@conarium-ai/core 0.2.35 → 0.2.37

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.
Files changed (45) hide show
  1. package/CHANGELOG.md +71 -0
  2. package/LIMITATIONS.md +32 -3
  3. package/LIMITATIONS.tr.md +67 -2
  4. package/README.md +6 -3
  5. package/bin/conarium-reconcile.mjs +297 -13
  6. package/bin/conarium-stamp.mjs +2 -2
  7. package/conformance/cases/reconciliation/v2-clocks-declared.json +16 -0
  8. package/conformance/cases/reconciliation/v2-clocks-undeclared-profile.json +16 -0
  9. package/conformance/cases/reconciliation/v2-excluded.json +16 -0
  10. package/conformance/cases/reconciliation/v2-indeterminate.json +14 -0
  11. package/conformance/cases/reconciliation/v2-matched.json +16 -0
  12. package/conformance/cases/reconciliation/v2-neg-unattributed.json +15 -0
  13. package/conformance/cases/reconciliation/v2-neg-undeclared-exclusion.json +16 -0
  14. package/conformance/cases/reconciliation/v2-neg-undeclared-multiplicity.json +16 -0
  15. package/conformance/cases/reconciliation/v2-neg-undeclared-skew.json +15 -0
  16. package/conformance/cases/reconciliation/v2-observed-without-receipt.json +14 -0
  17. package/conformance/cases/reconciliation/v2-receipted-without-observation.json +14 -0
  18. package/conformance/fixtures/reconciliation/after-customers-1.json +1 -0
  19. package/conformance/fixtures/reconciliation/after-customers-15.json +1 -0
  20. package/conformance/fixtures/reconciliation/after-empty.json +1 -0
  21. package/conformance/fixtures/reconciliation/after-set.json +1 -0
  22. package/conformance/fixtures/reconciliation/after-unattributed.json +1 -0
  23. package/conformance/fixtures/reconciliation/before-customers-10.json +1 -0
  24. package/conformance/fixtures/reconciliation/before-empty.json +1 -0
  25. package/conformance/fixtures/reconciliation/profile-clocks.json +9 -0
  26. package/conformance/fixtures/reconciliation/profile-exclusion.json +7 -0
  27. package/conformance/fixtures/reconciliation/profile-multiplicity.json +9 -0
  28. package/conformance/fixtures/reconciliation/profile-no-clocks.json +4 -0
  29. package/conformance/fixtures/reconciliation/receipts-customers-early.jsonl +1 -0
  30. package/conformance/fixtures/reconciliation/receipts-customers.jsonl +1 -0
  31. package/conformance/fixtures/reconciliation/receipts-empty.jsonl +0 -0
  32. package/conformance/run.mjs +59 -0
  33. package/dist/audit.d.ts.map +1 -1
  34. package/dist/audit.js +117 -35
  35. package/dist/audit.js.map +1 -1
  36. package/docs/API-STABILITY.md +1 -1
  37. package/docs/ARCHITECTURE.md +3 -1
  38. package/docs/RECEIPT-SPEC.md +92 -15
  39. package/docs/claims/retracted-phrasings.json +7 -0
  40. package/docs/claims/reviews/0.2.36.md +152 -0
  41. package/docs/claims/reviews/0.2.37.md +180 -0
  42. package/docs/claims/surfaces.json +115 -0
  43. package/docs/security/THREAT-MODEL.md +14 -4
  44. package/docs/security/best-practices-badge.md +5 -3
  45. package/package.json +2 -2
@@ -0,0 +1,152 @@
1
+ # Claims review — 0.2.36
2
+
3
+ - base: v0.2.35
4
+ - head: 7c4a424
5
+ - surface: sha256:4ddf21235110bec0c15f0a676a84aa21cbd406587a3856a869572871d616d72c
6
+ - verdict: pass
7
+ - reviewer: Claude (Opus 5)
8
+
9
+ Four surfaces changed. Three of them were sentences that had been true and
10
+ stopped being true; one is a threat-model paragraph rewritten twice in this
11
+ release, because the first rewrite described a lock that did not hold as tightly
12
+ as it said. The finding worth recording is not any single sentence but that two
13
+ of the four sat in files the claim gate could not see, and the reason it could
14
+ not see them was a rule about directory depth.
15
+
16
+ ## Findings
17
+
18
+ ### 1. src/audit.ts — the receipt sink had no lock, and the audit sink's lock could be taken twice
19
+
20
+ - Claim, as of 0.2.35: `docs/security/THREAT-MODEL.md` — *"Two OS processes
21
+ writing the same audit/receipt sink have no file lock."* Half stale (the audit
22
+ sink had one since 15 August), half true (the receipt sink did not).
23
+ - Mechanism: `acquireSinkLock` at construct time, against `this.sink` only, while
24
+ receipts are appended to `this.receiptSink` — a different path.
25
+ - Why it fell short: two configurations write one receipt file with no shared
26
+ lock — receipt-only installs (no audit sink, so no lock is taken at all) and
27
+ two gateways with separate audit sinks sharing one receipt file. Both fork
28
+ `seq` and `prevHash`, and a forked chain does not verify. Reproduced with two
29
+ OS processes: `duplicate seq: 1,2,3,4,5,1,6,7,8,2,3,4,5,6,7,8`.
30
+ - Disposition: fixed. `log()` now takes every sink path it is about to write, in
31
+ lexicographic path order — one fixed global order, so two locks cannot
32
+ deadlock — and `writeReceipt` re-reads the receipt tail under that lock rather
33
+ than trusting what the instance loaded at startup. A single digest over the
34
+ pair was rejected and the reason is in the code: `audit-a+receipt` and
35
+ `audit-b+receipt` hash to different lock files, which leaves the shared receipt
36
+ file unserialised, which is the bug.
37
+ - Second finding, inside the first: with that fix in place the regression test
38
+ still failed once in five runs. The lock file is created by `writeFileSync`
39
+ with `wx` and the owner pid lands in the next syscall; a second process
40
+ arriving in that gap saw the file exist with an empty body, could not read an
41
+ owner, fell past the `pidAlive` check because there was no pid to check, and
42
+ deleted a live process's lock. So the guard added on 15 August under the title
43
+ *"reject a second OS process on the audit sink lock"* rejected a second process
44
+ most of the time. Its test never landed in the window, and a test that has only
45
+ ever been green proves nothing about the case it was written for.
46
+ - Disposition: an unreadable lock is now honoured, not stolen, until it is 30s
47
+ old. `test/sink_lock_ownership.test.ts` states the window rather than racing
48
+ for it — it plants the empty lock file — and pins the two behaviours the fix
49
+ could have broken: a lock naming a dead pid is still reclaimed at once, and an
50
+ unreadable lock past the threshold is still reclaimed, so a process that dies
51
+ inside the gap does not wedge its sink forever. Red on all three before the
52
+ fix, green after, and the two-process test went 20/20 green where it had been
53
+ 1-in-5 red.
54
+ - Residual, stated in the threat model rather than left for a reader to find: a
55
+ process that dies inside that gap blocks its sink — and, being fail-closed,
56
+ blocks access — for those 30s.
57
+
58
+ ### 2. The claim gate could not see two of the four surfaces
59
+
60
+ - Claim: the release gate reviews the documents that carry promises.
61
+ - Mechanism: `test/denetci.mjs` — `SURFACES` is the reviewed list, and
62
+ `unlistedDocs()` warns about a changed document on neither list.
63
+ - Why it fell short: `unlistedDocs()` matched top-level `.md`/`.html` and
64
+ `docs/*.md` — one level. That reads as a rule about where promises live and is
65
+ a fact about where the function looked. `docs/security/NPM-PROVENANCE.md` was
66
+ reviewed because someone added it to `SURFACES` by hand;
67
+ `docs/security/THREAT-MODEL.md` beside it was on neither list, and nothing was
68
+ capable of saying so. `standards/README.md` was outside both patterns for the
69
+ same reason. Those are exactly the two files that carried stale claims into
70
+ 0.2.35 and were found by an outside reader, not by the gate.
71
+ - Disposition: working records are named as directories now instead of inferred
72
+ from depth, and anything under `docs/` or `standards/` that is neither listed
73
+ nor named trips the warning. Demonstrated failing first: a scratch file at
74
+ `docs/security/` depth 2 is reported by the new rule and is invisible to the
75
+ old one. Widening it immediately surfaced a third stale document nobody had
76
+ looked at — `docs/security/best-practices-badge.md` still said *"Do not submit
77
+ until they have been read once by a human"* four days after the badge was
78
+ filed and awarded. It now points at project 14160 and says it is not updated
79
+ when an answer changes there.
80
+ - Not fixed, recorded: `SURFACES` exists twice, in `test/denetci.mjs` (21 files)
81
+ and `test/claim_discipline.mjs` (12 files), both hand-maintained. Two
82
+ hand-written copies of one fact is the shape that produced most of this
83
+ release. Deriving one from the other is a change for its own release, not a
84
+ rider on this one.
85
+
86
+ ### 3. docs/ARCHITECTURE.md — "transmits nothing outward on its own"
87
+
88
+ - Claim: *"the gateway runs in the customer's network and transmits nothing
89
+ outward on its own."*
90
+ - Mechanism: `announceUpdate()` in `src/update-check.ts`, and the production
91
+ profile in `src/config.ts`.
92
+ - Why it fell short: false as written, and *"on its own"* made it worse rather
93
+ than hedging it — the version check is precisely the outbound call the operator
94
+ did not ask for. The production profile also sets
95
+ `CONARIUM_ANCHOR_SINK=opentimestamps` when the operator left it unset, which
96
+ submits digests to public calendars.
97
+ - Disposition: rewritten to keep the claim that survives measurement — protected
98
+ values do not leave — and to name both outbound connections with their source
99
+ file, what they send, and the variable that disables each. Verified against the
100
+ code rather than the diff: `CONARIUM_NO_UPDATE_CHECK` is read in
101
+ `src/update-check.ts`, `CONARIUM_ANCHOR_SINK` defaults to `none` in
102
+ `src/config.ts:111` and is set to `opentimestamps` at `:113`, and the calendar
103
+ hosts are in `src/ots/calendar.ts`. The first rewrite named `src/index.ts` as
104
+ the caller; `announceUpdate()` is also called from `src/http.ts:350`, and a
105
+ reader auditing outbound traffic needs both, so both are named.
106
+
107
+ ### 4. standards/README.md — draft status, and a warning that was deleted with it
108
+
109
+ - Claim: *"`-04` is being prepared here and has not been submitted."*
110
+ - Mechanism: the IETF Datatracker.
111
+ - Why it fell short: `-04` was posted on 18 August. Restating a status in a file
112
+ that does not hold it means the file is a copy that ages, which is the third
113
+ time in three releases the same shape has failed here.
114
+ - Disposition: the file now points at the Datatracker instead of restating it,
115
+ and the individual-submission limit — not an IETF standard, not adopted by a
116
+ working group — is kept. `test/datatracker_draft_status.mjs` enforces the
117
+ shape: the README must carry the canonical tracker URL, an unreachable
118
+ Datatracker is a printed SKIP rather than a silent green, and a reachable
119
+ mismatch is red. All three states measured here, not taken from the report:
120
+ red against a planted stale sentence, green against the current text, SKIP with
121
+ the endpoint pointed at a closed port.
122
+ - Recorded because it was a deletion, not a rewrite: the rewrite removed a
123
+ paragraph admitting this file had been wrong the same way about `-03`. Its
124
+ function — send the reader to the tracker — is now carried by the structure of
125
+ the file rather than by a warning, and the history is here. That is a
126
+ disposition, not an oversight.
127
+ - Known limit of the new check: the mismatch half matches a phrasing
128
+ (`-NN … has not been submitted`). A differently worded stale sentence slips
129
+ past it. The half that does not depend on phrasing is the requirement that the
130
+ tracker URL be present, and that is the half that makes restating the status
131
+ unnecessary in the first place.
132
+
133
+ ### 5. README.md and package.json — "a receipt of every access"
134
+
135
+ - Claim: the gateway *"hands you a signed, independently verifiable receipt of
136
+ every access it mediates."*
137
+ - Mechanism: `writeReceipt`, reached when `receiptSink` is configured.
138
+ - Why it fell short: unconditional as written. `conarium-init` configures the
139
+ sink, so the typical install does produce a receipt per access — the claim is
140
+ true of the default layout and not true as a property of the gateway.
141
+ - Disposition: both now state the condition and the default in the same
142
+ sentence. `package.json`'s `description` carried the same unconditional
143
+ wording and is corrected here; it was outside the scope given to the reviewer
144
+ who found the README instance, which is a scoping mistake worth naming — the
145
+ same sentence in two files, and only one of them was assigned.
146
+
147
+ ## What did not change
148
+
149
+ The policy gate, masking, the receipt format, the reconciliation tool and the
150
+ conformance vectors are untouched. The engine change is confined to when locks
151
+ are taken and to which paths. `npm run test:all`: 89 test files, 747 passed, 2
152
+ skipped, 52/52 checks.
@@ -0,0 +1,180 @@
1
+ # Claims review — 0.2.37
2
+
3
+ - base: v0.2.36
4
+ - head: ef7b1dd
5
+ - surface: sha256:ccfef14ad8a5dddddd579822a9513c16680ef757c4e3007ed55c6858936357ef
6
+ - verdict: pass
7
+ - reviewer: Claude (Opus 5)
8
+
9
+ Six surfaces changed. The release began as a claim with no call site and a
10
+ result object that decided on bounds nobody had declared; two of the five
11
+ findings below were found while reviewing the fixes for the other three, which
12
+ is the only reason they are in this record rather than in the next one. The
13
+ shape that recurs: a sentence corrected once, in one of the places it was
14
+ written.
15
+
16
+ ## Findings
17
+
18
+ ### 1. docs/RECEIPT-SPEC.md — "Receipts are anchored automatically"
19
+
20
+ - Claim, as of 0.2.36: *"Receipts are anchored automatically. Documents are
21
+ not…"* — unconditional, on a surface that ships in the tarball. The same claim
22
+ was repeated in `README.md` (*"Opt-in anchoring: `CONARIUM_ANCHOR_SINK=…`"*,
23
+ which reads as: setting the variable stamps receipts), in `LIMITATIONS.md`
24
+ (*"Receipts already show `pending`"*), in `docs/ARCHITECTURE.md` (*"that path
25
+ submits a digest"*), and in the header of `bin/conarium-stamp.mjs`.
26
+ - Mechanism: `AnchorScheduler.maybeAnchor`, `createAnchorSinkFromEnv` and
27
+ `appendAnchorSidecar` in `src/anchor.ts`.
28
+ - Why it fell short: none of them has a caller. `src/audit.ts` contains no
29
+ occurrence of the string `anchor`; `src/receipt.ts:399` gives every receipt
30
+ `anchor: null`; the only production files importing `./anchor.js` are
31
+ `src/console.ts`, which classifies an anchor for display, and
32
+ `src/anchor-service.ts`, which is the hosted endpoint. Measured end to end
33
+ with `CONARIUM_ANCHOR_SINK=opentimestamps` and a receipt sink configured: the
34
+ receipt carried `anchor: null` and no `<sink>.anchors.jsonl` was written. The
35
+ variable selects a calendar client for the manual tools; it does not put the
36
+ write path in touch with a calendar.
37
+ - Disposition: the text is narrowed to the mechanism in all five places, and the
38
+ scheduler was deliberately **not** wired. Wiring it adds an outbound network
39
+ hop to the receipt path and changes the two-connection list in
40
+ `docs/ARCHITECTURE.md`; that is a product decision, and making a sentence true
41
+ by adding a network call to satisfy it is the wrong direction of fit.
42
+ - Guard: `test/anchor_wiring.mjs` — an automatic-anchoring claim on a claim
43
+ surface requires a write call in `src/` or `bin/`, excluding tests and
44
+ excluding the definitions themselves. Demonstrated in both directions rather
45
+ than one: with the claim restored and no call it exits 1 and prints the file
46
+ and line; with the claim present and a call added to `src/console.ts` it exits
47
+ 0 and prints the call site. A guard that has only been shown to fail proves it
48
+ can fail.
49
+ - Residual, and it is the same residual the guard has by construction: the rules
50
+ are phrasings. Finding 3 is what that costs.
51
+
52
+ ### 2. bin/conarium-reconcile.mjs — an undeclared exclusion decided the outcome
53
+
54
+ - Claim: the `/2` object implements `coverage-reconciliation/2` as
55
+ `draft-dogru-scitt-disclosure-evidence-04` defines it.
56
+ - Mechanism: `projectResultV2`, and §Exclusions of that draft: *"Exclusion rules
57
+ MUST be stated in the Mapping Profile and are therefore covered by its digest…
58
+ An implementation MUST NOT exclude items by a rule that is not in the
59
+ profile."*
60
+ - Why it fell short: with no profile, an infrastructure pattern was emitted as
61
+ `excluded` with `rule.version: "hardcoded"`, and the run reported
62
+ `outcome: no-exceptions` while `bounds.exclusion` said `undeclared` on the same
63
+ object. It named the bound as undeclared and then decided with it. The
64
+ inversion is what makes it a defect rather than a rough edge: measured on one
65
+ input, a profile that merely omitted the rule produced `indeterminate` and
66
+ `exceptions`, while **no profile at all** produced `excluded` and
67
+ `no-exceptions`. Strictly less declaration bought a strictly cleaner verdict,
68
+ which is Proposition 2b run backwards, on the outcome the draft calls the one
69
+ through which a reconciliation can be made to come out clean.
70
+ - Disposition: with no profile, an infrastructure pattern is `indeterminate`
71
+ (`reason: exclusion-undeclared`), and `bounds.exclusion` is
72
+ `operator-declared` only when the profile lists at least one exclusion rule —
73
+ a profile carrying no exclusions declares nothing about exclusion. Measured
74
+ after the fix on the same input: no profile → `indeterminate` / `exceptions`;
75
+ profile without the rule → `indeterminate` / `exceptions`; profile with the
76
+ rule → `excluded` / `no-exceptions`. Declaration and verdict now move in the
77
+ same direction.
78
+ - Second finding, inside the first: `conformance/cases/reconciliation/v2-excluded.json`
79
+ had pinned the defect. Its `expectOutcome` was `no-exceptions` with
80
+ `expectCounts.excluded: 1`, and its own `rationale` field said *"Without a
81
+ Mapping Profile the exclusion bound is undeclared."* The fixture stated the
82
+ undeclared bound and asserted the decision taken with it. A green vector over
83
+ a defect is worse than no vector: it converts the defect into a contract. The
84
+ case now runs against a profile that declares the rule, and the profile-less
85
+ input is a negative case (`v2-neg-undeclared-exclusion`).
86
+
87
+ ### 3. LIMITATIONS.md — the anchoring correction had a second copy
88
+
89
+ - Claim: *"Bitcoin confirmation still takes hours; receipts still show `pending`
90
+ until upgrade."* (`LIMITATIONS.md`, OpenTimestamps client section.)
91
+ - Mechanism: the same one as finding 1 — a receipt is born `anchor: null`.
92
+ - Why it fell short: the file contradicted itself. Its anchors section, rewritten
93
+ earlier in this same release, says `pending` appears only after an operator
94
+ submits a stamp; two hundred lines down the old sentence was still there, and
95
+ it is the more specific of the two, so a reader resolving the contradiction
96
+ would resolve it the wrong way. `LIMITATIONS.tr.md` had both places corrected,
97
+ so the two languages had also come apart — with the English, the surface that
98
+ ships to npm, holding the stale half.
99
+ - Why the guard did not catch it: the rule was `receipts already show
100
+ `pending``. The sentence says *still*. A rule pinned to one adverb guards one
101
+ sentence, not the claim the sentence makes.
102
+ - Disposition: sentence corrected to match the mechanism; the rule now matches
103
+ *already* / *still* / *are*, and was shown red against the live tree before the
104
+ sentence was touched — the check named `LIMITATIONS.md:249` and exited 1.
105
+ - Recorded because of when it was found: during the release review, not by the
106
+ guard written in the same release for exactly this claim. The guard is worth
107
+ keeping and its limit is worth stating rather than describing it as coverage.
108
+
109
+ ### 4. Two hand-written copies of the claim-surface list
110
+
111
+ - Claim: the release gate reviews the documents that carry promises, and the
112
+ retracted-phrasing scan reads the same documents.
113
+ - Mechanism: `SURFACES` in `test/denetci.mjs` (22 entries) and `SURFACES` in
114
+ `test/claim_discipline.mjs` (12 entries).
115
+ - Why it fell short: the second was a silent subset of the first. Measured: a
116
+ strict subset, no entry in the second that was absent from the first, so ten
117
+ documents — including `standards/README.md` and
118
+ `docs/security/THREAT-MODEL.md`, the two that carried stale sentences into
119
+ 0.2.35 — were exempt from the phrasing scan for no stated reason. 0.2.36's
120
+ record named this and deferred it to its own release; this is that release.
121
+ - Disposition: one file, `docs/claims/surfaces.json`, read by both. Every entry
122
+ carries the reason it is a surface, and the phrasing exemption is an explicit
123
+ per-entry field with a required justification — measured after the change: 22
124
+ entries, 22 reasons, zero exemptions. Widening the scan from 12 to 22 produced
125
+ one new red, `docs/PRIOR-ART.md` matching the Turkish-letter check on a
126
+ registry field name (*Yayın Numarası*); it was resolved by naming the example,
127
+ not by narrowing the list back.
128
+
129
+ ### 5. LIMITATIONS.tr.md — three limitations reached one language and not the other
130
+
131
+ - Claim: the Turkish limitations file is the Turkish limitations file.
132
+ - Mechanism: nothing. Two files, written by hand, with no relation between them.
133
+ - Why it fell short: 24 sections against 21. Missing in Turkish: the Node 20
134
+ end-of-life floor, the audit sink hash not being JCS, and reconciliation
135
+ establishing object attribution rather than per-statement coverage. All three
136
+ are limits a reader is entitled to before choosing this package; a reader in
137
+ one language was told and a reader in the other was not. The same shape lost
138
+ the SOC 2 answer for a release in 0.2.26.
139
+ - Disposition: the three sections written, and `test/limitations_locale.mjs`
140
+ compares the two files by a section key carried in an HTML comment — not by
141
+ heading text, which is translated and cannot be the join. Shown red first: with
142
+ one key removed the check exits 1, names the missing key, and names the
143
+ orphaned heading beside it.
144
+
145
+ ## What this release adds rather than corrects
146
+
147
+ `--profile` reads a Mapping Profile that declares the multiplicity bound, the
148
+ exclusion rules, and the two clocks. The clock fields are **declarations and
149
+ nothing more**: the tool does not verify that `clocks.observation` is the clock
150
+ that stamped the snapshots or that `clocks.receipt` is the one that stamped the
151
+ receipts, and `/2` therefore reports them at `operator-declared`, never
152
+ `measured`. What the declaration does change is the comparison — measured, not
153
+ asserted: against a receipt 3s outside the window, a declared bound of 5s leaves
154
+ the item `indeterminate` (a declaration does not invent a match) and a declared
155
+ bound of 1s moves it to `observed-without-receipt`. `--skew` and `clocks.skew`
156
+ are the same kind of statement, so disagreement between them fails the run rather
157
+ than resolving silently; `5s` and `5000` are the same declaration.
158
+
159
+ The field list and its duration grammar are written out in
160
+ `docs/RECEIPT-SPEC.md` because another specification in the same working group
161
+ asked to adopt the shape rather than invent a second one. That is the reason the
162
+ encoding is stated to the millisecond and the reason `0ms` is documented as a
163
+ declaration rather than a default.
164
+
165
+ ## What did not change
166
+
167
+ The policy gate, masking, the receipt format, `conarium-verify`, and the `/1`
168
+ reconciliation body and its exit codes. `/1` was re-measured on the same input
169
+ after each change and is byte-identical; `test/spec_exitcode_drift.mjs` reports
170
+ the same set. `standards/` is untouched — the published `-04` is immutable, and
171
+ the promises made to that list were kept in code and in `docs/`, not by editing
172
+ a posted draft. `-04`'s Implementation Status now understates this
173
+ implementation in four places; correcting it is `-05`'s work and is not done
174
+ here, because a draft describing an unpublished tree would be the same defect
175
+ pointed the other way.
176
+
177
+ Measured and not a finding, recorded so it is not re-raised: `package-lock.json`
178
+ carries `version: 0.2.24`. The published SBOM does not inherit it — `npm sbom
179
+ --package-lock-only` takes the root component from `package.json` and reports
180
+ `@conarium-ai/core@0.2.37`, 81 components, CycloneDX 1.5.
@@ -0,0 +1,115 @@
1
+ {
2
+ "_comment": "Single source for the claim-surface list. test/denetci.mjs and test/claim_discipline.mjs both read this file. A second copy of the list is how a document became a promise that nobody was assigned to read. phrasing=false is an exemption from retracted-phrasing scan and requires phrasingWhy. phrasing=true is not a silent default — every entry states it.",
3
+ "surfaces": [
4
+ {
5
+ "path": "README.md",
6
+ "why": "Published front page. A reader takes the opening sentences as a promise about what the package does.",
7
+ "phrasing": true
8
+ },
9
+ {
10
+ "path": "README.tr.md",
11
+ "why": "Turkish front page. The same promise surface in the other language; it may be absent from a given tree.",
12
+ "phrasing": true
13
+ },
14
+ {
15
+ "path": "SECURITY.md",
16
+ "why": "Security promises: outbound calls, countersigning, what is and is not operated.",
17
+ "phrasing": true
18
+ },
19
+ {
20
+ "path": "LIMITATIONS.md",
21
+ "why": "The measured-limits surface. A missing limit is itself a claim: the English reader is told something the other language is not.",
22
+ "phrasing": true
23
+ },
24
+ {
25
+ "path": "LIMITATIONS.tr.md",
26
+ "why": "Turkish measured-limits surface. The SOC 2 answer failed to reach this file in 0.2.26 for exactly this reason.",
27
+ "phrasing": true
28
+ },
29
+ {
30
+ "path": "docs.html",
31
+ "why": "Website documentation page. Not in the tarball; a reader of conarium.dev still takes it as a promise.",
32
+ "phrasing": true
33
+ },
34
+ {
35
+ "path": "dpa.html",
36
+ "why": "Data-processing terms a reader can treat as a legal promise.",
37
+ "phrasing": true
38
+ },
39
+ {
40
+ "path": "terms.html",
41
+ "why": "Terms of use. A promise surface even though it is not in the npm tarball.",
42
+ "phrasing": true
43
+ },
44
+ {
45
+ "path": "privacy.html",
46
+ "why": "Privacy page. The 2026-08-17 overclaim ('never receive, see, or store your data') lived here.",
47
+ "phrasing": true
48
+ },
49
+ {
50
+ "path": "docs/ARCHITECTURE.md",
51
+ "why": "Names the outbound connections and the request path. A reader auditing traffic starts here.",
52
+ "phrasing": true
53
+ },
54
+ {
55
+ "path": "docs/RECEIPT-SPEC.md",
56
+ "why": "The receipt-format contract. Exit codes, field meanings, and anchoring mood are taken as specification.",
57
+ "phrasing": true
58
+ },
59
+ {
60
+ "path": "docs/API-STABILITY.md",
61
+ "why": "States which flags and exit codes are candidate-stable. A reader pins a script to this page.",
62
+ "phrasing": true
63
+ },
64
+ {
65
+ "path": "docs/CONTRIBUTING.md",
66
+ "why": "Tells a contributor what the project claims to require. Not a product promise to a customer, but it is public and it is read.",
67
+ "phrasing": true
68
+ },
69
+ {
70
+ "path": "docs/COUNTERSIGN.md",
71
+ "why": "Documents the countersigning endpoint a reader is told to run. The missing bin-registration of conarium-anchor-service lived next to this page.",
72
+ "phrasing": true
73
+ },
74
+ {
75
+ "path": "docs/BENCHMARK.md",
76
+ "why": "Publishes numbers. A number without a method is a claim; a number with a method is still a claim about what was measured.",
77
+ "phrasing": true
78
+ },
79
+ {
80
+ "path": "docs/PRIOR-ART.md",
81
+ "why": "States what was searched and what was not. A reader treating this as legal advice would be wrong; a reader treating the search record as a search record must still get a current one.",
82
+ "phrasing": true
83
+ },
84
+ {
85
+ "path": "docs/CONSENT-BINDING-SPEC.md",
86
+ "why": "A specification page. Field names and binding rules are taken as the format.",
87
+ "phrasing": true
88
+ },
89
+ {
90
+ "path": "docs/security/THREAT-MODEL.md",
91
+ "why": "Two levels down, and for five days that was enough to hide it. The sentence 'two OS processes … have no file lock' stayed in the tree after the audit sink got a lock, because unlistedDocs only looked one level under docs/ and nothing ever asked why this file was absent from both lists.",
92
+ "phrasing": true
93
+ },
94
+ {
95
+ "path": "docs/security/PENTEST-SCOPE.md",
96
+ "why": "States that no independent pentest is on file. That is a claim with an expiry date attached to an event we intend to cause.",
97
+ "phrasing": true
98
+ },
99
+ {
100
+ "path": "standards/README.md",
101
+ "why": "Not in the tarball, public on GitHub, and the place a reader checks what we say our IETF standing is. It carried '-04 has not been submitted' for two days after -04 was posted.",
102
+ "phrasing": true
103
+ },
104
+ {
105
+ "path": "docs/PRICING.md",
106
+ "why": "Ships in the tarball and states a price, a refund window and what each tier delivers. It arrived in 0.2.31 unlisted, and the review that listed it found the Button column pointing at a checkout route that redirects to the waitlist form — a promise surface that no reader was assigned.",
107
+ "phrasing": true
108
+ },
109
+ {
110
+ "path": "docs/security/NPM-PROVENANCE.md",
111
+ "why": "Ships in the tarball and tells a reader how to check where the package came from. Unlisted until 0.2.33, by which point it had gone stale in the worst direction: it said 'this is not a published release' inside eight published releases, and pointed at a verification command that answers 404 for npm provenance. A page about how to distrust us is one a reader must be able to trust.",
112
+ "phrasing": true
113
+ }
114
+ ]
115
+ }
@@ -100,10 +100,20 @@ Pending OpenTimestamps is disclosed as `pending`. Bitcoin upgrade is hours later
100
100
  - A shorter leftover chain still verifies. That is structural, not a bug.
101
101
  - Anchor calendars and `blockstream.info` are third parties. Unreachable →
102
102
  exit 15 ("could not check"), not exit 0.
103
- - Two OS processes writing the same audit/receipt sink have no file lock.
104
- In-process concurrent `log()` is synchronous and keeps the chain (measured:
105
- 50 queries, 0 `prevHash` breaks). Same-millisecond writers across processes
106
- can fork the chain. Single-process installs are the supported shape.
103
+ - The audit sink lock is taken at construct when `sink` is set. The receipt
104
+ sink was not: receipt-only installs, and two processes with different audit
105
+ sinks sharing one receipt file, could fork `seq` / `prevHash`. `log()` now
106
+ takes every sink path it will write, in lexicographic path order, and
107
+ reloads the receipt tail under that lock (see `src/audit.ts`). In-process
108
+ concurrent `log()` is still synchronous (measured: 50 queries, 0 `prevHash`
109
+ breaks). Cross-process writers of the same file are serialized; a lock wait
110
+ that times out throws and fail-closes. A lock file whose owner pid cannot be
111
+ read is honoured, not stolen, until it is 30s old — the window between
112
+ creating a lock and writing the pid into it is otherwise long enough for two
113
+ live writers to claim one sink. The bounded cost is the inverse case: a
114
+ process that dies inside that same window blocks its sink, and therefore
115
+ access, for those 30s. Same-file writers that do not go through `Audit.log`
116
+ (a hand-edited append, a second implementation) are still unsupported.
107
117
 
108
118
  ## Out of scope — the operator
109
119
 
@@ -1,8 +1,10 @@
1
1
  # OpenSSF Best Practices — draft answers
2
2
 
3
- Filing at <https://www.bestpractices.dev/> is a separate decision. These
4
- answers are what the public tree supports today. Do not submit until they
5
- have been read once by a human.
3
+ These were the answers prepared before filing. They were filed, and the filed
4
+ copy is the one that counts: <https://www.bestpractices.dev/projects/14160>.
5
+ Read that page for the current answers and the badge's current level — this
6
+ file is kept for provenance, is not updated when an answer changes there, and
7
+ says nothing about what the badge shows today.
6
8
 
7
9
  Project: Conarium (`@conarium-ai/core`)
8
10
  Homepage: https://conarium.dev
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@conarium-ai/core",
3
- "version": "0.2.35",
3
+ "version": "0.2.37",
4
4
  "mcpName": "io.github.dogrucanemek-alt/conarium",
5
5
  "//files-artefakt": "examples/ altindaki uretilmis dosyalar (ozel anahtar, makbuz/denetim JSONL) ELENIR. .gitignore npm'i baglamaz: izin listesinde 'examples' oldugu icin demoyu kosturmus bir makinede 'npm publish' _keys/audit-ed25519.pem'i (OZEL ANAHTAR) ve makbuzlari yayinliyordu. test/pack_artefakt.mjs bunu kilitler. Elemeler examples/ ile SINIRLI: test-vectors/ altindaki receipts.jsonl ve .pub.pem YAYINLANMAK ZORUNDA, yoksa ucuncu taraf konformans vektorlerini kosturamaz.",
6
6
  "//files": "Pakete NE gireceginin izin listesi. Liste olmadan npm her seyi gonderiyordu: 296 dosya / 1.5 MB — CI workflowlari, gitleaks ayari, 137 KB pazarlama gorseli, sitenin api/chat.js dosyasi. Bir bankanin guvenlik ekibi tarball'i acip bunlari gorur. scripts/ BILEREK var: conarium-reconcile calisirken scripts/pg-snapshot.sql'e atif yapiyor, olmazsa kullanici destek maili yazar.",
@@ -33,7 +33,7 @@
33
33
  "LIMITATIONS.tr.md",
34
34
  "CHANGELOG.md"
35
35
  ],
36
- "description": "Governed MCP gateway for AI data access: policy enforcement and PII masking, an Ed25519-signed receipt per access that a third party verifies offline, coverage reconciliation against the database's own query counters to surface access that bypassed the gateway, and conformance vectors so the receipt format can be implemented without us. Self-hosted, MIT.",
36
+ "description": "Governed MCP gateway for AI data access: policy enforcement and PII masking, an Ed25519-signed receipt per mediated access that a third party verifies offline (written when a receipt sink is configured, which conarium-init sets), coverage reconciliation against the database's own query counters to surface access that bypassed the gateway, and conformance vectors so the receipt format can be implemented without us. Self-hosted, MIT.",
37
37
  "keywords": [
38
38
  "mcp",
39
39
  "ai",