sheleg-design-skill 1.43.0 → 1.45.0

package/CHANGELOG.md

@@ -4,6 +4,233 @@ All notable changes to this project are documented in this file. The format
 follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/); versions
 follow [SemVer](https://semver.org/spec/v2.0.0.html).
 
+## [1.45.0] - 2026-08-20
+
+**Four packs contradicted their own token layers, and the layer is what an implementer
+copies.** `scoreboard` still shipped the `--accent-hover` comment the pack retracted in
+1.13.1 — *"the one orange that may carry a link"*; its elevation comment said "two soft
+stacks and a hairline" where four shadows ship, called both untinted while one is 16% and
+another a blue-black, and left `--shadow-panel` in no prose at all; it banned radii above
+8px while `--radius-pill: 999px` ships and its own section tick requires it.
+`instrument-console` legislated four statuses and shipped two colours, prescribed a pressed
+primary whose label lands at **3.06:1**, and banned "colored shadows besides
+`--accent-glow`" while mandating a composite the prose never named. `showroom` gave its
+signature specimen frame two different radii in two sections, with a subtraction rule keyed
+to the wrong one, and shipped one hex for both `--surface-2` and `--line-weak`, so a
+hairline on a sunken well is invisible. `atrium` had the same collision, unrecorded.
+
+**Six sweeps replace nine hand edits, and they found 24 more defects on their first run:**
+16 shadow or glow tokens no prose names across 11 packs (`proscenium`'s "Elevation is three
+shadows" was four), a card at two radii, three status-map errors, two more hairline
+collisions, a press outside the doctrine's band, three missing base layers. The motion band
+and the UI ceiling are **parsed out of `MOTION_DOCTRINE.md`** rather than restated, and
+`@role drawn-on:` is how a pack declares where a colliding hairline is actually drawn.
+
+Registers that had stopped:
+
+- **Run stamps complete to 1.44.0** — 26 rows reconstructed from `git log`, tags and the
+  CHANGELOG, with `Diverged?` left honestly *unrecorded*; a release at or above 1.5.0 with
+  no row now fails. The register had stopped at v1.26.0, eighteen versions back, which made
+  the "not fired in five run stamps" retirement trigger uncomputable.
+- **The palette gate's own numbers**: it claimed 71 of 121 unguarded ratio claims measured
+  at sixteen packs and asserted all 71 had been hand-verified. Measured today: **513 claims,
+  336 guarded, 177 unguarded at 22 packs** — and the run prints that pair every time, with
+  the hand-verification dated for what it was. Two more restated counts went the same way.
+- **`package.json` said twenty-nine packs and named 27** — `awning` and `bulletin` missing
+  from the surface an npm reader chooses from. It is an enumeration site now.
+- **Two `## [1.35.0]` sections** under one version, and four releases in the CHANGELOG with
+  no tag. The duplicate is resolved; the untagged four are **reported, not created** — a tag
+  nobody has is not this run's to invent.
+- **Every pack that bans a weight or a slant owes a base layer** and two of 29 shipped one:
+  `<strong>` renders 700 and `<em>` italic with no stylesheet involved, so the ban was
+  invisible to a grep over CSS.
+
+Two plants found holes in the checks they were written for, and two bugs the gate found in
+itself: one sweep was reading a declaration out of a header comment and silently excluding
+a whole pack, and a new check emitted a different number of checks with and without git
+tags, which broke the ratchet in its own self-test copy.
+
+**Found by the release, in the release's own gate.** The untagged-release check read
+`bool(tags)` as "the whole tag set is visible". A release checkout fetches the released
+ref and nothing else, so at `v1.45.0` exactly one tag was present and the other 49 shipped
+releases read as untagged — the release went red for a reason with nothing to do with the
+release. The same commit passed on `main` because a shallow checkout fetches NO tags, the
+check switched itself off, and the difference between *no view* and *a one-tag view* was
+invisible. Both are named as partial now, and a partial view is disclosed rather than
+answered. Verified against a `--depth 1 --branch v1.45.0` clone, which is the environment
+that failed.
+
+Checks: `validate.py` 3398 to **3543**, `validate_palette.py` 1912 to **1924**, `sloplint.py`
+635. Ten new plants, all caught; 49 planted defects across the three self-tests, 0 missed.
+
+## [Unreleased]
+
+### Four packs contradicted themselves, and six sweeps now say so
+
+A reader copies the token layer verbatim — every pack's Palette section tells
+them to — so when the layer and the prose disagree, the layer wins and the prose
+lies. Nine such contradictions were filed by hand across `scoreboard`,
+`instrument-console`, `showroom` and `prism`. All nine are fixed, and where the
+class repeats it is a sweep with a planted defect rather than nine edits.
+
+**`scoreboard`** shipped `--accent-hover: #e03d00; /* the one orange that may
+carry a link */` — the exact permission the pack retracted in 1.13.1, still in
+the file a reader copies (B-033). Its elevation comment counted "two soft stacks
+and a hairline" over four shadows, called `--shadow-card` untinted when
+`rgba(16, 24, 40, …)` is a blue-black, and left `--shadow-panel` in no prose at
+all (B-035). Its Bans forbade "radii above 8px" while `--radius-pill: 999px`
+ships and the section tick requires it, and cited `font-smooth: never` — not a
+legal value of the property the pack sets, and not a property any engine
+implements (B-036).
+
+**`instrument-console`** legislated four statuses and shipped two colours, so
+`var(--danger)` and `var(--info)` were undefined custom properties — invalid at
+computed-value time, falling back with no error anywhere — in the pack that
+required them (B-038). Its pressed primary kept a label at **3.06:1**, one row
+under its own warning that `--accent-dim` is a fill and not a label (B-039). And
+it called `--accent-glow` "the only permitted glow" while mandating
+`--signal-glow`, a composite its prose never named — so the pack banned its own
+signature motif (B-041).
+
+**`showroom`** specified its specimen frame at `--radius-2xl` in one section and
+`--radius-3xl` in another, with a subtraction rule keyed to the outer value, so
+every inner radius inherited the 4px error (B-022). It shipped one hex for two
+jobs — `--line-weak` and `--surface-2` are both `#edeff3`, which makes a hairline
+on the sunken well 1.00:1 — and a comment naming a colour the focus ring stopped
+using fourteen releases ago (B-023). Three of its motion values contradicted the
+doctrine in the same bundle (B-024). **`prism`** prescribed its CTA press over
+`--dur-fast` at 200 ms, past the doctrine's 100–160 ms press band, with nothing
+faster in the layer to reach for.
+
+### Six sweeps, twenty-four findings beyond the nine
+
+- **`validate_elevation_tokens_named()`** — every `--shadow-*` and `--*-glow` a
+  pack declares is named in the pack's prose. **Sixteen unnamed tokens across
+  eleven packs** on its first run; `proscenium`'s own "Elevation is three
+  shadows" turned out to be four.
+- **`validate_radius_single_valued()`** — a component is given one radius. It
+  attributes a token to the noun that FOLLOWS it, so the nesting sentence every
+  pack writes reads as one rule per noun. Found `scoreboard`'s card at two radii.
+- **`validate_motion_bands()`** — a press sits inside the doctrine's press band,
+  a duration literal in prose is not a value the layer already holds, and a
+  control stays under the UI ceiling. The band and the ceiling are **parsed out of
+  `MOTION_DOCTRINE.md`**, so the gate cannot disagree with the document it
+  enforces.
+- **`validate_emphasis_base_layer()`** — a pack that bans a weight or a slant
+  ships the base rule for it. `<strong>` renders 700 and `<em>` renders italic
+  with no stylesheet involved, so the ban was invisible to every grep over CSS
+  and to every browser. Two of twenty-nine shipped one; `atrium`, `notation` and
+  `field-notes` now do too (B-044).
+- **`validate_status_vocabulary()`** — the cross-pack status map in
+  `SURFACE_COMPOSITION.md` is a table, compared against every token layer with
+  set equality. It was prose, and wrong in three places (B-016's live half).
+- **`validate_line_surface_collision()`** in the palette gate — a line token and
+  a surface token resolving to the same colour fail unless the layer declares
+  `@role drawn-on:`. Found the unrecorded instance in `atrium` and the
+  deliberate-but-undeclared one in `maquette`.
+
+### The doctrine took two corrections
+
+"UI motion stays **under** 300 ms" is now "**at or under**", because a control may
+sit on the ceiling with a reason and 301 ms is over — and a gate needs a boundary.
+And the "hover/entrance motion stays sub-500ms" sentence in `SKILL.md`,
+`SHELEG_DESIGN.md`, `README.md` and the Cursor rule was the wrong half of its own
+contradiction: it lumped hover with entrance and set a bound that `manpage`,
+`tenor`, `bulletin`, `scoreboard`, `field-notes` and `roster` all exceed with
+measured values. An entrance is not UI motion, may run longer when the value is
+measured, and never gates content.
+
+### Three registers that had stopped describing the tree
+
+- `CHANGELOG.md` carried **two `## [1.35.0] - 2026-08-15` sections** — two runs
+  took the number, the tag went to the second, and the loser's notes sat *above*
+  the winner's, so a release extractor reading the first match would have shipped
+  the wrong notes. One heading now, and the superseded section says what happened.
+- `docs/evidence/retro.md`'s Run stamps table stopped at `v1.26.0` while `1.44.0`
+  shipped — eighteen versions — so the *has not fired in five run stamps*
+  retirement trigger read every standing instruction as dormant. Twenty-six rows
+  reconstructed from `git log`, the tags and the CHANGELOG, with `Diverged?` left
+  *unrecorded* rather than guessed (B-037, SG-04).
+- `validate_release_register()` gates all of it: no duplicate version, a stamp for
+  every release at or after `1.5.0`, and a missing tag reported against a declared
+  list. **Reported, never created** — a tag nobody has is not a gate's to invent.
+
+### Numbers that were restated instead of computed
+
+- `test/validate_palette.py` asserted "71 of 121 (59%) … all 71 were recomputed by
+  hand … nothing is wrong today" at sixteen packs. Counted today: **513 claims,
+  177 unguarded (35%) at 22 packs**, and the hand-verification covered 71 of 121
+  on 2026-08-12 and has never been repeated. The figures print on every run and
+  the comment says plainly which of them was ever verified.
+- `test/validate.py` said "the six packs shipped before the widening" (seven) and
+  "seven of the ten widened packs" (twenty-two widened). The split is printed now.
+- `package.json` said "twenty-nine locked style packs" over a list of **27** —
+  `awning` and `bulletin` were missing. `ENUMERATION_SITES` gained `package.json`,
+  which is the description npmjs.com renders: its count was already policed and
+  the names beside it were not.
+
+### Fixed
+
+- `_root_block()` stripped no comments, so `datasheet.css` — whose header explains
+  the reference's rem base with `html { font-size: 8px }` — sliced a
+  nine-character root block and was silently excluded from two sweeps. A block
+  scanner that reads a comment as code skips exactly the packs whose authors
+  explained themselves most.
+- `validate_release_register()` emitted one check when `git` could see no tags and
+  two when it could, so the check count moved with the environment and the floor
+  measured on a real checkout failed inside the self-test's copy. Both branches
+  emit two now.
+
+### Gate
+
+`npm test` → **3543 / 1924 / 635**, floors raised in `test/floors.json` with the
+delta attributed per function. Ten new plants, all watched being refused; two of
+them found holes in the checks they were written for.
+
+## [1.44.0] - 2026-08-19
+
+### Degrade to calm gained an observable
+
+The promise had none. `sloplint.py` asserted only that the **string**
+`prefers-reduced-motion` appeared in the doctrine — that the rule was mentioned, not that any
+shipped artifact obeyed it. Measured: **2 of 29 token layers shipped no reduce branch**,
+including the pack that mandates a particle field.
+
+`validate_reduced_motion()` asserts two things per layer: a layer declaring `--dur-*` or
+`--ease-*` ships a `prefers-reduced-motion` branch, **and that branch collapses at least one
+time-valued property to ≤1ms**. The second half is the row — doctrine §9 says motion collapses
+"to static or instant, not to slower", so a branch that merely names the query satisfies a grep
+and changes nothing a reader can feel. Durations are matched **by value, not by name**, so
+`paperclip`'s `--t-*` and `instrument-console`'s `--motion-ease` are not read as empty.
+
+`editorial-luxury` and `instrument-console` fixed. Where a motif cannot be reached from a custom
+property — a rAF particle field is not stoppable by a duration — the still is now stated for each:
+field to one static frame, spotlight to every section lit, instruments fully drawn. 29 of 29
+layers now carry a branch that bites. B-040 closed.
+
+### A count hyphenated onto its noun was invisible to the gate that counts it
+
+`docs/DOCMAP.md` said "a fourteen-kit build matrix" against **29** kits, and the counted-claims
+guard ran green over that string for weeks: `COUNTED` never matched a number spelled as a word and
+joined to its noun by a hyphen. A contributor had already routed around the hole rather than
+report it.
+
+The separator is now `[ -]`, **captured and back-referenced**, so a compound is read in one
+alphabet or the other and a half-and-half form is not silently accepted; the plural is required
+for the spaced form only, because a hyphenated count is a singular modifier by grammar.
+Blast radius measured old-pattern-against-new on one tree: the policed set 34 → 36 spans with
+**zero false positives**, the whole tree 333 → 384.
+
+The coordination claim is **derived** rather than reworded — DOCMAP now points at the generated
+`AGENT_SYNC.md`, and a new check holds the derivation from both ends so the page, the config and
+the document cannot disagree.
+
+### Plants
+
+Self-test 16 → 19, each asserting its **message** rather than merely the red — every token layer
+is copied into its kit byte for byte, so any edit trips the kit-drift check and a plant would
+otherwise report "caught" with the intended check never having run.
+
 ## [1.43.0] - 2026-08-17
 
 **The twenty-ninth pack, and the first one whose signature is a defect its reference
@@ -508,6 +735,75 @@ accessibility defects in it, and neither was reachable without building a page.
 
 ## [1.35.0] - 2026-08-15
 
+> **The number was already in use.** A concurrent run had written a
+> `## [1.35.0]` section for `ledger` before this one, and `v1.35.0` tags this
+> commit — so the ledger notes moved to a heading of their own, directly
+> below. Standing instruction 1 in `docs/evidence/retro.md` is about exactly
+> this, and it caught the collision late rather than never.
+
+**A twenty-second pack, and the reference it was asked for did not survive measurement.**
+The request was "a pack in the style of taskip.net". Taskip turned out to be WordPress and
+Elementor over a bought theme (Xilancer), carrying **three token systems that disagree** —
+the theme's declared primary `#6074f6` appears nowhere on the rendered page, the visible
+emerald lives in a second sheet, and a third set of values is Tailwind slate. Twelve radius
+values, six shadows, seven weights, four font families, and `.3s` / `0.3s` / `300ms` written
+as three different things. That is a page-builder output, not a vocabulary, and this library
+exists on the premise that a pack is the reference's own vocabulary rather than a
+reconstruction of one. It was declined as a source and the register was re-sourced instead.
+
+### Added — `awning`, the twenty-second pack
+
+Origin: <https://www.shopify.com>, read 2026-08-15 from the served HTML plus its six linked
+stylesheets. Chosen after measuring five candidates on the same axes; it was the only one
+that paired the register with a real system.
+
+- **The accent is black, and it is a resolved chain rather than a stylistic absence.**
+  `--color-component-button-primary-bg` → `--color-theme-bg-cta` → `#000`, with hover, active
+  and disabled declared beside it. No hue reaches the chrome at all, which is what leaves
+  every colour on the page belonging to the product screenshot inside it.
+- **A three-tier token layer** — primitives, semantic roles, per-component states — so the
+  pack ships the indirection rather than the resolved values. `--radius-component-button:
+  var(--radius-theme-full)` is the only place the system says *why* a button is a pill.
+- **420 and 550, and no 700 anywhere.** A variable grotesque used as one. Setting a heading
+  in 700 here is not a small deviation; it is the one number the system was built to avoid.
+- **Tracking crosses zero inside one family** — negative on the display ramp, positive on the
+  body ramp, crossover around 1.375rem — where most packs in this library split tracking
+  across two faces.
+- **Leading ships paired with size in `rem`**, not as a ratio, and the ratio *changes* down
+  the ramp: 1.08 at display, 1.12 at t2, 1.30 at t7.
+- **One shadow, three layers** — ambient, contact, and a `0 0 2px` hairline edge that is what
+  stops a card dissolving on pure white. That third layer is the one people drop when they
+  copy a shadow by eye.
+- Six Gotchas, three of them defects in the reference: `--ink-faint` at **4.40:1 on the
+  system's own second field**; three eases named as tokens against **ten unnamed inline
+  durations**; and `ease-in` shipped as a token while the doctrine bans it in UI.
+- Routed through all seven surfaces the validator checks, mirrored to `.cursor`, added to
+  `install.sh`, and shipped with `kits/awning` — the six-name spine on the canonical API plus
+  `ProductFrame`, `PlanCard` and `FeatureRow`.
+
+### Fixed
+
+- `paperclip` and `showroom` now link **back** to `awning`. The validator caught the one-way
+  fork: a neighbour reference that only points one direction is a dead end for anyone who
+  reaches the other pack first.
+- Nine stated counts moved from twenty-one to twenty-two across `SKILL.md`, `README.md`,
+  `package.json`, both manifests, `bin/cli.js`, `MOBILE_SURFACES.md`, `SURFACE_COMPOSITION.md`
+  and `DESIGN_SYNC_BRIDGE.md`, plus core-contract packs from six to seven. Every one was
+  caught by the gate rather than remembered.
+
+## [ledger — shipped inside 1.35.0] - 2026-08-15
+
+> **Two runs took the number 1.35.0, and the tag went to the other one.**
+> `v1.35.0` is `e9c0bf6`, the `awning` release below; this section's work is
+> `48f24d9`, which carried `1.35.0` in every manifest and was never tagged —
+> `git checkout v1.35.0` gets you both, because the ledger commit is an
+> ancestor of the tag. Two identical `## [1.35.0]` headings sat here from
+> 2026-08-15 until 1.45.0, the ledger one ABOVE the tagged one, so a release
+> extractor reading the first match for `## [1.35.0]` would have shipped the
+> wrong notes. The heading is no longer a version because the version was
+> never this section's to hold, and `validate_release_register()` now refuses
+> a duplicate.
+
 **The twenty-first pack, and two counts that had been wrong for eight releases.**
 `ledger` was extracted from `basedash.com` — a warm cream console for a product
 that answers questions about data. Adding it walked the library past twenty for
@@ -570,58 +866,6 @@ table that stopped at twenty and a regex that read one modifier word.
   check that would have caught it — a name in a context that routes to it — is
   not written.
 
-## [1.35.0] - 2026-08-15
-
-**A twenty-second pack, and the reference it was asked for did not survive measurement.**
-The request was "a pack in the style of taskip.net". Taskip turned out to be WordPress and
-Elementor over a bought theme (Xilancer), carrying **three token systems that disagree** —
-the theme's declared primary `#6074f6` appears nowhere on the rendered page, the visible
-emerald lives in a second sheet, and a third set of values is Tailwind slate. Twelve radius
-values, six shadows, seven weights, four font families, and `.3s` / `0.3s` / `300ms` written
-as three different things. That is a page-builder output, not a vocabulary, and this library
-exists on the premise that a pack is the reference's own vocabulary rather than a
-reconstruction of one. It was declined as a source and the register was re-sourced instead.
-
-### Added — `awning`, the twenty-second pack
-
-Origin: <https://www.shopify.com>, read 2026-08-15 from the served HTML plus its six linked
-stylesheets. Chosen after measuring five candidates on the same axes; it was the only one
-that paired the register with a real system.
-
-- **The accent is black, and it is a resolved chain rather than a stylistic absence.**
-  `--color-component-button-primary-bg` → `--color-theme-bg-cta` → `#000`, with hover, active
-  and disabled declared beside it. No hue reaches the chrome at all, which is what leaves
-  every colour on the page belonging to the product screenshot inside it.
-- **A three-tier token layer** — primitives, semantic roles, per-component states — so the
-  pack ships the indirection rather than the resolved values. `--radius-component-button:
-  var(--radius-theme-full)` is the only place the system says *why* a button is a pill.
-- **420 and 550, and no 700 anywhere.** A variable grotesque used as one. Setting a heading
-  in 700 here is not a small deviation; it is the one number the system was built to avoid.
-- **Tracking crosses zero inside one family** — negative on the display ramp, positive on the
-  body ramp, crossover around 1.375rem — where most packs in this library split tracking
-  across two faces.
-- **Leading ships paired with size in `rem`**, not as a ratio, and the ratio *changes* down
-  the ramp: 1.08 at display, 1.12 at t2, 1.30 at t7.
-- **One shadow, three layers** — ambient, contact, and a `0 0 2px` hairline edge that is what
-  stops a card dissolving on pure white. That third layer is the one people drop when they
-  copy a shadow by eye.
-- Six Gotchas, three of them defects in the reference: `--ink-faint` at **4.40:1 on the
-  system's own second field**; three eases named as tokens against **ten unnamed inline
-  durations**; and `ease-in` shipped as a token while the doctrine bans it in UI.
-- Routed through all seven surfaces the validator checks, mirrored to `.cursor`, added to
-  `install.sh`, and shipped with `kits/awning` — the six-name spine on the canonical API plus
-  `ProductFrame`, `PlanCard` and `FeatureRow`.
-
-### Fixed
-
-- `paperclip` and `showroom` now link **back** to `awning`. The validator caught the one-way
-  fork: a neighbour reference that only points one direction is a dead end for anyone who
-  reaches the other pack first.
-- Nine stated counts moved from twenty-one to twenty-two across `SKILL.md`, `README.md`,
-  `package.json`, both manifests, `bin/cli.js`, `MOBILE_SURFACES.md`, `SURFACE_COMPOSITION.md`
-  and `DESIGN_SYNC_BRIDGE.md`, plus core-contract packs from six to seven. Every one was
-  caught by the gate rather than remembered.
-
 ## [1.34.0] - 2026-08-15
 
 **The pack was applied correctly and the result still read as somebody else's typography.**

package/README.md

@@ -88,7 +88,8 @@ motion layer never hard-codes a palette.
 3. **Hold, then redeploy.** Hold a formation ~80% of a section, then morph in a
    short, phase-staggered, arc-curved wave. No crossfades.
 4. **Earned motion.** Scrub is for instruments that narrate state over time;
-   entrance motion stays sub-500ms and never gates content.
+   hover and press stay inside the doctrine's bands, an entrance may run past
+   them when measured, and neither gates content.
 5. **Degrade to calm.** Reduced-motion / coarse pointer / no-WebGL collapse to a
    static, fully-legible page.
 

package/cursor/rules/sheleg-design.mdc

@@ -119,7 +119,7 @@ Frequency decides before taste gets a vote. (Full doctrine:
   `linear`. Curves: `--ease-out: cubic-bezier(.23,1,.32,1)`,
   `--ease-in-out: cubic-bezier(.77,0,.175,1)`,
   `--ease-drawer: cubic-bezier(.32,.72,0,1)`.
-- **UI motion stays under 300 ms.** Button press 100–160ms · tooltip 125–200ms ·
+- **UI motion stays at or under 300 ms.** Button press 100–160ms · tooltip 125–200ms ·
   dropdown 150–250ms · modal/drawer 200–500ms. Marketing and scrollytelling are
   deliberately longer.
 - Never animate a property that triggers layout (`width`, `height`, `top`,
@@ -143,7 +143,8 @@ Frequency decides before taste gets a vote. (Full doctrine:
    a short, phase-staggered, arc-curved wave. Crossfades between point sets
    are banned.
 4. **Earned motion.** Scrub only for instruments that narrate state over time
-   (charts, step flows); hover/entrance motion stays sub-500ms and never
+   (charts, step flows); hover and press stay inside the duration bands above,
+   an entrance may run past them when the value is measured, and neither ever
    gates content visibility.
 5. **Degrade to calm.** `prefers-reduced-motion` / coarse pointer / no-WebGL
    collapse to a static, fully-legible page. The effect is a bonus, never a

package/kits/atrium/src/styles.css

@@ -9,7 +9,11 @@
   --surface-ink: #2a2b2f; /* the ONE inverted surface (media wells, banner) */
   --line: #d1c9bf; /* hairline — decorative strength, see Gotchas */
   --line-strong: #737373; /* borders that carry an affordance (inputs) */
-  --line-ink: #2a2b2f; /* ghost-control border */
+  /* @role drawn-on: --bg, --surface — the ghost-control border, and the SAME hex
+     as --surface-ink (#2a2b2f): on the inverted banner it disappears completely.
+     A ghost control that lands on --surface-ink borders in --on-ink instead
+     (13.47:1), which is what the pack's Gotchas now say. */
+  --line-ink: #2a2b2f;
 
   /* ---- Ink ------------------------------------------------------------- */
   --ink: #2a2b2f; /* 13.5:1 on --bg */
@@ -167,6 +171,13 @@
   color-scheme: light;
 }
 
+/* The base layer. This pack bans bold outright — its entire emphasis vocabulary is
+   the italic terracotta aside — so `<strong>` is pointed at the one weight above
+   body that the system owns. Left to the UA it renders 700, a weight no face here
+   ships, and inside the 300-weight serif the browser fakes it. `<em>` is deliberately
+   untouched: italic IS this pack's emphasis, and the serif has a real one. */
+strong, b { font-weight: var(--weight-semibold); }
+
 /* The hero effect is a bonus, never a dependency: at reduced motion the canvas
    is REMOVED and the still image underneath is the hero. Do not slow it down —
    a slow refraction is still refraction. */

package/kits/editorial-luxury/src/styles.css

@@ -76,6 +76,32 @@
   --bg: var(--paper); /* role alias -> --paper, this pack's field */
 }
 
+/* Degrade to calm — MOTION_DOCTRINE.md §9: "Shipping an animation without a
+   reduced-motion path is a bug, not a polish item." All eight durations this
+   layer declares are spent — the two product-register values by the kit's
+   hovers, the brand and the SHELEG set by the page's entrances — so all eight
+   collapse to 0s rather than to anything a reader could still perceive as
+   motion. The kit's .el-btn, .el-card and .el-dossier read --dur-fast and
+   --dur-base, so a hover lands on its final state instead of rising into it.
+
+   The two curves stay as declared: a curve spent over 0s is inert, and neither
+   overshoots (both stay inside 0..1 on the value axis), so there is no bounce
+   left to tame. What a custom property cannot express is an infinite animation
+   — this pack declares none, and one added later needs its own stop in the
+   component layer, not a smaller number here. */
+@media (prefers-reduced-motion: reduce) {
+  :root {
+    --dur-hover: 0s;
+    --dur-state: 0s;
+    --dur-brand: 0s;
+    --dur-fast: 0s;
+    --dur-base: 0s;
+    --dur-slow: 0s;
+    --dur-epic: 0s;
+    --stagger: 0s;
+  }
+}
+
 /* ── components ── */
 
 /* Everything below consumes var(--…) only. The token block above is the whole

package/kits/field-notes/src/styles.css

@@ -288,6 +288,15 @@
   --focus-ring: var(--brand);
 }
 
+/* The base layer, and it is a synthesis rule rather than a weight or a style.
+   Bricolage Grotesque has no italic, so an `<em>` in a display heading gets a
+   browser-slanted fake — the "irregularity is the whole personality" argument in the
+   Gotchas, undone by a shear transform. `font-synthesis-style: none` refuses the
+   fake without touching Geist's real italic, which the body copy does use. The
+   declared `font-style` is the UA's own value, stated so the rule says what it keeps
+   as well as what it removes. */
+em, i { font-style: italic; font-synthesis-style: none; }
+
 /* The reference ships a reduced-motion branch for every animated class, which
    is the right posture and this pack requires it. Durations collapse; the
    reveal resolves to its final state instead of transitioning to it. */

package/kits/instrument-console/src/styles.css

@@ -14,15 +14,41 @@
   --accent-dim: #1f5fb0;
   --accent-bright: #6bb3ff;
   --accent-ink: #0a0e14; /* text ON the accent — 6.17:1. White on #3392ff is 3.14 and fails. */
+  /* Two glows, two jobs, and both are named in the pack's Bans. --accent-glow is
+     the translucent halo behind a focused control; --signal-glow, declared below,
+     is the composite box-shadow the ONE active signal element wears. Until 1.45.0
+     the prose called --accent-glow "the only permitted glow" and then mandated the
+     composite, which it never named -- so the pack banned its own signature. */
   --accent-glow: rgba(51, 146, 255, 0.18);
+
+  /* ---- Status: four roles, and the pack owns three hues ------------------
+     The prose has legislated success, warning, danger and info together since
+     this pack's first release, and the token layer shipped two of them -- so
+     `var(--danger)` was an undefined custom property, which is invalid at
+     computed-value time and falls back with no error anywhere (B-038).
+     --info is the signal hue by decision, not by accident: an infrastructure
+     console's "here is a fact" and its "here is the thing to press" are the same
+     blue, and giving info its own hue would break the pack's one-accent rule.
+     --danger is a PACK DECISION, not a measurement: the reference this pack was
+     read off paints no danger state, so there was nothing to measure. Chosen to
+     clear AA on every field this pack ships (6.02:1 on --base, 5.77 on
+     --surface-1, 5.10 on --surface-3) and to stay above the library's separation
+     floors against every peer, including under simulated CVD -- the numbers are
+     in the pack's Palette. */
   --ok: #46d39a;
   --warn: #e0a030;
+  --danger: #f4526b;
+  --info: var(--accent);
 
   --r-sm: 4px;
   --r-md: 8px;
   --r-lg: 14px;
   --r-pill: 999px;
 
+  /* THE signature shadow: the one active signal element wears it, and nothing
+     else on the page does. Named here and in the pack's Texture, Micro-interactions
+     and Bans -- a mandated composite that appears in no prose is a value a reader
+     has to reconstruct from a literal. */
   --signal-glow: 0 0 0 1px rgba(51, 146, 255, 0.4),
     0 8px 30px rgba(51, 146, 255, 0.18);
 
@@ -47,6 +73,39 @@
   --bg: var(--base); /* role alias -> --base, this pack's field */
 }
 
+/* Degrade to calm — MOTION_DOCTRINE.md §9. The doctrine's hardest line lands on
+   this pack, because this is the pack that mandates a WebGL particle field, a
+   scan/dim attention spotlight and scrubbed SVG instruments, and §9 says WebGL
+   and canvas scenes degrade to their CSS or SVG still. One of those three can
+   be reached from here and two cannot, and saying which at the declaration is
+   the point:
+
+     - The four durations and the stagger collapse. Every transition and every
+       reveal that spends them lands on its final state without travelling —
+       that includes the kit's only transition, .ic-btn's colour swap.
+     - The particle field is a rAF loop, and no custom property can stop one.
+       The scene branches on the reduced-motion hook and renders a single static
+       frame (SHELEG_DESIGN.md:364); `## Motion flavor` in instrument-console.md
+       now states the still for each motif, which is where an implementer of the
+       backdrop is actually reading.
+     - The scan/dim spotlight and the instrument scrub are driven by the scroll
+       clock, not by these tokens: under reduce every section renders lit and
+       every instrument renders drawn. The progress rail was already static.
+
+   `--motion-ease` is left alone: it is an ease-out, not an overshoot, and a
+   curve spent over 0s is inert. That is pigeonhole's lesson applied rather than
+   relearned — a duration collapses what a duration drives, and the rest is
+   stated where the implementer reads it. */
+@media (prefers-reduced-motion: reduce) {
+  :root {
+    --dur-fast: 0s;
+    --dur-base: 0s;
+    --dur-slow: 0s;
+    --dur-epic: 0s;
+    --stagger: 0s;
+  }
+}
+
 /* ── components ── */
 
 /* Everything below consumes var(--…) only. Colour never appears as a literal.

package/kits/maquette/src/styles.css

@@ -53,7 +53,12 @@
   --model-top: #fff9f4; /* the lit top face — 17.49:1 */
   --model-face: #fff3d6; /* the front face, warmer — 16.56:1 */
   --model-side: #e8ddc8; /* the side in shade */
-  --model-line: #151515; /* the model's own outline: the field colour */
+  /* @role drawn-on: --model-top, --model-face, --model-side — the model's
+     own outline, and it is the SAME hex as --bg on purpose: the cream solids are
+     separated by the table showing between them, which is why the model reads as
+     cut paper rather than as an SVG. It is drawn on the model's own faces and
+     never on the field, where it would be 1.00:1. */
+  --model-line: #151515;
   --leader: #8a8f98; /* the dotted leader from a block to its label */
 
   /* ---- Elevation: ONE shadow, and its direction is not adjustable ------- */

package/kits/notation/src/styles.css

@@ -122,6 +122,13 @@
    data-theme="dark" from a prefers-color-scheme listener — keep one switch,
    never two sources of truth for the theme. */
 
+/* The base layer. `--w-semi` and `--w-bold` are both 500 because this pack has no
+   bold — emphasis is the serif, the mono or a rule — so `<strong>` resolves to that
+   500 rather than to the UA's 700, which is the one weight the system does not
+   contain. Without this rule the ban lived only in the Bans list, where no browser
+   reads it. */
+strong, b { font-weight: var(--w-bold); }
+
 @media (prefers-reduced-motion: reduce) {
   :root {
     --dur-state: 0s;

package/kits/prism/src/styles.css

@@ -125,6 +125,12 @@
 
   /* ---- Motion ----------------------------------------------------------- */
   --ease: cubic-bezier(0.4, 0, 0.2, 1);
+  /* Three durations, and the press has its own because the doctrine gives press
+     feedback a band of its own: 100-160 ms. Until 1.45.0 this layer held two, the
+     pack prescribed --dur-fast for the press, and 200 ms is 40 ms past the top of
+     that band with nothing faster in the layer to reach for. --dur-fast stays at
+     the reference's 200 ms for hover and colour, which is inside the UI ceiling. */
+  --dur-press: 0.14s;
   --dur-fast: 0.2s;
   --dur-base: 0.3s;
 

package/kits/scoreboard/src/styles.css

@@ -50,7 +50,11 @@
      marks": a 3x18px tick, a chip edge, a rule. --accent-ink is the one that may carry a
      word, and the 1.13.1 correction says no orange here may carry a link. */
   --accent: #ff4801;
-  --accent-hover: #e03d00; /* 4.12:1 on --bg — the one orange that may carry a link */
+  /* 4.12:1 on --bg — the hover step of a MARK, and it may not carry a link either.
+     1.13.0 called this "the one orange that may carry a link"; 1.13.1 retracted that
+     in the pack's Gotchas and left the sentence here, where a reader copies it. A
+     link is --ink with an --accent underline. */
+  --accent-hover: #e03d00;
   --accent-wash: #fff7ed; /* MEASURED — selected row, hovered cell */
   /* The accent's one filled use: a selected chip. 4.92:1 under --on-accent,
      which clears AA; white on the same fill is 3.40:1 and does not. The accent
@@ -105,11 +109,16 @@
   --chart-grid: #e4e4e7;
   --chart-axis: #71717b;
 
-  /* ---- Elevation: two soft stacks and a hairline ------------------------ */
-  /* MEASURED. Both stacks are two layers at 4–5% and neither is tinted. This
-     pack lifts almost nothing: a card is a 1px --line-weak rule with the
-     hairline under it, and --shadow-card is for the one report surface a
-     section is built around. */
+  /* ---- Elevation: a hairline, a panel wash, one stack and one lift ------ */
+  /* MEASURED. Four shadows, and the pack's Texture section names all four:
+     --shadow-hairline (1 layer, 2.5%) sits under every card; --shadow-panel
+     (1 layer, 4%) is the [data-surface="panel"] band's own seat; --shadow-card
+     (2 layers, 5% and 5%, and it IS tinted — rgba(16,24,40) is a blue-black, on
+     purpose: the report surface is the one place the page reads as screen rather
+     than paper) is for the one report surface a section is built around; and
+     --shadow-lift (1 layer, 16%, untinted) is for something genuinely floating.
+     This pack lifts almost nothing: a card is a 1px --line-weak rule with the
+     hairline under it. */
   --shadow-hairline: 0 1px 2px 0 rgba(0, 0, 0, 0.025);
   --shadow-panel: 0 1px 6px rgba(0, 0, 0, 0.04);
   --shadow-card: