svelte-kit-vice/CLAUDE.md

# CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

> **Branch note (`active-uix`):** the legacy layers `src/lib/`, `src/uix/terra/`,
> `src/uix/air/` and the old `src/routes/` test tree were removed during the
> UIX refactor. The new ecosystem lives in `src/arts/`, `src/libs/`, `src/svrs/`,
> and `src/uix/{active-uix,soma,sema,eidos,morfo}`. Demos are being re-authored
> under `web/routes/` from scratch.

## Build / Test Commands

```bash
npm run dev          # Start dev server (routes resolve from web/routes)
npm run build        # Production build (static via adapter-static)
npm run test         # Run all tests once
npm run test:unit    # Run tests in watch mode
npx vitest run src/uix/morfo/compile.test.ts   # Run a single test file
npx vitest run -t "describe name"              # Run tests matching a pattern
npm run check        # Type check with svelte-check
npm run lint         # Check formatting (Prettier)
npm run format       # Auto-format
```

## Code Style

- Tabs for indentation, single quotes, no trailing commas, 100 char print width (Prettier).
- Comments in **English**. (Some legacy Spanish comments remain — don't add new ones.)
- Svelte 5 runes mode is enforced globally (`svelte.config.js` `dynamicCompileOptions`).
- DOM event handlers: lowercase (`onclick`, `onfocus`). User callbacks: camelCase (`onComplete`, `onInteractOutside`).
- State: `$state()` without `#`. Derived: `readonly x = $derived.by(...)`.

## Path Aliases

Aliases are kept in sync between `svelte.config.js` and `vite.config.ts`. Source
of truth is `vite.config.ts` (single `aliases` const reused by both top-level
resolve and the server-test project).

| Alias | Target |
| --- | --- |
| `@` | `src/` |
| `$active-app`, `$adom`, `$auth`, `$bus`, `$cache`, `$clipboard`, `$color`, `$connection`, `$format`, `$http`, `$langs`, `$logger`, `$motion`, `$orca`, `$perm`, `$prefs`, `$session`, `$sium`, `$storage`, `$timer` | `src/arts/{name}` |
| `$libs`, `$locale`, `$reactive` | `src/libs`, `src/libs/locale`, `src/libs/reactive` |
| `$svrs` | `src/svrs/` |
| `$uix`, `$active-uix`, `$soma` | `src/uix`, `src/uix/active-uix`, `src/uix/soma` |

## Vitest Two-Project Structure

`vite.config.ts` defines two test projects:

- **client**: Browser tests via Playwright for `*.svelte.{test,spec}.{js,ts}`.
- **server**: Node environment for `*.{test,spec}.{js,ts}` (excludes svelte tests).

## Architecture

The UIX is layered around a declarative contract (morfo) that the other layers
consume. Each consumer reads the morfo via `compileMorfo(morfo)`, never by
walking the raw declaration.

```
arts/   → runtime artifacts: active-app, adom, motion, perm, prefs, ...
libs/   → pure helpers (zero-dep): reactive, days, dom, locale, ...
svrs/   → server-authoritative engines

uix/morfo      → declarative contract: parts, data-attrs, ARIA, keyboard,
                 events with semantic family/intent. Compiler at compile.ts
                 emits CompiledMorfo (cached by WeakMap).
uix/sema       → semantic engine + channels (visual / sound / vibra).
                 The visual channel writes data-event* during a hold window.
                 State attrs (data-state, data-disabled, ...) are NEVER
                 touched by sema — they belong to the morfo runtime.
uix/soma       → headless component layer. Each provider consumes morfo +
                 sema via `runtime.trigger(eventName)` which sequences:
                 prewrite → semantic.emit → handler → effect-driven state attrs.
uix/eidos      → CSS layer that selects against the data-attrs the morfo
                 contract promises. lint.ts validates a CSS file against
                 compileMorfo(morfo).contracts.cssSelectors.
uix/active-uix → composition root. Two boot modes:
                 - createActiveUix(options)        — standalone (owns services)
                 - attachActiveUix(activeApp)      — attach to external app
```

**Motion** lives in `arts/motion` (`EngineMotion`, alias `$motion`), exposed as
`uix.motion` and consumed by BOTH soma (`Presence.motion` → `motion.run`) and
eidos (CSS generation + preset registration). It's an art — not part of eidos —
so soma can drive JS motion (spring) without a soma→eidos dependency. The DOM
arrives injected via a structural `MotionDom` port (`adom` satisfies it).
Two-moment model + F1–F7: `src/arts/motion/README.md` +
`src/uix/eidos/eidos-motion.md`.

### morfo (`src/uix/morfo/`)

Single source of truth for a component. A `Morfo` declares:
- `parts[]` — kebab + archetype + `data` + `aria` + optional `keyboard`
- `events[]` — name + `target` partRef + `semantic` + optional `prewrite` + `commits`
- `focus` — initial / trap / return / restore (for overlay components)

`compileMorfo(morfo)` returns a `CompiledMorfo` with pre-resolved `AttrPlan[]`,
`KeyboardPlan[]`, `ActionPlan[]`, and `contracts.cssSelectors` (the closed set
of selectors eidos is allowed to use). Cached by morfo identity.

### eidos linter

`scripts/eidos-lint.ts <component>` and `scripts/eidos-lint-all.ts` classify every
`[data-*]` selector in eidos CSS as morfo-backed / eidos-only / invalid. Use
this to detect drift between the morfo contract and the eidos rules.

## Key Conventions

- **Never modify the morfo declaration files** without updating the eidos
  selectors and provider sources that depend on them.
- **Read before acting.** When told to read a file, read it. Don't paraphrase.
- **Verify before reporting done.** Run `npm run check` and the relevant
  vitest scope. UI claims need a browser check; if you can't run a browser, say so.
- **No backward-compat shims** when relocating code. Update consumers and delete
  the old path; don't leave a re-export.
- **No re-export façades** between layers (e.g. soma must not re-export dias —
  consume `$libs/days` directly).
- **Provider naming**: child providers reference the parent as `provider`, never
  `root`. The exported component is `<Xxx.Provider>`. `createAttrs` still uses
  the part kebab `'provider'` internally.
- **Data-attr naming**: `data-{component}` (provider) and `data-{component}-{kebab}`
  (sub-parts). Never `data-soma-*`. The compiler emits these via
  `compiled.parts.attrs`; runtime queries match exactly.
- **CSS token naming (eidos recipes)**: bare-prefixed by component, never
  by layer. Public tokens: `--{component}-…` (e.g. `--dialog-content-bg`,
  `--toggle-radius-md`). Internal recipe tokens: `--_{component}-…`.
  Never `--eidos-…`, `--air-…`, `--terra-…`, `--soma-…`. Migrating
  `air-` baselines means stripping the `air-` prefix, not replacing it.
- **DOM helpers**: `$adom` is the single public surface for DOM-related
  functionality at component / soma level. It re-exports all of
  `$libs/dom` wholesale, plus the reactive `createActiveDom` runtime
  and stateful primitives (`BodyScrollLock`, `DOMContext`, `RovingFocusGroup`).
  Components and soma NEVER import from `$libs/dom` directly —
  consume `$adom`. Only `arts/adom/*` (the implementation), `arts/frontend`
  (peer service fallback), and tests of `$libs/dom` may reach into it.
  This includes responsive types (`ResponsiveProp<T>`, `Breakpoint`,
  `Breakpoints`) — they travel with the runtime, so they live in `$adom`.
- **Iframe / popup correctness**: when you have a `uix.dom`, prefer
  `uix.dom.getDocument(node?)` and `uix.dom.getWindow(node?)` over the
  bare functions of the same name. The instance methods respect the
  active-dom's `targetWindow`; the bare functions fall back to the
  global `document` / `window`, which is wrong in iframe / popup /
  happy-dom-test contexts.

## Critical Rules

- **NEVER delete files without explicit instruction** ("delete", "remove",
  "borra", "elimina"). If ambiguous, ASK first.
- **NEVER create fake translators / mocked services** in tests for components.
  Use real instances composed via `createActiveUix` or `attachActiveUix`.
- **NEVER skip hooks (`--no-verify`) or bypass signing** unless explicitly asked.
  If a hook fails, fix the root cause.
- **`$reactive` survives, `$lib` does not.** The `$reactive` alias points to the
  current `src/libs/reactive`; the singular `$lib` alias was removed in the
  cleanup phase and any code that resurfaces it is a regression.

## Reference Documents

Architecture overview and motivations: [`src/uix/active_architecture.md`](src/uix/active_architecture.md).

Per-layer references:
- [`src/uix/morfo/README.md`](src/uix/morfo/README.md)
- [`src/uix/sema/README.md`](src/uix/sema/README.md)
- [`src/uix/soma/SOMA_ARCHITECTURE.md`](src/uix/soma/SOMA_ARCHITECTURE.md) +
  [`src/uix/soma/COMPONENT_GUIDE.md`](src/uix/soma/COMPONENT_GUIDE.md)
- [`src/uix/eidos/README.md`](src/uix/eidos/README.md) +
  [`src/uix/eidos/components/README.md`](src/uix/eidos/components/README.md)

The canonical guide for sema vocabulary, morfo event shape, color
tokens, persistence and a11ySemantic lives in
[`src/docs/GUIA_IMPLEMENTACION_SEMAUIX.md`](src/docs/GUIA_IMPLEMENTACION_SEMAUIX.md).
That document is authoritative — when in doubt about families, intents,
verbs, color subsets per component, or holds, that's the source.

## Sema cascade selectors must use the typed builder

Cascade rules in `src/uix/sema/components/*.ts` MUST construct their
`selector` via `semaSelector(morfo, partKebab, matchers?)` from
`$uix/morfo`. **Never hand-write selector strings** that target morfo-
emitted attrs (`[data-{component}]`, `[data-{component}-{part}]`,
`[data-event="..."]`). The helper:

- Compile-error if `partKebab` doesn't exist on `morfo.parts`.
- Compile-error if `matchers.eventName` doesn't match `morfo.events`.
- Type-checks `eventFamily`/`eventIntent` against the canonical sema
  unions.
- Output is the same CSS string the cascade rule needs — zero runtime
  cost beyond concatenation.

Hand-written selectors in cascade rules are an architecture violation:
when morfo renames a part or event, the strings go stale silently. The
helper wires the cascade to the morfo's contract so renames break at
the type level.

```ts
// good
import { semaSelector } from '$uix/morfo'
import { dialogMorfo } from '$uix/morfo/components/dialog'

{
  selector: semaSelector(dialogMorfo, 'content', { eventFamily: 'commit' }),
  haptic: { kind: 'tap' }
}

// bad — string drifts silently if morfo renames
{
  selector: '[data-dialog-content][data-event-family="commit"]',
  haptic: { kind: 'tap' }
}
```

## Sema: intent policy per family

Per book canon there are **8 families**, encoded in `SEMA_FAMILY_POLICY` at
`src/uix/sema/types.ts`. Each family declares two independent axes:

```ts
{
  contact:  { intentRequirement: 'optional', intentGuidance: 'discouraged' },
  commit:   { intentRequirement: 'required', intentGuidance: 'expected' },
  signal:   { intentRequirement: 'required', intentGuidance: 'expected' },
  handle:   { intentRequirement: 'optional', intentGuidance: 'contextual' },
  emerge:   { intentRequirement: 'optional', intentGuidance: 'contextual' },
  shift:    { intentRequirement: 'optional', intentGuidance: 'contextual' },
  sustain:  { intentRequirement: 'optional', intentGuidance: 'contextual' },
  delegate: { intentRequirement: 'optional', intentGuidance: 'contextual' }
}
```

- **`intentRequirement`** (`'required' | 'optional' | 'forbidden'`) — compile-time gate. `required` → `intent` REQUIRED in `MorfoEventSemantic` (discriminated union). `forbidden` is reserved for future use (no family uses it today).
- **`intentGuidance`** (`'expected' | 'contextual' | 'discouraged'`) — doctrinal hint, not type-enforced. Drives lint warnings and editor tooltips.

Type derivation: `SemaEvent` and `MorfoEventSemantic` discriminate over `intentRequirement`. Editing the const reshapes the discriminated union.

Runtime enforcement: `validateSemaEvent` throws when an `intentRequirement: 'required'` family declares an event without intent. Optional families pass silently.

**`delegate`** is the 8th family per book cap. 29 — reparto de iniciativa entre usuario y sistema. Structural; carries no intent of its own. Active channels empty (delegate composes with sustain / signal / commit for perceptual layering, doesn't own a base signature).

The earlier flat `intentPolicy: 'allowed' | 'expected' | 'optional'` mixed type-requirement with doctrinal guidance. The split was applied in commit `00f0b740`.

## Sema: open channel registry + flat CSS-style cascade

The perceptual layer is **NOT closed**. The book defines **8 expression
channels** (time · motion · presence · depth · shape · color · sound · haptic);
the framework SPLITS them by owner — and this is the single canonical narrative:

- **Sema executes 2 runtime channels** — `sound` + `haptic`, the ONLY two
  declared in `SemaChannelSignatures` — plus projects the `visual` meta-channel
  (it stamps `data-event-*` during the hold). `motion`/`color`/`presence` were
  explicitly removed from sema (`sema-map.ts`); the engine knows no DOM/CSS.
- **Eidos materializes the visual channels** (motion / presence / depth / shape /
  color) by reacting in CSS to those `data-event-*` + the morfo's
  `data-state`/`data-intent`. Eidos is the sole visual owner.

Apps add more *sema* channels (a11y, voice, …) via TypeScript declaration
merging of `SemaChannelSignatures`.

**The semantic tokens are `data-event-*`**: the engine stamps `data-event`,
`data-event-family`, `data-event-intent`, `data-event-phase`, `data-event-id`
on `signal.target` BEFORE the cascade resolves. Cascade rules use CSS
selectors that target those tokens directly — same surface that eidos CSS
already reads. Engine owns stamp/unstamp; VisualChannel only owns the hold.

Cascade rules are **flat** (no `overrides: { eventLabel: ... }` middle
layer): a single selector + per-channel deltas, like a CSS rule.

```ts
{
  selector: '[data-dialog-content][data-event-intent="threat"]',
  sound:  { sampleUrl: '/sounds/dialog-fail.wav' },
  haptic: { kind: 'error', pattern: [50, 80, 50, 80, 50] }
}
```

Resolution cascade (each layer overrides the previous):

```
1. SEMA_MAP.families[F].base                  — per-channel signatures
2. SEMA_MAP.intents[I].deltas                 — valenced families only;
                                                numbers ADD (modificadores)
3. signal.overrides + signal.channels         — per-event morfo overrides;
                                                numbers REPLACE
4. engineOpts.overrides.runtime               — path-based globals; REPLACE
5a. engineOpts.components[].cascade           — per-component packs
                                                (sema/components/{name}.ts)
5b. engineOpts.overrides.cascade              — app-level rules; appended
                                                AFTER packs so they win
                                                on equal specificity
```

**Per-component packs live in `src/uix/sema/components/{name}.ts`** —
symmetric to morfo / soma / eidos. Each exports a `Sema` with
`cascade` rules + optional `preloadSamples` (WAVs decoded at boot).

**Override semantics — numbers**:
- intent.deltas (capa 2): numbers ADD (modificadores compositivos).
- overrides (capas 4, 5, 6): numbers REPLACE (`color: red`, not "add red").
- Use `{ op: 'add' }` / `{ op: 'multiply' }` / `{ op: 'replace' }` for
  explicit semantics either way.

**`intent.deltas` owns the perceptual signature of the intent** — the
sound primitives `pitch`, `gain`, `contour` (and the haptic `intensity`
when shaped per-intent) come from capa 2. **Cascade rules (capas 6a/6b)
MUST NOT override these primitives** — doing so flattens the perceptual
difference between `risk` / `threat` / `loss` (or `affirm` / `fulfill`).
Cascade rules add character (`haptic.kind`, `haptic.pattern`,
`sound.sampleUrl`, `motion.easing`, `presence.backdrop`) — not the
intent's evaluative profile. If a per-component cascade rule needs to
*shift* a numeric primitive (e.g. close-* in emerge goes descending),
use `{ op: 'add', value: -150 }` so it composes on top of the intent
rather than erasing it.

`HapticChannel` (V1, real — Vibration API) is opt-in via
`new EngineSemantic({ haptic: true })`. Sound and haptic both default
off because they have audible / tactile side effects.

Canonical sources: `src/uix/sema/sema-map.ts` (SEMA_MAP value +
SemaChannelSignatures registry + Sema), `src/uix/sema/resolver.ts`
(cascade implementation + CSS specificity), `src/uix/sema/stamp.ts`
(stamp/unstamp), `src/uix/sema/components/*.ts` (per-component packs),
`src/uix/sema/README.md`.

## Eidos drift defense — types over lint

Selector drift between morfo's emitted attrs and the layers that style /
react to them (eidos CSS, sema cascade) is defended **at compile time
via typed builders**, not at lint time.

- **Sema cascade rules** (`sema/components/*.ts`) consume
  `semaSelector(morfo, partKebab, matchers?)` from `$uix/morfo`. Renames
  in morfo break the cascade at type-check time. See "Sema cascade
  selectors must use the typed builder" above.
- **Eidos plain `.css` recipes** are still raw CSS today — there is no
  CSS-side typed builder. For those, `scripts/eidos-lint.ts` remains as
  an **opt-in safety net** that classifies each `[data-*]` selector as
  `morfo-backed` / `eidos-only` / `invalid`. It is not the architectural
  contract; the contract is the morfo declaration.

Rule: when a layer can consume the morfo via TypeScript (anything in
`.ts` / `.svelte`), it MUST use the typed builder. Lint is for the
remaining surface (plain CSS recipes) until those gain a builder of
their own. Hand-written morfo-targeting selector strings in TypeScript
files are an architecture violation, not a lint warning.

## Session hand-off — 2026-06-05 (sistema de color — cierres + theme builder runtime + wide-gamut + a11y)

Sprint dando "un giro de vueltas" al sistema de color de Eidos hasta reference-grade. 6 commits. El motor `uix.color` (`$color`, art puro isomórfico: OKLCH↔sRGB, APCA, alpha compositing-inverse, `deriveScheme`/`generateScale`/`temper`) ya existía; este sprint lo CONSUME end-to-end + cierra los ítems de calidad de `THEMING_AUDIT`.

**Cierres rápidos** (`cc37bdce`):
- `themes/base.ts`: `loss` era `purple` ≡ primary → mapeado a `plum` (escala canónica). Última colisión de roles del tema base cerrada (tras tertiary→indigo).
- `temas/grafito`: la sección "override por componente" pasaba nombres de escala (teal/amber/…) a `Button.color`, que solo acepta el override jerárquico (`primary|secondary|neutral`) — type error + inerte. Reescrita a los 2 ejes reales: `color` (jerarquía) + `intent` (paleta evaluativa). Cierra el último error de `svelte-check` → **0 errores**.

**API runtime — theme builder** (`c4d2e34d`):
- Nuevo `src/uix/eidos/lib/build-scheme.ts` (PURO): `buildScheme(seed, opts)` compone `deriveScheme` + `generateScale` + APCA on-solid + alpha en el mapa de override `--primitive-{role}-*` (+ `--color-{role}-contrast`). `seed → { variables, roles }`. Sin DOM.
- `ActiveEidos.applyColorScheme(seed, opts)` / `clearColorScheme()`: resuelve escalas-donantes + background del tema activo, escribe el bloque `uix-eidos-scheme` DESPUÉS del de tema (gana cascada), RE-DERIVA al cambiar de modo (sigue light/dark). Devuelve `BuildSchemeResult`. `opts`: `variant` (tonal|vibrant|monochrome) + `temper` (cohesión de intents, mantiene hue) + `overrides` per-rol + `selector`. Exportado de `$uix/eidos`. Demo `/temas/color` dogfooda `buildScheme`.

**Wide-gamut OKLCH** (`909ab7f9` output + `0bb03559` generador):
- Output OKLCH-nativo default-on (RFC §7 estrategia A): `render-css > appendColorScaleDeclarations` emite por cada paso de paleta el hex (fallback) + un hermano `oklch()` que gana donde se soporta. SIN flag.
- Generador wide-gamut-TRUE: `buildScheme`/`applyColorScheme` retienen el OKLCH raw de `generateScale` (sin clamp) → `result.wideGamut` + `result.roles[].stepsOklch`. Nuevo `schemeDeclarations(result, { fallback })` apila hex+oklch (default) u oklch-only (inline).
- **HONESTIDAD**: la paleta Radix shipped es hex sRGB → sus `oklch()` son sRGB-equivalentes (idéntico hoy). El wide-gamut REAL vive en el generador: seed con croma > sRGB sale más saturado en P3. Demo `/temas/color`: slider **vivacidad P3** + badge «fuera de sRGB → P3» (`isInSrgbGamut`). Verificado: ×1.70 → croma primary-9 0.18→0.31. La paleta autorada NO se migró a semillas (regresaría los valores exactos de Radix sin añadir wide-gamut visible).

**P3-2 forced-colors + P3-3 ramp de bordes** (`4eab3306`):
- **forced-colors (Windows HCM)**: el navegador auto-mapea bordes/texto/fondos (`forced-color-adjust: auto`) pero ELIMINA `box-shadow` → el focus ring (`--focus-ring`, box-shadow) desaparecía. Fix: foundation emite siempre `renderForcedColorsBlock` → `@media (forced-colors: active) { :focus-visible { outline: 2px solid Highlight } }`. Componentes con outline propio (Button) lo conservan por especificidad. Pendiente: `prefers-contrast: more`.
- **ramp de bordes**: slot de rol `border` 6 (separador sutil) → **7** (UI element border de Radix). `DEFAULT_COLOR_ROLE_SLOT_STEPS`. element/hover/active=3/4/5 se quedan. Verificado: `--color-{role}-border` → primitive-7 (checkbox OK).

**Verificación**: `check` 0 errores. Suite eidos 155/158 (3 fallos pre-existentes del track `words`, confirmados con `git stash` baseline). `build-scheme` 8 + scheme 4 + forced-colors 1. `generated/base.css` regenerado y en sync.

**Decisiones con efecto duradero**:
- **Capas**: `uix.color` = matemática · `eidos/lib/build-scheme` = composición (pura) · `ActiveEidos` = aplicación DOM. No mezclar.
- **Wide-gamut default-on, sin flag** (estrategia A); fallback hex → sRGB idéntico, no-disruptivo.
- **Paleta autorada = exacta-Radix-sRGB** (sin regresión). Wide-gamut VISIBLE de la paleta = Fase 3 futura (autorar/generar en OKLCH).
- **forced-colors**: el box-shadow muere en HCM → el foco debe ser `outline`. Migrar los componentes que aún usan `var(--focus-ring)` (box-shadow) a `outline` propio es trabajo futuro per-componente (hoy dependen del fallback global).
- **slot `border` = step 7** (no 6).

**Pendiente (solo higiene del engine, NO calidad de color)** — _todo resuelto-o-decidido en la continuación, abajo_: `THEMING_AUDIT` P3-4/5/6/7/8/9/11 + mitades P2.

Docs: `COLOR_ENGINE_RFC.md` §6.2/§7, `THEMING.md` §26/§27/§28 + TOC, `README.md` tabla, `THEMING_AUDIT_2026-06-01.md` P3-1/2/3 resueltos.

**Continuación (mismo día) — fix de UX + cierre del backlog**:

- **Checkbox lag — NO era color, era timing del sema** (`e6fd014d` + `cd834384`). El usuario reportó el check "muy lento". Medido frame-a-frame: el `data-state` tardaba **244ms** en cambiar tras el click. Causa: el provider del checkbox fija el estado en el HANDLER del `runtime.trigger`, y el morfo declaraba `sequence: 'pre'` → el runtime hace `await runEmit()` (que **espera el hold del canal visual ~240ms**, `engine.emit()` → `await visualChannel.handle`) ANTES del handler. Fix: `commit-toggle-check`/`commit-toggle-uncheck` → `sequence: 'post'` (handler primero, pulso después). 244ms → 46ms. Secundario: trazo del checkmark `stroke-duration` 220ms hardcoded → `var(--duration-fast)`. **Verificado que radio-group (47ms) y tabs (31ms) NO laggean** aunque son 'pre' — fijan estado en el call-site, no en el handler; no se tocaron. Toggle/Switch ya eran 'post'.
  - **DOCTRINA NUEVA**: un control cuyo estado se fija en el HANDLER del trigger DEBE usar `sequence: 'post'`; con `'pre'` el hold perceptual del emit bloquea el cambio funcional. Los que fijan estado en el call-site toleran 'pre' sin lag.

- **P3-11 surface ladder** (`76e18772`). Light `overlay` era `neutral-3` == `muted` (popovers indistinguibles de paneles muted en light); dark ya tenía overlay=4. Light overlay → `neutral-4` → ladder consistente en ambos modos: `default(1) < raised(2) < muted(3) < overlay(4)`. Verificado (popover light: overlay L93% ≠ muted L95.5%).

- **prefers-contrast: more** (`21b2329a`). Completa el a11y de color junto a forced-colors. `renderPrefersContrastBlock` → `@media (prefers-contrast: more) { :root:root { … } }` refuerza bordes (neutral 7/8/9) + texto de-enfatizado (12/11). `:root:root` (0,2,0) gana al `:root` del tema; aditivo, gated, estrictamente más fuerte. THEMING §28.

- **Cierre del backlog del engine** (`afa15aea`). `THEMING_AUDIT` P3 todo fixed-or-decided:
  - ✅ **P3-5** guard `appendScaledMetricDeclarations` (solo escala números finitos ≠0; keywords/`var()`/`calc()` verbatim — evita `calc(auto * …)`).
  - ✅ **P3-6** confirmado ya resuelto (dispose vía `dom`, sin `document` directo).
  - ✅ **P3-8** index ya NO re-exporta los render-fns crudos (API pública = clase `ActiveEidos`; `./lib/render-css` para uso interno). Consumidor de test redirigido al módulo.
  - ⏸️ **P3-4** deferido (densidad gana por orden de fuente determinista, estable; restructure `:where(:root)` = coste alto por nit teórico).
  - ⏸️ **P3-7** deferido (reactividad callback-driven vía `apply()` POR DISEÑO; runes = refactor riesgoso sin bug que lo justifique).
  - ⏸️ **P3-9** deferido (forwarders huérfanos `_accent` — cirugía de recipe con riesgo de cascada, valor bajo).
  - Único pendiente real: 2 mitades **P2** (poda de contrato + validación de identificador-color pelado) — edge-case, deferidas por bajo valor.

**Estado final del color/theming**: apariencia + a11y (forced-colors + prefers-contrast) + wide-gamut + theme builder runtime + backlog del engine — todo resuelto-o-decidido, cero ítems colgando. `check` 0 errores. Suite eidos 156/159 (3 fallos de words, pre-existentes, confirmados con baseline). Docs extra: `THEMING.md` §28 (prefers-contrast), `THEMING_AUDIT_2026-06-01.md` scorecard P3 completo.

## Session hand-off — 2026-05-27 (persistence + a11ySemantic + polymorphic)

Sprint cerrando los 3 ítems diferidos del hand-off 2026-05-08: persistence + holds-by-intent (libro §6), a11ySemantic per event (libro §9), polymorphic events (libro §5.3). Detalle completo en `src/docs/LIBRO_VARIACIONES_Y_EXTENSIONES.md` D.9 / D.10 / D.11.

**Persistence (libro §6.1)** — separa `hold` (duración perceptible mínima) de `persistence` (lifecycle real):

- Nuevo tipo `SignalPersistence = 'transient' | 'untilAction' | 'untilFix' | 'stateBound'` en `src/uix/sema/types.ts`, propagado a `SemanticSignal`, `MorfoEventSemantic` y `TriggerOptions`.
- Tabla canónica `SEMA_HOLDS_BY_INTENT` en `src/uix/sema/holds.ts` (referencia, NO auto-aplicada — default conservador `'transient'`).
- `EngineSemantic.emit()` ahora devuelve `Promise<string>` (el id). Para `persistence !== 'transient'` mantiene la proyección viva pasado el hold; expone `engine.clear(id)`, `engine.clearTarget(target)`, `engine.hasActive(id)`.
- `SomaRuntime.trigger()` devuelve `TriggerResult { id?, persistence? }`. Expone `runtime.clearSignal(id)`, `runtime.clearTarget(target)`, `runtime.partRef(part)` (helper que devuelve el HTMLElement registrado de un part — útil para que providers limpien sin trackear ids).
- 6 morfos declaran `persistence` explícita: `announce.signal-alert` (untilAction), `dialog/drawer.close-after-fail` (transient explícito), `file-upload.signal-warn-reject` + `form.signal-warn-invalid` (untilFix), `password-field.signal-notify-caps-state` (stateBound).
- 3 providers cabledados con `clearTarget(provider)`: form (limpia al validar / reset), file-upload (limpia antes de cada nueva ronda de accept/reject + en remove / clear), password-field (limpia indicator al apagarse caps lock).

**a11ySemantic (libro §9.1)**:

- Nuevo `MorfoA11ySemantic` interface en `src/uix/morfo/types.ts`: `requiresPersistentTrace?`, `requiresLiveRegion?`, `requiresFocusMove?`, `keyboardEquivalent?`, `reducedMotionFallback?: 'state' | 'text' | 'focus' | 'none'`. Añadido como `a11ySemantic?` opcional en `MorfoEvent`. Pasa por compile a `ActionPlan.a11ySemantic`.
- Helper de reduced-motion en `src/arts/adom/reduced-motion.svelte.ts` (paralelo a `viewport.svelte.ts`): `ReducedMotionTracker` reactivo del media query con SSR-safe fallback. Expuesto en `ActiveDom.prefersReducedMotion.matches`.
- `ActiveUix.announce(message, priority?, timeout?)`: live region compartida lazy-creada vía `dom.writeNode`. No depende de soma. Polite/assertive son regiones separadas, cleanup en `dispose()`.
- `Soma.runtime()` auto-wires `sources.announce = (m, p) => uix.announce(m, p)`. Tests usan fake `announce`.
- `SomaRuntime.trigger()` honra a11ySemantic después del emit: live region (con `opts.message`), focus move, reduced-motion fallback (incluye forzar `channels: []` cuando fallback = 'state').
- 6 morfos anotados (los mismos consumidores).

**Polymorphic events (libro §5.3)** — ADITIVO sobre el shape concreto:

- Morfo declara `family` + `intent` + `verb` como default. Si añade `allowedFamilies: readonly SemaFamily[]`, los providers pueden override la family en `runtime.trigger(name, { semantic: { family, intent?, verb? } })`. El default family del morfo es IMPLÍCITAMENTE allowed.
- `isPolymorphicSemantic(semantic)` helper exportado de `$uix/morfo`.
- `SomaRuntimePolymorphicError` cuando override no está en allowedFamilies.
- Diseño aditivo (no variante separada con `defaultSemantic`) elegido para mantener backwards-compat con 29 morfos existentes y demos que acceden `event.semantic.family` directamente.

**Tests**: 671/671 pasan en `src/uix/sema + morfo + soma + src/arts/adom/test`. 4 fallas pre-existentes en `src/uix/contracts.test.ts` (words component translation keys) — confirmadas en baseline antes de mis cambios via `git stash`.

**`npm run check`**: 1 error pre-existente en `src/uix/soma/components/command/command-provider.svelte.ts:384` (`Cannot find name 'attrs'`) que ya estaba en main.

**Pendientes deliberadamente fuera del sprint** (resueltos en la segunda mitad — ver hand-off 2026-05-27 #2 abajo):

- ~~Extender persistence a tags-input + textarea~~ — DONE
- ~~Caller messages para announce~~ — DONE (form / file-upload / password-field / dialog / drawer)
- ~~Demo polymorphic real~~ — DONE (Dialog refactorizado a polymorphic close)

## Session hand-off — 2026-05-27 #2 (messages + persistence ext + Dialog polymorphic)

Segunda mitad del sprint cerrando el backlog del hand-off anterior. 3 grupos de cambios.

**M — Caller messages para live region**:

Cada provider que dispara un evento con `a11ySemantic.requiresLiveRegion` ahora pasa `opts.message` con texto localizado. La live region (ya cabledada via `Soma.runtime() → sources.announce`) finalmente anuncia algo:

- `form-provider`: `signal-warn-invalid` → "1 form error..." / "{N} form errors..." (count = `Object.keys(form.issues).length`). Usa `FORM_LANGS.ERROR_SUMMARY_*`.
- `file-upload-provider`: `signal-warn-reject` → "1 file was rejected" / "{N} files were rejected". Nuevas entradas `FILE_UPLOAD_LANGS.REJECT_SUMMARY_SINGLE/MULTI` + langs catalog.
- `password-field-provider`: `signal-notify-caps-state` → "Caps Lock is on" (usa `PASSWORD_FIELD_LANGS.CAPS_WARNING` que ya existía).
- `dialog-provider.dismissWith(action, { message })`: signatura aceptando `message` opcional, forwarded al trigger.
- `drawer-provider.dismissWith(action, { message })`: igual que dialog.

**P — Persistence extendido a tags-input + textarea**:

Misma doctrina §6.2 — signal+risk = untilFix:

- `tags-input.signal-warn-reject` → persistence: `untilFix` + a11ySemantic + provider clear via `clearTarget(input)`. Nueva entrada `tags-input.reject-warning` en langs catalog. Provider gana método privado `emitWarnReject(target)` que centraliza clear + emit + message. Successful add también clearea (la adición VÁLIDA es el "fix" de un reject previo).
- `textarea.signal-warn-count-overflow` → persistence: `untilFix` + a11ySemantic + provider clear en transición OUT-of-overflow. Nueva entrada `textarea.overflow-warning` en langs catalog. Provider emite SOLO en transición INTO overflow (no en cada keystroke al cap) — el `untilFix` proyecta persistente, no necesita re-anuncio.

**POLY — Dialog refactor a polymorphic close**:

Caso real de uso del feature de eventos polymorphic (§5.3). 5 eventos `close-*` colapsados en 1 evento `close`:

- **Morfo** (`dialog.ts`): un evento `close` con `family: 'emerge', verb: 'close', allowedFamilies: ['emerge', 'commit', 'signal']`. SIN `prewrite` (el provider escribe `data-last-action` imperativamente).
- **Validador relajado**: el invariante "cada `values[]` debe ser prewritten por algún event" cayó al sentido único "cada prewrite con value debe estar en `values[]`". Razón: con polymorphism los values pueden ser escritos imperativamente. El comentario en `schema.ts` explica.
- **Provider** (`dialog-provider.svelte.ts`): `DISMISS_CAUSES` mapa de acciones a `{ lastAction, semantic }`. `dismissWith(action, opts)` traduce a `dom.apply` + `runtime.trigger('close', { semantic, ... })`. `triggerClose` es privado ahora.
- **Cascade sema** (`sema/components/dialog.ts`): selectores `eventName: 'close-dismiss-outside'` → `eventName: 'close', state: { attr: 'data-last-action', value: 'dismissed-outside' }`. `eventNamePrefix: 'close-'` → `eventName: 'close', eventFamily: 'emerge'`.
- **Eidos CSS** (`dialog.css`): NO cambió — ya leía `data-last-action` para tintar, no nombres de evento.
- **API pública preservada**: consumidores externos no notan diferencia (`dismissWith` mantiene firma + comportamiento observable).
- **Drawer / Popover**: NO refactorizados. Mismo patrón aplicable; defer porque cada uno requiere reescribir su cascade + decisión separada de timing.

**Tests**:

- `npx vitest run src/uix/sema src/uix/morfo src/uix/soma src/arts/adom/test`: **679/679 pass**.
- `npm run check`: 1 error pre-existente (`command-provider.svelte.ts:384` — ya en baseline).
- 4 fallas pre-existentes en `src/uix/contracts.test.ts` (words translations) — no relacionadas.

**Pendientes a futuro** (cubierto en hand-off 2026-05-27 #3 abajo):

- ~~Refactor analógico de Drawer + Popover a polymorphic close~~ — DONE
- Otros `signal.warn-*` no incluidos: hay variantes con `verb: 'warn'` que podrían también beneficiarse de persistence untilFix (revisar caso por caso).

## Session hand-off — 2026-05-27 #3 (Drawer + Popover polymorphic close)

Tercera tanda del sprint — completa el rollout del patrón polymorphic close (book §5.3) a los dos componentes hermanos de Dialog. Mismo refactor aplicado de forma sistemática.

**Drawer**:
- Morfo: 5 close-* → 1 polymorphic `close` con `family: 'emerge', allowedFamilies: ['emerge', 'commit', 'signal']`, sin prewrite. Drag events (drag-start / drag-progress / drag-end / resize) intactos.
- Provider: `DISMISS_CAUSES` map + `dismissWith(action, opts?)` traduce a `dom.apply(data-last-action)` + `runtime.trigger('close', { semantic })`. `triggerClose` privado. 4 callsites internos migrados a `dismissWith` (Content escape, Content keydown escape, Overlay click-outside, Close button).
- Sema cascade (`sema/components/drawer.ts`): selector `eventNamePrefix: 'close-'` → `eventName: 'close', eventFamily: 'emerge'`.
- Test (`drawer-provider.svelte.test.ts`): `triggerClose('close-dismiss', ...)` → `dismissWith('dismiss')`; expect name='close'.

**Popover**:
- Morfo: igual refactor. 5 close-* → 1 polymorphic `close`.
- Provider: igual patrón con DISMISS_CAUSES + dismissWith. Internal callsites (5 lugares: scheduleHoverClose, trigger toggle click, trigger toggle keydown, escape, outside) migrados a `dismissWith`. Close button: usa el prop `action` traducido a dismissWith.
- Sema cascade (`sema/components/popover.ts`): mismo patrón — `eventName: 'close'` + `state: { attr: 'data-last-action', value: 'dismissed-outside' }` para la regla de dismiss-outside passive.
- Test (`popover-provider.svelte.test.ts`): no había refs a close-* directos. Ningún cambio necesario.

**Test fixtures migrados antes del refactor**:
- `compile.test.ts`: prewrite test usaba `drawerMorfo` → ahora usa `colorPickerMorfo`.
- `runtime.svelte.test.ts`: prewrite test usaba `drawerMorfo` → ahora usa `colorPickerMorfo`.

**Resultado**:
- 3 componentes overlay (Dialog / Drawer / Popover) usan el patrón polymorphic close de forma consistente.
- 5 picker components (color-picker / date-picker / date-range-picker / time-picker / time-range-picker) mantienen el shape per-event. Son la canonical reference para `prewrite` declarativo + sirven como fixtures de tests.
- API pública intacta — `dismissWith` mantiene signatura en los tres overlays.
- Eidos CSS no tocó (ya leía `data-last-action`).

**Tests**: 680/680 pass en `src/uix/sema + morfo + soma + src/arts/adom/test`. `npm run check`: 1 error pre-existente.

**Pendientes a futuro** (picker family cubierto en hand-off 2026-05-27 #4):
- ~~Picker family refactor~~ — DONE
- Otros `signal-warn-*` no cubiertos (textarea / tags-input ya hechos; revisar caso por caso si emergen más).

## Session hand-off — 2026-05-27 #4 (Picker family polymorphic close)

Cuarta tanda — completa el rollout polymorphic close al picker family (5 componentes). Más decoupling de tests.

**Refactor de los 5 pickers**:
- `color-picker`: 4 close-* → 1 polymorphic `close`
- `date-picker`: 4 close-* → 1 polymorphic `close`
- `date-range-picker`: 4 close-* (incl. `close-range-commit`) → 1 polymorphic `close`
- `time-picker`: 4 close-* → 1 polymorphic `close`
- `time-range-picker`: 4 close-* → 1 polymorphic `close`

Cada morfo: `family: 'emerge', verb: 'close', target: v.partRef('calendar'|'clock'|'content'), persistence: 'transient', allowedFamilies: ['emerge', 'commit', 'signal']`.

**Hallazgo importante**: los picker providers NO disparan los close events. Solo togglean `opts.open = false`. Los eventos estaban declarados pero **inertes** — su único consumidor era el schema validator y los compiler tests. El refactor es alineación doctrinal, no de comportamiento.

**Sema cascade**: solo `color-picker` tiene un sema pack y no referenciaba close-* (solo handle-*). Cero updates necesarios en cascades.

**Test fixtures decoupling**:
- Nuevo `src/uix/morfo/test-fixtures.ts` con `prewriteFixtureMorfo` sintético.
- `compile.test.ts`: migrado de `colorPickerMorfo` → `prewriteFixtureMorfo`.
- `runtime.svelte.test.ts`: igual migración.
- Beneficio: los tests validan el contrato del compiler/runtime sin depender del catálogo de componentes. Futuros refactors del catálogo no rompen estos tests.

**Estado final del rollout polymorphic** (3 sprints combinados):

| Componente | Polymorphic morfo | Provider cabledado |
|---|---|---|
| Dialog | ✓ | ✓ (`dismissWith` → `runtime.trigger`) |
| Drawer | ✓ | ✓ |
| Popover | ✓ | ✓ |
| color-picker | ✓ | ✗ (toggle opts.open) |
| date-picker | ✓ | ✗ |
| date-range-picker | ✓ | ✗ |
| time-picker | ✓ | ✗ |
| time-range-picker | ✓ | ✗ |

Único morfo con shape pre-polymorphic restante: el fixture sintético `prewriteFixtureMorfo` — vivo sólo para tests.

**Tests**: 680/680 pass. `npm run check`: 1 error pre-existente.

**Pendiente a futuro**:
- Cabledar `runtime.trigger('close', { semantic })` en los providers picker cuando justifique disparar el evento (telemetría, sound on commit picker, etc.).

## Session hand-off — 2026-05-27 #5 (TSC v2.2 + cobertura universal de theming)

Cierra la migración universal del Token Scope Contract. El usuario rechazó el framing "excepciones arquitectónicas" para los 3 componentes que vivían fuera de TSC v2.1 (`select`, `avatar`, `toggle-group`) — la respuesta fue extender TSC con dos features nuevas (`parts` + `composition`) y migrar los 3.

**Extensiones TSC v2.2** (`src/uix/eidos/lib/config-types.ts` + `lib/render-css.ts` + `lib/config.ts`):

- **Multi-part scope** — `RecipeTokenMultiDeclaration.parts?: readonly string[]`. Cuando un token tiene `parts: ['x', 'y']` el generador emite selectores comma-separados (`[data-{c}-x][...], [data-{c}-y][...]`) en lugar del default `[data-{c}][...]`. Cubre el caso del select donde `data-color` vive en Trigger + Content (Content portaliza fuera del árbol del Trigger). Único consumer hoy.
- **Cross-recipe composition** — nuevo sibling field `composition?: { foreignComponent: { targetSelector, tokens } }`. Permite que un recipe declare overrides de tokens de OTRO recipe scoped a su propia cascade, con un selector de descendant para alcanzar la part foránea. Genera `{host-scope} {targetSelector} { --{foreign}-{tokenName}: ... }`. Validator rechaza root-scoped composition (un override no-scoped pertenece al foreign recipe). Único consumer hoy: `toggle-group` modifica `--toggle-palette-*` en sus items.

Validador (`config.ts`) extendido con `validateRecipeComposition`. Generator (`render-css.ts`) gana `stripCompositionKey` + `emitComposition`. Contract builder (`contract.ts`) skip-list de la reserved key `composition` para que no aparezca como `--{c}-composition` knob en el contrato público. Test helpers (`tokenKeys`/`tokenEntries` en `recipe-css-contract.test.ts`) filtran `composition` en todos los iteradores.

**Migraciones de los 3 componentes restantes**:
- `select`: 4 → 3 private `_accent-*` con `parts: ['trigger', 'content']`. (`_accent-solid` no se migró porque la CSS nunca lo consumía — era orphan; se removió de la recipe.)
- `avatar`: 6 tokens (`_bg`/`_fg`/`_border` + `_badge-bg`/`_badge-fg`/`_badge-border`) con declarations[] generadas por un helper inline `matrix()` que recorre los 8 colores × 3 variants. `_badge-*` usa `parts: ['badge']` para retargetear a `[data-avatar-badge]`. Reemplaza ~150 declarations CSS por las mismas declarations, generadas desde TS.
- `toggle-group`: bloque `composition: { toggle: { targetSelector: '[data-toggle-group-item]', tokens: { 'palette-*': {...} } } }` con 8 palette tokens × 4 colors. Reemplaza los 4 bloques CSS per-color.

**Bug post-migración del toggle-group (toggle-group color cascade)** — encontrado y arreglado:

La composition correctamente override `--toggle-palette-*` en `[data-toggle-group-item]`, pero la cascade se rompía aguas abajo. Después de la migración TSC v2 original de toggle, los tokens DERIVADOS (`--toggle-solid-on-bg`, `--toggle-outline-fg`, etc.) viven en scope `[data-toggle]`. El `[data-toggle-group-item]` es SIBLING (no descendant) de `[data-toggle]`, así que `var(--toggle-solid-on-bg)` resolvía a undefined en el item.

Fix (en `toggle-group.css`): **inlined las derivation expressions** que leen palette directamente en `[data-toggle-group-item]` y sus variant cascades. Las expresiones reflejan las de `recipes/base.ts > toggle.{solid,outline,ghost}-*`. Duplicación documentada en el header. Alternativa estructural (hacer el item carrier de `[data-toggle]` + propagar `data-color` morfo/soma) no se aplicó — change too big.

**Documentación**:
- `THEMING.md` §18 reescrito como "Cobertura universal de TSC" (sin excepciones). §7 extendido con subsecciones "Multi-part scope" y "Cross-recipe composition" con ejemplos completos. TOC actualizado.
- `eidos/README.md` tabla de referencia ampliada con entradas para TSC v2.2 + §18.

**Tests**: 786/786 pass en `src/uix/{eidos,morfo,soma,sema}`. `npm run check`: los mismos 6 errores pre-existentes en `lib/_demo`/`soma/components/internal`/`web/routes/active` (no relacionados).

**Pendiente a futuro**:
- Si emergen patrones similares al toggle-group (otros wrappers compositivos como button-group, nav-menu), la composition TSC v2.2 los cubre — no requiere más extensiones.
- Los `inlined derivation expressions` en `toggle-group.css` son la única duplicación entre recipes/base.ts y CSS. Si Toggle's derivations cambian, hay que actualizar ambos.

## Session hand-off — 2026-05-27 #6 (variants canon — `EIDOS_VARIANTS` + lint)

Pregunta arquitectónica del usuario: "si activeUIX quiere ser referencia como framework, ¿qué es lo lógicamente coherente respecto a la extensibilidad de variants?". Respuesta firme: **variants son canon del eidos, NO del theme** — paralelo a las 8 sema families del libro.

**Cambios**:

- `src/uix/eidos/lib/types.ts`: nueva constante `EIDOS_VARIANTS` (5 archetypes: `control`/`selection`/`chip`/`marker`/`tabs`) como single source of truth. Los 5 union types se derivan via `[number]` indexed access — valor y tipo no pueden desincronizarse. Nueva constante `EIDOS_VARIANT_VALUES` (Set flat de todos los valores canónicos + utilidades cross-component como `'plain'` y `'subtle'`).
- `src/uix/eidos/recipe-css-contract.test.ts`: nuevo test "variant CSS selectors per component match the declared type union". Por cada componente: extrae el union type de `components/{c}/types.ts` via regex (literal-union + archetype-alias patterns soportados, Extract<>/conditional types caen a advisory mode); compara con los `[data-{c}][data-variant='X']` selectores en `{c}.css`; reporta typos y unauthorized extensions bidireccionalmente. 101/101 tests pasan.
- `src/uix/eidos/THEMING.md` §19: nueva sección "Variants son canon del eidos, NO del theme" con argumentación (portabilidad, type safety, archetypes perceptuales), tabla de las 3 capas de la cebolla (sema → variants → palette), referencia a `EIDOS_VARIANTS`, comparación con Radix Themes/Mantine/Chakra v3/Ark UI/shadcn. TOC actualizado.
- `src/uix/eidos/README.md`: tabla de referencia ampliada con §19.

**Doctrina sostenida**: Theme = retintar lo perceptualmente fijo. Cambia QUÉ color es `affirm`, no QUÉ significa `outline`. Si una app necesita un look brandeado, hace override de tokens en `EidosConfig.recipes` o crea un wrapper composicional — NO inventa un nuevo variant.

**Variants component-specific permitidos** (Banner `inline`/`overlay`/`persistent`, Spinner `bars`/`dots`/`ring`, Button `'plain'`): viven en cada `components/{c}/types.ts` y el lint los valida contra la CSS del componente.

**Tests**: 101/101 pass en `src/uix/eidos`. `npm run check`: los mismos 6 errores pre-existentes (lib/_demo, soma/components/internal, web/routes/active).

## Session hand-off — 2026-05-28 (toggle-group structural identity)

Cierra la deuda dejada explícitamente abierta en el hand-off 2026-05-27 #5 ("Los `inlined derivation expressions` en `toggle-group.css` son la única duplicación entre recipes/base.ts y CSS. Si Toggle's derivations cambian, hay que actualizar ambos."). Ahora son cero.

**Pregunta arquitectónica del usuario** tras verificar el fix EXT-FIX (color cascade): "¿qué es lo recomendado en referencia al ecosistema y su diseño y arquitectura?". Respuesta firme: **resolver la duplicación estructuralmente, no con un test de drift**. El item de toggle-group ES un toggle (mismo press, mismo variant/color/size, misma máquina on/off) — la doctrina "morfo declara DNA" pide declararlo.

**Cambios**:

- `src/uix/morfo/components/toggle-group.ts`: el part `Item` declara `{ attr: 'data-toggle', value: v.literal(''), severity: 'required' }`. Cada item proyecta `data-toggle=""` como atributo de presencia. Cero overhead, captura la identidad estructural.
- `src/uix/eidos/components/toggle-group/context.ts` (nuevo): contexto Svelte tipado (`ToggleGroupEidosCtx` con getters reactivos para `variant` y `size`) — propaga las dos perillas eidos-only del root a cada item.
- `src/uix/eidos/components/toggle-group/toggle-group.svelte`: setea el contexto en el root. Getters mantienen reactividad cuando el prop cambia.
- `src/uix/eidos/components/toggle-group/toggle-group-item.svelte`: lee el contexto y escribe `data-variant={ctx?.variant}` + `data-size={ctx?.size}` en el button. Combinado con `data-toggle` del morfo, el elemento es DOM-equivalente a un `<Toggle>` standalone.
- `src/uix/eidos/components/toggle-group/toggle-group.css`: borradas ~150 líneas (todas las derivaciones base + variant cascade + size cascade + focus-visible / disabled / icon-only duplicados). El CSS conserva SOLO grouping concerns (flex layout, orientation, attached con first/last/border-radius, block, focus z-index, group-level disabled). El header comment se rescribe documentando la nueva división de responsabilidades.
- `src/uix/eidos/components/toggle-group/README.md`: sección "Recipe" + "Decisiones" actualizadas — la entrada "Structural identity" documenta el patrón.

**Por qué TSC v2.2 composition se queda en la mezcla**: la composition sigue siendo necesaria para overridear `--toggle-palette-*` en el item bajo `[data-toggle-group][data-color='X'] [data-toggle-group-item]`. El color NO se propaga via contexto porque la composition ya hace el trabajo a nivel CSS, sin overhead reactivo. Variant/size sí se propagan porque tienen muchos derivados (height, padding, gap, font, radius × 5 sizes; bg, fg, border, hover, on, on-hover × 3 variants) que solo Toggle's recipe ya emite — no había ningún beneficio en mantener cascadas paralelas.

**Verificación**: probe DOM en `/uix/components/toggle-group` capturando computed values en las 12 combinaciones (4 colores × 3 variants) × 2 estados (on/off) — match bit-a-bit con baseline pre-refactor. Ejemplos:
- `affirm/solid` on: `rgb(18,165,148)` = `#12a594` (palette-solid affirm)
- `risk/outline` on: `color(srgb 0.2 0.118 0.043)` ≈ `#331e0b` (color-mix outline on-bg)
- `threat/ghost` on: `rgb(25,17,17)` = `#191111` (palette-track threat)

**Tests**: 163/163 pass en `src/uix/morfo` + `src/uix/eidos`. 4/4 pass en `src/uix/soma/components/toggle-group`. `npm run check`: 16 errores pre-existentes (los mismos del sprint Words), CERO añadidos por esta migración.

**Coste real vs estimado**: el hand-off #5 estimó el cambio como "too big" — incorrecto. Total = 1 entrada en morfo + 1 archivo de contexto (~30 líneas) + 2 ediciones puntuales en wrappers (~5 líneas cada) + 1 rewrite de CSS reduciendo ~150 líneas a ~85. La parte engañosa era pensar que requería tocar Toggle's CSS — no lo hace.

**Doctrina reforzada**: cuando un wrapper componente reusa visualmente otro, la respuesta canónica NO es duplicar el cascade ni extender TSC con un tercer feature. Es declarar la identidad estructural en el morfo del wrapper. Patrón aplicable si emerge button-group, link-group, etc.

## Session hand-off — 2026-05-28 #2 (Words editor — 6 sprints UX)

Segunda mitad de la sesión (2026-05-28) dedicada a fixes del editor Words por feedback iterativo del usuario. Seis sprints EV-* + cleanup del check.

**Cleanup pre-sprint — `npm run check` 16 → 0**:
- 4 errores `orientation` drift en `soma/components/words-toolbar*` (deuda mía del sprint anterior — removí orientation del type pero no del soma component). Quitado el prop + create() call.
- 12 errores en el sprint Words activo (post-2026-05 F2/F3/COLOR): `onUploadImage` threading en soma `Words.Provider`, `'insert-image'` añadido a 2 Records `WordsToolbarButtonCommandName`, `leafItem` snippet hoisted FUERA de `<SomaWords.Provider>` (snippets dentro de un component element son props en Svelte 5; el snippet era helper local), 5 arrays inline en `words-drawer.svelte` extraídos a constantes typed `as const satisfies readonly { id: WordsCommandName | WordsMark; ... }[]`.

**EV-A — spam de eventos del canvas**:
- `isInsideWordsTool` extendido a los 5 overlays añadidos post-DRAWER: `data-words-drawer`, `data-words-block-handle`, `data-words-block-handle-menu`, `data-words-block-inserter`, `data-words-image-float-bar`. Antes, cualquier click sobre estos overlays se interpretaba como blur EXTERNO → fire `commit-save-content` + re-focus → fire `contact-focus` = 2 sonidos por interacción.
- `contact-focus` target movido de `content` a `provider` en morfo + sema cascade (doctrina: focus es evento de componente, no de body).

**EV-B — toolbar slim**:
- Presets demo (`minimal`/`formatting`/`full` + custom) reducidos a acciones GLOBALES: history (undo/redo) + insert + link + tools + find-replace. Text/block/list/align/table OUT porque el drawer ya los cubre por scope.

**EV-C — drag handle UX**:
- Borrado `e.dataTransfer.setDragImage(hoverBlockEl, 12, 12)`. El browser usa su snapshot por defecto (= el grip button) como ghost — el ghost viaja con el cursor mientras el bar en la gutter queda fijo como ancla visual. Nuevo `data-dragging` + CSS fade del ancla a 0.35 opacity.

**EV-D — inserter al borde inferior**:
- `seam.y` para seams entre bloques cambia de `(a.bottom + b.top) / 2` (midpoint) a `a.bottom` (borde inferior del bloque anterior). Half-open interval `[top, bottom)` para que el píxel exacto del bottom pertenezca al seam, no al bloque.

**EV-E — scroll interno del content**:
- Nuevo token `content-max-block-size-sm/md/lg` (50/60/70vh) en el recipe. `[data-words-content]` gana `max-block-size: var(--_words-content-max-block-size)` + `overflow-y: auto`. El min-block-size baseline queda como starting height para editores vacíos.
- block-handle / block-inserter / image-float-bar: scroll listeners migrados de `window.addEventListener('scroll')` a `document.addEventListener('scroll', { capture: true })`. Razón: scroll events NO burbujean — la versión anterior solo captaba scroll del root document; con la rail/handle/inserter ahora dentro de un content scrollable, había que capturar también scroll DENTRO del content para que las overlays se re-midieran.

**EV-F — paquete de 6 fixes en uno**:
1. Cascade sema de `contact-focus` REMOVIDA → canvas mudo en focus (EV-A fixaba solo el target; la cascade seguía sonando).
2. Rail bg → flat silver (`#d4d4d4`) + borde derecho `#9a9a9a`, sin dot pattern. Tokens `words.rail-bg / rail-border` en el recipe. **Fixed-tone, NOT theme-aware** — la intención es emular un margen físico de cuaderno, debe verse igual en dark/light theme.
3. Toolbar: `undo` + `redo` DIRECTOS (no popover); `insert-menu` eliminado. Resultado: 5 items.
4. Family-panel base 12.5 → 16rem · tools 17rem · link 20rem. Sin scroll horizontal.
5. Link popover sin scroll vertical (consecuencia del #4).
6. Engine `insertParagraph` con guarda explícita: paragraph vacío + Enter = no-op; heading vacío + Enter = demote a paragraph (canonical Notion UX).

**EV-G — último pass (parcial)**:
- `runtime.trigger('contact-focus')` comentado en soma. Causa real del sonido residual: la family `contact` en `SEMA_MAP` tiene BASE signature de sound (pitch 800, gain 0.25) que suena aunque no haya cascade per-component. Solución: no disparar el evento. Telemetry vacía para focus, aceptado.
- Drag handle + inserter ahora cubren `data-words-node="list"` y `data-words-node="table"`, no solo `"block"`. Los selectores en `findBlockElement`, `blockUnderCursorY`, `listBlockBoundaries` aceptan los tres tipos para top-level blocks. Listas y tablas tienen drag handle + inserter seam.
- `:hover` rules eliminadas en `[data-words-block-inserter-button]` y `[data-words-block-handle]`. Las overlays son AMBIENT — estado visual cambia solo en data attrs (`data-open`, `data-grabbed`, `data-dragging`). No se iluminan en hover.

**Decisiones arquitectónicas con efecto duradero**:
- **Canvas perceptualmente silencioso** (Words): `contact-focus` no se emite. Si en el futuro un consumidor necesita el evento, descomentar el `runtime.trigger` en `words-provider.svelte.ts > onfocus`.
- **Tokens FIXED-TONE para concepto físico** (Words rail): `--words-rail-bg/border` son hex literal en el recipe, no `var(--color-*)`. Justificación: el concepto visual es "margen de papel" — debe verse igual en todos los themes.
- **Drag-handle predicate cubre 3 tipos de node**: `block`/`list`/`table`. No solo `block`. Cualquier nuevo top-level node type tiene que entrar en la lista.
- **Overlays ambient, no interactive buttons**: gutter overlays (block-handle, block-inserter button) no tienen `:hover`. Estado visual via data attrs solamente.
- **Engine guard sobre bloque vacío**: `insertParagraph` ya no duplica párrafos vacíos. Convierte heading→paragraph en vacío.

**Pendientes documentados en CONTINUE.md**:
- P1 — Drag handle aparece FUERA del rail sobre code blocks (no diagnosticado).
- P1 — Heading inline level change (h1↔h2↔h3) — la UI no permite cambiar nivel inline; el drawer Block panel lo tendría que ofrecer cuando `currentBlock === 'heading'`.
- POLISH-1b / POLISH-3 / WORDS-F2.3-F2.6 / WORDS-DRAGDROP-SOMA — heredados.

**Tests al cierre**: 467/467 pass en morfo + sema + eidos + soma/components/words. 85/85 engine pass. `npm run check`: **0 errors**.

## Session hand-off — 2026-05-09 (selector discipline + lint reframing)

- **Typed selector builder applied** in
  `src/uix/sema/components/dialog.ts`. Every cascade rule now constructs
  its selector via `semaSelector(dialogMorfo, 'content', matchers?)`
  instead of hand-writing strings. The helper lives in
  `src/uix/morfo/selectors.ts` and is exported from `$uix/morfo`.
- **Eidos-lint reframed** from "architectural drift mechanism" to
  "opt-in safety net for plain CSS recipes". The architectural mechanism
  is compile-time typing. The lint script + scripts continue to work
  unchanged but are no longer the doctrinal source of safety.
- **Intent.deltas signature ownership documented** — cascade rules MUST
  NOT override `pitch` / `gain` / `contour` on `sound`, because doing so
  collapses the per-intent perceptual difference. Captured both inline
  in the dialog cascade (NOTE comment on the intent block) and in the
  "Override semantics — numbers" rule above.
- **Documentation updated**:
  - `src/uix/morfo/README.md` — new "Typed selector builder —
    `semaSelector`" section.
  - `src/uix/eidos/README.md` — replaced "Linter" section with
    "Defensa contra drift de selectores" framing the two-tier defense
    (compile-time builder for TS/Svelte, opt-in lint for plain CSS).
  - `src/uix/sema/README.md` — already documented the builder + the
    cascade discipline + the SEMA_FAMILY_POLICY shape (from the prior
    refactor).

**Verification at hand-off**: `npm run check` reports 0 errors, 2
pre-existing Svelte warnings unrelated to this work. `npm run test`
reports 1883/1883 tests green across 154 files (~22s).

## Session hand-off — 2026-05-08 (post-canon alignment)

The codebase has been aligned to the canonical guide. Most recent
architectural move:

- **Runtime relocated + renamed** — what was `MorfoRuntime` in
  `src/uix/morfo/runtime.svelte.ts` is now `SomaRuntime` in
  `src/uix/soma/runtime.svelte.ts`. Reason: morfo must remain pure
  declarative TypeScript (DNA); the runtime imports Svelte runes,
  `$adom`, `$uix/sema`, and `$libs/reactive` — those are soma's
  territory. Per the doctrine "morfo declares, soma executes" the
  runtime now lives where it conceptually belongs. Active-uix's
  `uix.runtime(morfo, sources)` factory stays unchanged; it imports
  `createSomaRuntime` from `$soma`.
- **Sema is now genuinely ornamental** — `ActiveUix.semantic` returns
  `EngineSemantic | undefined` (no longer throws). `SomaRuntime.trigger()`
  skips the `emit` step and the target-resolution check when no engine
  is present. Components stay functional in SSR / headless tests /
  audio-disabled environments.

The earlier swept phases (still in effect):

- **Sema verbs** (`src/uix/sema/verbs.ts`): restructured to
  family-keyed `Record`. `select`/`toggle`/`acknowledge` moved to
  `commit`; `edit` removed from `handle` (now expressed as
  `shift.enter-mode`). New verbs added per the canon (tap, focus,
  release, complete, restore, remind, route, step, syncing, etc.).
  `validateEventName` recognises both `{verb}-{variant}` and
  `{family}-{verb}` shapes.
- **Morfo event shape** (`src/uix/morfo/types.ts`): `target` moved
  inside `semantic`; added optional `verb` (advisory, validated against
  `SEMA_VERBS[family]`) and `sequence: 'pre' | 'coincident' | 'post'`.
  All 22 events across toggle/toast/popover/drawer/dialog refactored
  to the new shape.
- **Color tokens** (`src/uix/eidos/themes/base/{light,dark}.css`):
  doctrinal 8 tokens — `primary, secondary, neutral, affirm, fulfill,
  risk, threat, loss`. Renamed `success → fulfill`, `warning → risk`,
  `danger → threat` (same hex). New primitives for `secondary`
  (slate-blue), `affirm` (teal-mint, low activation), `loss` (deep
  violet-grave, posterior). `info` palette deleted entirely. Legacy
  aliases in `_static.css` deleted (clean cut, no transition).
  29 consumers (recipe CSS + tokens) migrated to doctrinal names.

Earlier in the session:

- **Toggle is the eidos pilot** (`src/uix/eidos/components/toggle/`):
  recipe + Svelte wrapper + types + index + README. The pattern is
  documented in `src/uix/eidos/components/README.md`.
- **Eidos has a shared lib** at `src/uix/eidos/lib/types.ts`. First
  type is `Size` (`'xxs' | 'xs' | 'sm' | 'md' | 'lg' | 'xl' | 'xxl' | 'full'`).
  Components narrow with `Extract<Size, 'sm' | 'md' | 'lg'>` per the
  per-component-subset doctrine.
- **No `Eidos` prefix on types**. The path
  `$uix/eidos/components/{x}` already conveys layer. Public types are
  `ToggleProps`, `ToggleVariant`, `ToggleSize`.
- **API doctrine**: soma stays compound (`Toggle.Provider`); eidos
  exports both `default` and `Provider` for single-part components so
  `<Toggle>` works for the ergonomic case while `<Toggle.Provider>`
  stays available.
- **SoundChannel eager-init landed** in `src/uix/sema/chans/sound.ts`
  to fix the autoplay race: the AudioContext is created + resumed
  synchronously on the first user gesture (capture-phase listener on
  `document`).
- **Demo page (`web/routes/toggle/+page.svelte`)** has the live
  preview rendered ALWAYS above the tablist (so Sema-tab Play buttons
  can fire on the real toggle) and the motion preview amplifies scale
  ×8 visually only — doctrinal values stay in the `<dl>`.

**Mandatory checklist before any eidos component migration**:

1. Recover the air baseline from git:
   ```
   git ls-tree -r 0a391408^ --name-only | Select-String "air.*{name}"
   git show 0a391408^:src/uix/air/components/{name}/types.ts
   git show 0a391408^:src/uix/air/components/{name}/{name}.svelte
   ```
2. Inventory air's props, slots, effects, perceptual emits.
3. Build a comparison table: feature × (air had it | canon allows it |
   eidos exposes it). Flag each gap as `regression`, `enhanced`,
   `dropped-by-canon`, or `nuevo`.
4. Present the table BEFORE coding. Get explicit per-gap decision.
5. Only then write the wrapper.

Switch was migrated WITHOUT this discipline and shipped a regression
(no `ResponsiveProp<Size>` for breakpoint-aware sizing). Don't repeat.

Next concrete steps:

1. Migrate `collapsible → dialog → drawer → popover → toast → avatar`
   to the eidos wrapper pattern. **For each: walk the air-baseline
   checklist above first.** Toggle / switch are templates for the
   subdirectory shape, but the per-component prop surface must come
   from air's recovery, not from copying toggle's props.
2. **Persistence + holds-by-intent** (guide §6) — defer until the
   first concrete `signal.warn` / `signal.alert` consumer appears.
   Today every signal is `transient` with a numeric hold. Need to add
   `persistence: 'transient' | 'untilAction' | 'untilFix' | 'stateBound'`
   to `SemanticSignal` + the holds-by-intent table per the canon.
3. **a11ySemantic per event** (guide §9) — defer until there's a
   reduced-motion runtime helper + live-region helper to consume the
   contract.
4. **Polymorphic events** (guide §5.3, `allowedFamilies` +
   `defaultSemantic`) — defer; no current component needs polymorphism.
5. The docs site's topbar prefs (sound mute toggle, theme switcher)
   are wired up but the sound mute does NOT yet drive
   `SoundChannel.masterGain`. Wiring belongs on layout.svelte's `$effect`.

Baseline `npm run check` shows 39 pre-existing errors in test files
unrelated to this work. Tests are 217/217 green across uix.

## Session hand-off — 2026-05-26 (sema canon + soma emission + READMEs)

Sprint completing the alignment of the project with the book *Diseñando lo que ocurre*. Nine commits closed P1 #3, P1 #4 and P1 #5 of the audit-codex:

```
504780ad  expression field + sema coverage check
2d562f37  11 sema packs + doctrina samples (D.5/D.6/D.7)
826ca2bf  D.8 channels scope doctrine
43b15372  picker-shell INTERNAL (P1 #5 closed)
66318c47  tree-view target → branch for emerge events
adf89a08  batch A — soma emission for 7 packs-shipped components
0784cae5  batch B — 8 components with sema scope, no pack
cbf76b66  batch C — drag-drop + virtual-list/grid
7afa7057  23 component READMEs with Sema events sections
```

**Architectural deltas from this sprint:**

- **`Morfo.expression`** new field (`'pack' | 'family-default' | 'delegated' | 'none'`) declares how each morfo with declared events materialises its perceptual signature. Schema-validated; lint enforces coverage. See `src/uix/morfo/types.ts` + `LIBRO_VARIACIONES_Y_EXTENSIONES.md` D.4.
- **`SemaFamily` is now 8** (added `delegate` per book cap. 29). `SEMA_FAMILY_POLICY` carries two axes (`intentRequirement` + `intentGuidance`) — see updated section above.
- **`SEMA_VERBS`** extended with `signal.inform`, `commit.unselect`, contextual verbs (`apply`, `partial`, `block`, `move`, `upload`, `acknowledge`, `confirm`).
- **38 sema packs** under `src/uix/sema/components/*.ts` compose 9 canonical tuning profiles (`form.commit.soft/subtle`, `form.toggle.silent`, `tooltip.silent`, `emerge.soft`, `emerge.exit/.soft`, `emerge.medium`, `emerge.dismiss.passive`, `tabs.select.soft`). Apps override at `defineEngineSemantic({ overrides: { cascade: [...] } })`.
- **Soma emission cabled** in 18 components (toggle-group, menubar, navigation-menu, dropdown-menu, context-menu, tree-view, tree-grid, listbox, grid-list, table, feed, command, carousel, announce, clipboard, drag-drop, virtual-list, virtual-grid). Pattern: emit from the central state-mutator with `fallbackTarget` resolving the specific element so the cascade matches the right instance.
- **`handle-scroll*` intentionally NOT emitted** in virtual-list / virtual-grid — family.handle activates haptic only; emitting on every pixel of scroll would buzz the device nonstop. Apps that want scroll-feedback wire their own throttled emit. Documented in both READMEs.
- **picker-shell relocated** from `src/uix/morfo/components/` to `src/uix/morfo/internal/` — declares its INTERNAL status by file location. Apps never write `<PickerShell>`; the five composite pickers (date / date-range / time / time-range / color) re-export the shell parts under their own namespace. README added at `src/uix/eidos/components/picker-shell/README.md`.
- **`LIBRO_VARIACIONES_Y_EXTENSIONES.md`** is the authoritative registry for project decisions vs the book canon. Sections D.4 (`expression` field), D.5 (toggles soft-tuned packs), D.6 (menus/trees packs), D.7 (samples doctrine — `SOUND_LIBRARY` = resources, `SOUND_TUNINGS` = canon, packs compose tunings never samples; `sampleOverlay` rejected permanently), D.8 (channels scope — `sound` + `haptic` are the only canonical runtime channels; ARIA structural lives in morfo, ARIA dynamic in soma, visual in eidos — never canonized as channels).
- **23 soma component READMEs** now include a `## Sema events` table with event / family / verb / target / intent / when + pack reference + doctrinal notes for the verb-rename corrections applied across the sprint.

**Verification at hand-off**: `npm run morfo:vocabulary` EXIT 0 (clean). `npx vitest run src/uix/sema src/uix/morfo` 195/195. `npx vitest run src/uix/soma` 419/419. Pre-existing words/* type errors unchanged (separate dev track).

**What's left from the audit-codex**: P2 (34 demos NEEDS-WORK) — explicitly out of scope (full refactor of demos + web routes pending separately).

---

# Behavioral Guidelines

These reduce common LLM coding mistakes. They bias toward caution over speed —
for trivial tasks, use judgment.

## 1. Think before coding

Don't assume. Don't hide confusion. Surface tradeoffs.

- State assumptions explicitly. If uncertain, ask.
- If multiple interpretations exist, present them — don't pick silently.
- If a simpler approach exists, say so. Push back when warranted.
- If something is unclear, stop. Name what's confusing. Ask.

## 2. Simplicity first

Minimum code that solves the problem. Nothing speculative.

- No features beyond what was asked.
- No abstractions for single-use code.
- No "flexibility" or "configurability" that wasn't requested.
- No error handling for impossible scenarios.
- If you write 200 lines and it could be 50, rewrite it.

## 3. Surgical changes

Touch only what you must. Clean up only your own mess.

- Don't "improve" adjacent code, comments, or formatting.
- Don't refactor things that aren't broken.
- Match existing style, even if you'd do it differently.
- If you notice unrelated dead code, mention it — don't delete it.

When your changes create orphans, remove the imports / variables / functions
your changes made unused. Don't remove pre-existing dead code unless asked.

## 4. Goal-driven execution

Define success criteria. Loop until verified.

- "Add validation" → "Write tests for invalid inputs, then make them pass."
- "Fix the bug" → "Write a test that reproduces it, then make it pass."
- "Refactor X" → "Ensure tests pass before and after."

Strong success criteria let you loop independently. Weak criteria require
constant clarification.