- sends them with no code change; a test says so.
2418
-
2419
- ## [4.7.1] — 2026-10-08
2420
-
2421
- ### Changed
2422
- - **README rewritten**: install, a first run against the public demo world (its output shown,
2423
- run as written), your own world with an agent seat for writes, then the reference. It no
2424
- longer calls a `world` field in a payload an error: the API ignores it.
2425
- - Comments, the RFC and pin-file prose no longer name people or internal places.
2426
- - **Schema repinned `v0.30.1-dist.15` → `v0.30.2-dist.16`** (canonical **00.30.02**).
2427
- `ONLYWORLDS_VERSION` is now `'00.30.02'`. Field descriptions move with the schema: Location's
2428
- `rivalries` and `partnerships`, Marker's `order`, and Pin's `element` pair. No field shape and
2429
- no `FIELD_SCHEMA` entry changed: canonical now requires only `name` on every type, which is
2430
- what `FIELD_SCHEMA` already said, so its hand-made exception for the old Pin and Marker
2431
- lists is gone. If the schema lists more than `name` again, codegen stops and asks for a
2432
- ruling instead of dropping the list. 31/31 file hashes recomputed from the tag.
2433
-
2434
- ## [4.7.0] — 2026-10-08
2435
-
2436
- ### Added
2437
- - **`OwApiError.isResyncRequired`**: 409 `resync_required` from `/changes`. Guests can follow the
2438
- feed since keel D79; when a guest's view or a key's role changes, its stored cursor is refused. Drop
2439
- it, walk from zero, and replace the local copy rather than merging.
2440
-
2441
- ### Documented
2442
- - `uploadImage` accepts webp, png, jpeg and avif. 4.6.0's JSDoc and README also listed gif, which
2443
- the edge refuses (415 `unsupported_type`); checked against the Worker's type table.
2444
- - `changes()`: the old rewind rule compared a stored cursor with `head`, which an opaque cursor
2445
- cannot do; replaced with the reset rule (`isResyncRequired`).
2446
- - The doc comments of `OwClientConfig`, `create` and `uploadImage` are written for users: internal
2447
- ruling numbers and test-harness notes are gone, and each opens with what the option or method does.
2448
- The docs site's TypeScript reference is generated from these comments.
2449
-
2450
- ## [4.6.0] — 2026-10-06
2451
-
2452
- Published from `e198298` (tag `v4.6.0`); the tarball was checked against a build of that commit.
2453
-
2454
- Includes 4.5.1's documentation fixes: 4.5.1 was staged and never published.
2455
-
2456
- ### Added
2457
- - **`uploadImage(image, { key?, ticket? })`**: upload one image (webp, png, jpeg, avif;
2458
- never SVG; 4.6.0's docs also said gif, which the edge refuses with 415) and get its permanent public URL. Two requests: a ticket from keel
2459
- (`POST /media/ticket`), then the bytes straight to the edge at the ticket's `upload_url`
2460
- (keel never sees them). Only the ticket goes to the edge, never the API key or PIN. An image
2461
- larger than the ticket's `max_bytes` throws 413 `too_large` before anything is sent; the edge's
2462
- refusals (`ticket_used`, `ticket_expired`, `bad_key`, `exists`, `unsupported_type`, ...) arrive
2463
- as `OwApiError` with their code. Set the returned `url` as an element's `image_url` yourself.
2464
- Checked live on 2026-10-06: one upload through keel's ticket to the edge (201), and the same
2465
- ticket again refused with 401 `ticket_used`.
2466
- - **`createMediaTicket()`**: the first step alone, for callers that upload themselves (progress
2467
- bars, say). Types `OwMediaTicket` and `OwUploadedImage`.
2468
-
2469
- ### Documented
2470
- - `ListParams.filter` lists what keel accepts now: exact `name`, and `characters=<id>` on every
2471
- type with a `characters` link (keel D76d).
2472
-
2473
- ## [4.5.1] — 2026-10-05 (never published; shipped in 4.6.0)
2474
-
2475
- Docs only: no code path changes.
2476
-
2477
- ### Documented
2478
- - **`isOwnerOnly`'s JSDoc in 4.5.0 was wrong about who sees `owner_only`.** It said owners and
2479
- co-builders never see it. A member key, a co-builder's too, gets it on `PATCH /world`; only the
2480
- owner's key is exempt (keel spec section 9; confirmed from keel's `patch_world`,
2481
- 2026-10-04).
2482
- - **`world_gone` is reserved.** `OwAuthErrorCode` and `isAuthError` still carry it, but keel never
2483
- emits it: a deleted world's keys are deleted with it, so the key answers 401 `invalid_credentials`
2484
- (2026-10-04; keel spec section 2). Found by the wire gate reading the SDK's own
2485
- built file for the codes it keys on.
2486
-
2487
- ## [4.5.0] — 2026-10-04
2488
-
2489
- What the staging write probes found on the wire (Python package, 2026-10-04): one new getter, and
2490
- three behaviours the server has always had that no client doc said. Additive; nothing is removed.
2491
-
2492
- ### Added
2493
- - **`OwApiError.isOwnerOnly`**: 403 `owner_only`, a contributor attempting a world-level change
2494
- (`patchWorld`, publishing a show page). Owners and co-builders never see it. keel's OpenAPI
2495
- document does not name the code yet (the schema side has that to close); the wire returns it.
2496
-
2497
- ### Documented
2498
- - **An atomic `/bulk` rollback still returns 201 slots** for the items that would have
2499
- succeeded, with the ids and timestamps they would have had (keel spec: "counterfactual 201s").
2500
- After `atomic: true` with `errors: true`, nothing was written. On `bulk()`, `OwBulkItemResult`
2501
- and the README.
2502
- - **`name` is required as a key only**: `''` and `null` are accepted and stored as `''` (keel
2503
- ruling 2026-09-28, spec §6; nameless Markers exist). On `create()` and the README.
2504
-
2505
- ## [4.4.0] — 2026-10-04
2506
-
2507
- Membership, as far as the server has shipped it: who created an element, and the refusal a
2508
- contributor meets on someone else's. Additive; nothing is removed or narrows.
2509
-
2510
- ### Added
2511
- - **`created_by`** on `OwElementBase` (so on every element type): `string | null`, the id of the
2512
- membership that created the element, `null` for the world's owner and for anything made
2513
- before memberships existed (keel D72 phase 2, live 2026-09-28; keel spec §4). Read-only. The
2514
- schema does not carry it (like `created_at` it is server bookkeeping), so the generator adds
2515
- it to the wire base under the declared mapping. A write body that carries it has it dropped
2516
- by the server, so the client does not strip it and a read body still round-trips.
2517
- - **`OwApiError.isNotAuthor`**: 403 `not_author`, a contributor changing, replacing, relinking
2518
- or deleting an element someone else created. `/bulk` answers 200 and reports it in the item
2519
- slot, so check the slot (documented on `OwBulkItemResult`).
2520
-
2521
- ### Fixed
2522
- - **`FIELD_SCHEMA` no longer marks `marker.map/zone/x/y/order` and `pin.map/element_type/element_id/x/y`
2523
- as `required: true`.** The wire requires `name` and nothing else (rulings.yaml `nullable-by-default`,
2524
- ruling 2026-07-28; keel's OpenAPI write schemas say `required: [name]`), but the schema files still
2525
- list those fields and the generator copied the lists, so a form built from the table demanded
2526
- coordinates the server never did. Found by comparing the SDK with keel's OpenAPI document (the
2527
- new `verify-sdk-wire` gate). **If you read `required` from this table, marker and pin
2528
- fields now read as optional**, which is what the server accepts. The generator notes the lists it
2529
- ignores on every run and says when canonical stops carrying them.
2530
-
2531
- ### Not yet
2532
- - `me()` and `members()` wait for the server routes (keel's M2b). Nothing in this release
2533
- guesses at their names.
2534
-
2535
- ## [4.3.0] — 2026-09-28
2536
-
2537
- Errors you can act on: the two 409s told apart, and the busy server's `Retry-After`, in Node and
2538
- browsers. One getter narrows (see Fixed); nothing is removed.
2539
-
2540
- ### Fixed
2541
- - **`OwApiError.isIdempotencyConflict` no longer claims every 409.** keel sends two: 409
2542
- `idempotency_error` (a key reused with a different body) and 409 `id_conflict` (the id is
2543
- taken: a create with an existing id since D39, or a PUT / bulk id held by another world since
2544
- D70). The getter was true for both, so an id conflict read as a key problem, and retrying with
2545
- a fresh key could never help. It now matches `idempotency_error` only, and the new
2546
- **`isIdConflict`** names the other. A caller who relied on the old catch-all for `id_conflict`
2547
- should switch to `isIdConflict`. A 409 with no code (a proxy, a non-JSON body) now reads
2548
- false for both. `/bulk` is unaffected: it answers 200 and reports a per-item 409 in the slot.
2549
-
2550
- ### Added
2551
- - **`OwApiError.retryAfter`** (seconds, or null) is parsed from `Retry-After`, as a delay or an
2552
- HTTP date, and **`isBusy`** marks keel's 503 `server_busy` (admission control, keel D71: six
2553
- requests at once per worker, up to 10 s of queueing, then 503 with `Retry-After: 5`). The client
2554
- still never retries on its own; it gives callers what they need to back off. A malformed
2555
- `Retry-After` is null, never "retry now". **Works in browsers too** since keel `4949406`
2556
- (2026-09-28): every response exposes `Retry-After`, `Idempotent-Replay` and
2557
- `X-OW-Schema-Version` over CORS (probed from an allowed origin), and keel reports that the
2558
- admission gate's 503 now carries CORS headers, so a busy server reads as busy, not as
2559
- `OwNetworkError` (keel's word; the 503 can't be triggered on demand to probe). Six tests; the
2560
- four core ones watched failing on the previous `errors.ts`.
2561
-
2562
- ## [4.2.0] — 2026-09-28
2563
-
2564
- Ids, honest filter docs, and the schema repin. Nothing is removed; no call needs to change.
2565
-
2566
- ### Changed
2567
- - **`create()` now mints an RFC 9562 UUIDv7** when the element has no id (was v4). The first
2568
- 48 bits are the creation millisecond, so ids this client mints a millisecond or more apart
2569
- sort by creation (while the clock does not step backwards), which gives databases better
2570
- index locality. **This holds within one client only**: worlds also hold v4 ids and legacy
2571
- v1-server ids (`06x…`, which also carry version nibble 7 but put seconds first), so never
2572
- order elements by id. For creation order use `created_at`; `change_seq` is last-write
2573
- order. This is a **default, not a
2574
- requirement**: a v4 or v7 id a caller supplies is still accepted as-is, and
2575
- every id already stored stays valid. Keel mints v7 server-side too (ruling,
2576
- 2026-09-28). Pinned by six tests: version and variant, the big-endian timestamp above
2577
- bit 32, fractional and pre-1970 clocks floored consistently, creation order across 50
2578
- consecutive milliseconds, 1,000 distinct ids inside one millisecond, and the no-`crypto`
2579
- fallback (asserting `Math.random` is actually used). Five injected defects (v4 nibble,
2580
- 32-bit truncation, a dropped random fill, a dropped variant mask, an unfloored clock)
2581
- each fail the suite on every run, on Node 20 and 22. An independent reviewer's decoder agreed on 20,000 random timestamps and the 48-bit
2582
- edges, and keel's `uuid7()` agrees on timestamp, version and variant for a pinned clock.
2583
- - **`elementColor()` on an unknown type now throws a `TypeError` that names the type.** It
2584
- used to crash with `Cannot read properties of undefined (reading 'dark')`. Found while
2585
- building a colour gate: atlas's own copy falls back to the `world` family for an
2586
- unknown type, so swapping it for this export is not a pure re-export for that input, and the
2587
- docstring now says so. Whether the package should fall back is left to its consumers.
2588
- - The generated field-schema comment no longer calls `maximum:` an open question or counts
2589
- its occurrences (the count was 41; since 00.30.01 it is 15). The question was ruled on
2590
- 2026-07-29.
2591
- - **`ListParams.filter` JSDoc names what the server actually accepts**: `name__icontains`,
2592
- `supertype`, `subtype`. It used to list `__in`, `__gte`, `__lte` and `__isnull`, copied
2593
- from keel's spec, which described them but never built them. Probed live 2026-09-28: the
2594
- three accepted keys answer 200; every other one, and `?ordering=`, answers 422.
2595
- - **Schema repinned `v0.30.1-dist.13` → `v0.30.1-dist.15`** (canonical unchanged, 00.30.01).
2596
- Two generated values move: `ELEMENT_ICONS.institution` `business` → `account_balance`,
2597
- `ELEMENT_ICONS.marker` `place` → `location_on` (both Material Symbols names). dist.15
2598
- also adds `minimum: 0` to `ability.potency`; the walk does not surface bounds, so nothing
2599
- generated changes for it. 31/31 file hashes recomputed from a fresh download.
2600
- - **`AGENTS.md` re-read against the new pin** (its gate fired on the repin, as designed).
2601
- It still called the v1 client "frozen legacy" in this package; v1 was removed at 4.0.0.
2602
- It now also names the wire behaviours keel deployed on 2026-09-28 (keel D70): the 64 KB
2603
- extension cap, the three built filters and the `?ordering=` 422, 409 `id_conflict` for a
2604
- PUT or bulk id that belongs to another world, and the 422 for unpaired surrogates. None of
2605
- them needs client code.
2606
-
2607
- ### Also in this release (staged earlier)
2608
- - **`SCHEMA.md` now opens with its own provenance** (dist tag, canonical version, publish
2609
- date — rendered from `schema-pin.json`, never the wall clock). It ships in the tarball
2610
- and is read cold by agents outside this repo, where the pin file is not present; a
2611
- generated reference that cannot name its source was the one remaining self-dating gap.
2612
- - **`AGENTS.md` carries a "Current as of" line, asserted by `codegen:check`** — the tag it
2613
- names must match the pin or CI fails. A hand-maintained agent doc with an ungated date
2614
- is how this repo's previous agent doc went a major version stale. Gate watched firing
2615
- (tampered tag → exit 1, restored → 0).
2616
-
2617
- ## [4.1.0] — 2026-07-29
2618
-
2619
- Public-surface hygiene. Nothing breaks; one member is now marked for removal, and one
2620
- piece of long-standing speculation is retired by measurement.
2621
-
2622
- ⚑ **4.0.2 was tagged in git and superseded before it reached npm.** Everything in it ships
2623
- here — the tag stays as a record rather than being moved or deleted.
2624
-
2625
- ### Deprecated
2626
- - **`FieldType.integer_max` and `FieldInfo.max`** — removal scheduled for **5.0.0**. No
2627
- `FIELD_SCHEMA` entry has ever carried either, in this repository's entire history. They
2628
- existed to surface the schema's `maximum:` constraint, and that constraint is **advisory**:
2629
- keel declares no `MaxValueValidator`, and a `charisma: 9999` write against a `maximum: 100`
2630
- field returns 201 and stores it verbatim. The canonical schema walk therefore stays silent
2631
- on bounds permanently. There is no source to wire them to and no promise they could keep —
2632
- and a public type member meaning "hint the wire ignores" is one consumers read as
2633
- validation. Deprecating now rather than at the major so the signal arrives early; the
2634
- `@deprecated` tags surface in editors via the shipped `.d.ts`.
2635
-
2636
- ### Changed
2637
- - **`TokenResource` is confirmed staying.** RFC-001 §5 asked the platform owner whether the
2638
- token routes were carried long-term or deprecated wire-side, and the question was never
2639
- answered in writing — so this package's own barrel carried "wire fate under review; may be
2640
- removed in a later 4.x" for months, on nobody's authority. Probed against production:
2641
- `GET /api/v2/tokens/status/` and `/tokens/encryption-info/` both return **200**, with a 404
2642
- control on a nonexistent route proving the check meant something. The wire carries it. The
2643
- speculation is retired and the comment now records the evidence instead.
2644
-
2645
- ## [4.0.2] — 2026-07-29
2646
-
2647
- Re-pinned to `v0.30.1-dist.13` (canonical **00.30.01**). No field shape changed and no
2648
- `FIELD_SCHEMA` entry changed — the schema walk is byte-identical between the two pins, so
2649
- nothing about decoding moved. Three things changed and nothing else.
2650
-
2651
- ### Fixed
2652
- - **`construct.relations` and `event.languages` shipped with no description at all** — in
2653
- `types.generated.ts` and in `SCHEMA.md`. Their descriptions were nested one level too deep
2654
- inside `items:` in the canonical YAML, which made them invisible to every consumer that reads
2655
- field descriptions, this package included. `SCHEMA.md` is the package's AI-legibility artifact,
2656
- so the gap landed where it did the most harm.
2657
- - **Five description typos** corrected in published JSDoc and `SCHEMA.md`: `beapplied`,
2658
- `phyiscal`, `relating the`, `object grant`, `eventuated`. The rendered docs site had already
2659
- fixed all five by hand — the downstream copy was the correct one, and nobody noticed because
2660
- the fix went where it was visible rather than where it was true.
2661
-
2662
- ### Changed
2663
- - `ONLYWORLDS_VERSION` `'00.30.00'` → `'00.30.01'`. It is a public `as const`, so its **literal
2664
- type** changes. Depending on that literal is pathological, but it is a type-level change and
2665
- should not be discovered rather than announced.
2666
- - The pin now carries canonical's numeric-bounds correction: the `maximum: 0` sentinel is gone
2667
- from 26 fields, 8 fields gained `minimum: 0`, and `rulings.yaml` carries the numeric-bounds
2668
- row. Nothing in this package consumes bounds — `maximum:` is **advisory** (keel does not
2669
- enforce it; a `charisma: 9999` write returns 201 and stores verbatim) and the walk stays
2670
- silent on bounds permanently. `integer_max` / `max` therefore remain declared no-ops here.
2671
-
2672
- ## [4.0.1] — 2026-07-29
2673
-
2674
- **Metadata correction release.** No wire-path change: the client's reads and writes never
2675
- consulted `FIELD_SCHEMA`, and the generated interfaces carried the correct targets throughout.
2676
- The exposure is anything that builds UI or validation by **iterating `FIELD_SCHEMA`**.
2677
-
2678
- ⚑ **One way this can surface as a compile error**: `relation.relations` is removed, so
2679
- `FIELD_SCHEMA.relation.relations` is now a TypeScript error rather than a value. That is the
2680
- intended outcome — the field does not exist in the standard and the API rejects it — but it can
2681
- break a build rather than only a behaviour.
2682
-
2683
- ### Fixed
2684
- - **`FIELD_SCHEMA.collective.equipment` targeted `construct`; the standard says `object`.**
2685
- This is the founding case of the schema ruling table
2686
- (`collective-equipment-target`, ruled 2026-07-23): the v1 implementation used
2687
- Construct, v1 is decommissioned, and keel serves per YAML. The generated code path
2688
- in this repo was corrected the same week. The hand-maintained `FIELD_SCHEMA` copy of
2689
- the same fact, in the same package, kept shipping the decommissioned value on
2690
- `latest` — the fix went where someone happened to be looking.
2691
- - **`FIELD_SCHEMA.relation.relations` removed** — a `multi_link` to `relation` that does
2692
- not exist in `relation.yaml`. A phantom field, publicly exported, that consumers
2693
- building forms from this table would have sent to an API that 422s unknown keys.
2694
-
2695
- ### Changed
2696
- - **`FIELD_SCHEMA` is now GENERATED** from the pinned schema distribution and gated by
2697
- `codegen:check`, joining `ELEMENT_ICONS` / `ELEMENT_SECTIONS` / `ELEMENT_FAMILIES`. It
2698
- was ~650 hand-maintained lines whose test compared nothing to the schema. Regenerating
2699
- it changed exactly 2 of 467 entries — the two above. Runtime shape and the deeply
2700
- readonly public types (`as const`) are unchanged.
2701
- Two declared deviations from a naive schema read are now stated in the generated file:
2702
- `pin.element` (a `generic-link`) splits into `element_type` + `element_id`, as the wire
2703
- serves it; and `integer_max` / `max` remain in the `FieldType` union unused, because
2704
- the walk does not surface the schema's `maximum:` constraint (41 across 17 types).
2705
- - `codegen/generate_types.py` imports the vendored schema walk instead of carrying its
2706
- own copy of it. Output byte-identical.
2707
-
2708
- ## [4.0.0] — 2026-07-23
2709
-
2710
- **v2-native only.** See `docs/migrating-3-to-4.md`. 3.x stays published forever for
2711
- pinned consumers.
2712
-
2713
- ### Removed (BREAKING)
2714
- - The v1 client (`OnlyWorldsClient`), the v1 `ElementType` enum, `icon-utils`.
2715
- `ElementType` at the root is now THE v2 slug union (was `V2ElementType` alias in 3.x).
2716
- - CJS build — the package is ESM-only with a sealed `exports` map (`"type": "module"`,
2717
- `sideEffects: false`).
2718
-
2719
- ### Changed
2720
- - Metadata tables (`ELEMENT_ICONS`, `ELEMENT_LABELS`, `ELEMENT_SECTIONS`, `FIELD_SCHEMA`,
2721
- `getElementIcon`, `getElementLabel`) ported onto the v2 slug union — same export names,
2722
- same shapes, contents byte-identical EXCEPT: `FIELD_SCHEMA` `type: 'number'` →
2723
- `'integer'` (70 entries; 'number' dropped from `FieldType`). Migration: consumers
2724
- switching on `'number'` switch on `'integer'`.
2725
- - `TokenResource` now takes a structural `TokenTransport` (`{ request<T>(method, path,
2726
- body?) }`) instead of the removed v1 client. Token surface retained pending the
2727
- wire-fate ruling (RFC-001 §5) — may be removed in a later 4.x.
2728
-
2729
- ### Added (post-alpha.0, same night)
2730
- - **`ELEMENT_ICONS` GENERATED** from keel's `icon:` wrapper key (keel `56c124a`).
2731
- - **`ELEMENT_SECTIONS` DERIVED** from the canonical schema's own document structure
2732
- (ruling: sections are the standard tier's property groups; document order is
2733
- display order). Divergence check vs the old hand table found and fixed three fossils:
2734
- creature "Behaviour"→"Behavior", pin's triple-listed generic link → `element`,
2735
- relation "Involves" listing a nonexistent `relations` field. Canonical is truth.
2736
- - **`SCHEMA.md`** — full generated schema reference (every field with its canonical
2737
- description, link directions, families, icons, sections), ships in the tarball;
2738
- covered by the codegen drift guard. AGENTS.md points agents at it first.
2739
- - **`OwV2Client.request()` is public** (typed escape hatch; does not sanitize) and
2740
- structurally satisfies `TokenTransport` — `new TokenResource(client)` just works.
2741
- Token ruling: keel keeps `/tokens/*` long-term; ported, not deleted.
2742
- - README rewritten v2-native (v1 sections, branded types, and the encrypted-key
2743
- walkthrough removed; bulk partial-failure and idempotency-key hygiene promoted).
2744
-
2745
- ### alpha.1 — consumer-pin fixes (atlas review, same night)
2746
- - **`parseEnvelope` → `parseErrorEnvelope`** (it parses the platform *error* envelope;
2747
- renamed before the 4.0 name freeze to avoid collision with the world-export envelope).
2748
- - The never-whitelist LAW now lives as a comment on `READ_ONLY_FIELDS` itself.
2749
- - Stale v1 example in the shipped d.ts fixed (TokenResource JSDoc).
2750
-
2751
- ### Release gates — ALL GREEN (2026-07-23 night)
2752
- - **Gate 3 (wire-log v1-traffic query)**: GREEN — no unknown v1-SDK consumer on
2753
- the wire; observed v1 traffic is pinned deployed builds (council proxy, legacy plugin
2754
- walks) that an npm major cannot touch. The v1 API dialect stays served regardless.
2755
- - **`/api/v2/tokens/` live on prod + staging** (keel `e181689`; three-base byte
2756
- agreement test-pinned). `TokenResource` annotates a 404 on token routes as
2757
- "older server", not "no tokens". Requirement noted in the migration guide.
2758
- - **`ONLYWORLDS_VERSION` now GENERATED** from canonical's `VERSION` file (carried
2759
- into keel `schema/` by the refresh script, keel `492168c`) — the last hand-synced
2760
- constant is dead. Consumer soak: the atlas app pinned the alpha against its full
2761
- static surface — strict-clean compile, correct runtime, package shape approved.
2762
- - **No `npm deprecate` on 3.x** (landscape research; 3.x stays plainly supported).
2763
- - 4.x wishlist: account client (list/mint/create — atlas first consumer), world-export
2764
- envelope reader (§8c), per-status error classes, richer codegen JSDoc, retry/backoff
2765
- parity audit, FIELD_SCHEMA generation with `required:` from YAML.
2766
-
2767
- ## [3.1.0] — 2026-07-23
2768
-
2769
- ### Added
2770
- - **Canonical element colour palette** (`src/v2/palette.ts`, exported from root):
2771
- `ELEMENT_FAMILIES` (type → family), `FAMILY_COLORS` (family → `{light, dark}` hex),
2772
- `familyOf(type)`, `elementColor(type, mode)`, `FAMILY_ORDER`, type `ElementFamily`.
2773
- Four semantic families (agents / world / abstract / temporal) ruled 2026-07-22 after
2774
- CVD-validated measurement.
2775
- Colour carries the FAMILY; `ELEMENT_ICONS` carries the TYPE. Keyed on the v2 slug
2776
- union. First proven live in atlas and council, whose local copies become re-exports.
2777
- **`ELEMENT_FAMILIES` is GENERATED** from the `family:` key in keel's schema YAML
2778
- (keel `c69366b`, a keel presentation-wrapper key — not part of the council-governed
2779
- OnlyWorlds standard); hexes are hand-authored design constants beside it. Membership
2780
- and hex invariants test-gated (`test/palette.test.mjs`); light hexes render-proven on
2781
- the ruled OnlyWorlds light mode (ow-house-light) 2026-07-23.
2782
- - **Codegen drift guard**: `python codegen/generate_types.py --check` fails if
2783
- `types.generated.ts` doesn't match the schema YAMLs (release gate; mirrors keel's
2784
- schema CI job). Codegen also hard-fails on missing/invalid `family:` keys — the
2785
- canary for a canonical-refresh that overwrote keel's wrapper layer.
2786
- - **CI**: GitHub Action building + running the full suite against the compiled bundle.
2787
- - **AGENTS.md** shipped in the tarball — usage guide for AI agents reading the package
2788
- locally, including the SDK-vs-MCP division of labor and the wire gotchas.
2789
- - README: worked create→link→read-back round-trip, schema-verified field by field
2790
- (also fixes the old example patching `'character'` with a location's id); canonical
2791
- colour section; SDK-vs-MCP guidance.
2792
- - This CHANGELOG.
2793
-
2794
- ### Fixed
2795
- - `FieldType` union now includes `'number'` — the legacy alias that 70 `FIELD_SCHEMA`
2796
- entries actually carry (consumers switch on it; data normalization deferred to 4.0).
2797
- - `generate_types.py` exits with a clear message when the sibling keel checkout is
2798
- absent instead of a bare traceback.
2799
-
2800
- ## [3.0.0] — 2026-07-18
2801
-
2802
- - v2-native client (`OwV2Client`) absorbed from an earlier v2 client (ow-v2-client v0.9.0),
2803
- wire-corrected against live staging fixtures (S22). Generated element types from the
2804
- canonical keel schema YAML (`src/v2/types.generated.ts`), validated against all 1,086
2805
- live W11 elements. v1 surface unchanged and frozen.