@hyperfixi/patterns-reference 2.11.1 → 3.1.0

package/CHANGELOG.md

@@ -7,6 +7,305 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [3.1.0] - 2026-09-04
+
+### Added
+
+- **`hyperscript.use(frontEnd)`** — register the multilingual front-end
+  `compile()` consults for a non-English program (and `toLSE`/`fromLSE` for
+  a semantic node and a renderer). The contract is `FrontEnd` in
+  `@hyperfixi/core` (`parseToAST`, optional `parse`/`render`);
+  `@lokascript/semantic` satisfies it through
+  `createBridgeFrontEnd(new SemanticGrammarBridge())` from
+  `@hyperfixi/core/multilingual`. The full browser bundle registers it at
+  boot. The library entry registers nothing and, through 3.x, builds the same
+  bridge lazily on the first non-English compile — so nothing changes for a
+  consumer with `@lokascript/semantic` installed. **4.0 will remove that
+  default:** a non-English `compile()` with no front-end registered will
+  return the same result the core-parser-only path returns today, and
+  `@lokascript/semantic` becomes an optional peer. Engine-migration plan
+  Arc 1, steps 2 and 3.
+
+### Fixed
+
+- **Language server: valid non-English code no longer gets a `parse-error`.**
+  In `lokascript`/`hyperscript-i18n` mode the server ran core's English
+  parser on every region regardless of what the semantic front-end said, so
+  `al hacer clic alternar .activo` (accepted at 0.89 confidence) was published
+  as `Unexpected token: hacer`. Core's errors are now suppressed for a region
+  the front-end accepted in a non-English language. The integration suite
+  that claimed to cover this had been discarding every notification; it now
+  records `publishDiagnostics` and asserts on it.
+
+- **Language server: settings actually reach the server.** Neither VS Code
+  extension set `synchronize.configurationSection`, so the client pushed
+  `settings: null` and every `mode`/`language`/`maxDiagnostics` value was
+  ignored; `initializationOptions` was ignored too. The server now reads
+  `initializationOptions`, a pushed `lokascript`/`hyperscript` section, or
+  pulls `workspace/configuration` when the push is empty — and merges partial
+  objects over the defaults instead of replacing them (a bare
+  `{ maxDiagnostics }` used to leave `language` undefined and error on every
+  document). Both extensions push their section, and the LokaScript one now
+  contributes `lokascript.mode`.
+
+- **Both VS Code extension server bundles build and run.** The standalone
+  hyperscript extension's `bundle:server` had failed since #1016 (no shim for
+  `@hyperfixi/core/multilingual`); the LokaScript extension externalized
+  `@hyperfixi/core` without shipping it, so every install ran with core
+  absent ("diagnostics and hover degraded"). Core is bundled in now, and CI
+  runs both bundle scripts.
+
+- **`@lokascript/language-server` declares what it imports.**
+  `@lokascript/semantic` was an optional peer but a static import (the
+  server crashed at startup without it); `@lokascript/framework` was
+  undeclared and got inlined together with `@lokascript/intent` (261 KB) —
+  and without `--clean`, `dist/` shipped every previous build's chunk (three
+  dead ones, 63% of the tarball). Both are regular dependencies now, the
+  build cleans, the 20-byte `--dts` output is gone, and `npm test` rebuilds
+  the server it spawns.
+
+- **Hyperscript-compat mode stops flagging portable code.** `as Int`, `Float`,
+  `JSON`, `Date`, `Set`, `Map` are upstream conversions (the extension list
+  is now `Math`/`Values`, the ones core registers and upstream lacks);
+  `my.x`/`your.x` evaluate upstream (only `its.` is flagged); and the
+  extension-command scan no longer fires inside `y.replace(`, `arr.push(`,
+  string literals, `:process`, `--` comments, or the upstream trailing
+  `… unless …` modifier.
+
+- **Language-server false positives and misses.** A `.hs` file containing a
+  query literal like `<li/>` was classified as HTML and lost every feature;
+  possessives after `)`/`]`/`>` and apostrophes in `--` comments raised
+  `unmatched-quote`; the outline reported the `on` in `toggle .x on me` as a
+  handler and `.init` as an init block; `install Foo(args)` produced
+  overlapping rename edits VS Code rejects; `on draggable:start` became a
+  variable named `:start`; and keyword boundaries were ASCII `\b`, so
+  definitions and symbols were silently absent in the ten non-Latin
+  languages. Completion-context and document-symbol logic moved out of the
+  entry script into `completion-context.ts` / `document-symbols.ts` so the
+  unit suite tests the shipped code rather than the hand-copied twins it
+  used to carry.
+
+- **MCP `get_diagnostics`, `get_completions` and `get_document_symbols` read
+  the AST.** The bridge guarded those paths on helper names core never
+  exported, so all three had only ever taken their token-based fallback.
+  They now read the interchange: the core parser's own errors become
+  diagnostics, completions inside a command offer that command's argument
+  shapes, symbols carry their commands as children, and a document with
+  several `end`-terminated handlers keeps all of them.
+
+- **`hyperscript.process()` honours the event grammar.** The API-side DOM
+  processor carried its own listener installer for `on …` attributes and
+  silently dropped what the runtime's installer handles: `on
+click[event.shiftKey]` fired on a plain click, `on mouseenter or click` never
+  fired on mouseenter, `on click from document` never fired. Every browser
+  bundle already went through the runtime; `process()` now does too, and
+  `config.logAll` moved into the runtime so both paths log. Pinned by
+  `api/dom-processor.test.ts`, which runs each case on both paths.
+
+- **The browser bundles' attribute processor now runs the element lifecycle,
+  and its lazy stub only takes the shapes it can run faithfully.** Measured
+  side by side with `hyperscript.process()`: the bundle path never dispatched
+  `hyperscript:before:init` / `hyperscript:after:init` and never set
+  `data-hyperscript-powered` — both lived only in the API processor, so a
+  morph engine reading the marker saw nothing on a bundle-processed page, and
+  canceling `before:init` was impossible there. It also compiled every
+  attribute as English; it now detects the element's language (`data-lang`,
+  closest `lang`, the document's) the way `process()` always did. And the
+  `lazyParsing` stub read only the event NAME from the header, so for the
+  first event it ignored a filter (`on click[event.shiftKey]` fired on a
+  plain click), listened for the first name of an `or` list only, and never
+  saw `from <target>` at all; a header with a filter, `(args)`, `or`, `from`,
+  `elsewhere`, `queue`, `debounced`, `throttled` or `in` is now processed
+  eagerly. `dom/processor-parity.test.ts` runs every row on all three paths
+  (API, eager, lazy) and is the gate the DOM-processor collapse landed under.
+
+- **`hyperscript.process()` and the browser bundles share one DOM
+  processor.** `process()` and `cleanup()` now run on the attribute
+  processor every bundle uses, which receives its compiler and runtime from
+  the API by injection instead of importing it (the `dom -> api` layering
+  edge is gone). What changes for `process()` callers, all of it upstream
+  parity: an element is initialized once however many times it is passed or
+  scanned (it used to re-install on every call); `<script
+type="text/hyperscript">` tags inside the tree are processed, with or
+  without `for=`, before the elements that may `install` what they define;
+  each element gets a non-bubbling `load` event once its program has run,
+  and an unparseable attribute is reported through `console.error`, the
+  `hyperfixi:compile-error` event and `config.onCompileError` (it was logged
+  at debug level only). And `cleanup(container)` now un-marks every processed
+  descendant — it stripped the root's `data-hyperscript-powered` only — and
+  drops a lazy stub still waiting on an element, so cleanup-then-process
+  re-initializes the tree once. A second scan in `lazyParsing` mode no longer
+  registers a second stub.
+  The API-side processor module is deleted; the multilingual bundle's own
+  copy of the language walk now calls the shared one, narrowed to the
+  languages that bundle ships; and a subtree the `MutationObserver` picks up
+  takes the same entry as `process()`.
+
+- **`set *<css-property>` writes inline style in every spelling upstream
+  accepts.** `set *opacity to 0.5`, `set *opacity of me to 0.5` and
+  `set *background-color of me to "red"` were silent no-ops and
+  `set the *opacity of me to 0.5` threw; only the possessive `set my *opacity`
+  worked. The tokenizer classes `*opacity` as a selector (for `measure`), and
+  `set`'s parser now re-types it into the identifier property the runtime's
+  style rungs already handle, so one path serves all of them — bare, `of
+<target>`, `the … of <target>`, `<target>'s`. Ten rows pinned against
+  `_hyperscript` 0.9.93.
+
+- **`require('@hyperfixi/core')` returned `{}`, and the subpath `require`s
+  threw — since `"type": "module"`, on 3.0.0 too.** Every `exports.*.require`
+  and `main` pointed at a `.js` file built as CommonJS, which Node reads as
+  ESM under that flag. The CJS outputs are `.cjs` now (main entry, the 21
+  subpaths, the parser modules); the `import` conditions are unchanged. The
+  bare-Node check in CI now `require()`s the entries as well as importing
+  them, which is the check that would have caught this. Found by the
+  end-to-end probe that landed Arc 1's build half.
+
+### Changed
+
+- **`@hyperfixi/core`'s library entry no longer bundles the multilingual
+  front-end.** `dist/index.mjs`/`index.js` had `@lokascript/semantic`,
+  `@lokascript/intent` and `@lokascript/framework` inlined whole (3.33 MB, zero
+  dynamic imports left); they are external now (1.04 MB) and load on the first
+  non-English `compileAsync`, the way the source always said they would. No
+  API change: semantic and intent stay `dependencies`, framework stays an
+  optional peer — and is now genuinely one. A consumer that imported both core
+  and semantic no longer holds two copies of semantic. `dist/index.min.js`
+  (UMD, not in `exports`) stays self-contained. Gated by
+  `scripts/check-node-import.mjs` on the sourcemap. Engine-migration plan Arc 1
+  step 2, build half.
+
+- **`@lokascript/semantic` drops its `asyncSchema`.** Core deleted the `async`
+  command in 3.0.0 (#1102); the semantic front-end never emitted the action
+  anyway — its `stripAsyncModifier` pass removes the keyword (in all 24
+  languages) and parses the following command, so `async fetch /api as json`
+  is a `fetch`. The schema, its registration, the `'async'` `ActionType`
+  member and the AST-builder mapper row are gone. Kept on purpose: every
+  profile's `keywords.async` (the stripper matches on it) and the
+  `'then' | 'and' | 'async'` chain type (a different concept). The IR-level
+  `async all/race` node lives in `@lokascript/intent` and is untouched.
+
+## [3.0.0] - 2026-09-03
+
+Full notes: [GitHub Releases](https://github.com/codetalcott/hyperfixi/releases/tag/v3.0.0).
+
+One major, two themes. `@lokascript/i18n`'s grammar transformer is gone,
+because `@lokascript/semantic` overtook it on every row of the translation
+corpus (#973–#1001). And the engine-migration plan
+(`docs-internal/ENGINE_MIGRATION_PLAN.md`) closed its last arc, which deleted
+the exported dead code it had been carrying and collapsed the browser bundle
+lineup to two names (#1099–#1105). One arc is still open: Arc 1's steps 2
+and 3 — `@hyperfixi/core`'s library entry still bundles the semantic
+front-end (recorded in the plan's History, 2026-09-03).
+
+### ⚠ BREAKING
+
+- **`@lokascript/i18n` no longer translates** (#1001). `translate()`, `toLocale()`,
+  `toEnglish()`, `GrammarTransformer` and `createTransformer` are deleted; the
+  package is per-language vocabulary and grammar profiles. Migrate to
+  `import { translate } from '@lokascript/semantic'` (same signature; it throws
+  on unparseable input where the transformer returned nonsense). The classic-i18n
+  browser bundle, `@hyperscript-tools/i18n` and `@lokascript/types-browser` drop
+  the same members (#998, #999); the patterns corpus writer is semantic-only (#1000).
+- **The prebuilt browser bundles are two names** (#1105): `hyperfixi-hx.js`
+  (small: hybrid parser, blocks, expressions, htmx v1/v2 attributes) and
+  `hyperfixi.js` (everything); `hyperfixi-hx-v4.js` and
+  `hyperfixi-multilingual.js` stay as separate products. `hyperfixi-lite.js`,
+  `-lite-plus.js`, `-minimal.js`, `-standard.js` and their `./browser/*`
+  exports are gone — use `hyperfixi-hx.js`, `hyperfixi.js`, or the Vite
+  plugin, which picks the tier itself. `./browser/hybrid-complete` remains as
+  the plugin's internal fallback.
+- **Exported dead code deleted from `@hyperfixi/core`** — none had a
+  production caller: the six `@deprecated` `features/` families and the
+  modular bundle's `features` namespace (#1099); `Lexer`/`Tokens` (#1100 —
+  use `tokenize`); the `unified-types` `Validator`, the `types.d.ts` shim and
+  `registry/multilingual` (#1101); the `async` command (#1102 — it was
+  unreachable from parsed hyperscript; the manifest is 58 commands);
+  `ContextProviderRegistry`, the context-provider `Proxy`, the registry's
+  `context` slot and the plugin `contextProviders` field (#1104 — set
+  request-scoped values as `context.locals`).
+
+### Fixed
+
+- **A small bundle fails loudly on a command it lacks and names `hyperfixi.js`**
+  (#1103). The hybrid parser had been dropping an unrecognised word silently;
+  making it loud also exposed and fixed three silent mis-parses in the hybrid
+  bundles: unquoted `fetch /api/data` fetched `/`, `send custom:event` sent
+  `custom` to `me`, and `repeat forever` ran zero iterations.
+- **English→foreign rendering is gated** (#931–#972, #953): 3105/3105 corpus
+  renders parse on the canonical engine; the bare (handler-less) surface has its
+  own gate. `as JSON` stays part of the value (#991); a flattened loop header
+  closes (#992).
+- Plugins report their real versions (#926); `analyze_content` is unshadowed in
+  the MCP server (#927); four dependency advisories resolved; the `domain-flow`
+  → `server-bridge` route contract is pinned (#1000s).
+
+### Changed
+
+- `hyperfixi.js` and `hyperfixi-hx-v4.js` carry the 24-language render
+  vocabulary (~19 KB gzip each) so `translate` works in the browser (#931).
+- Deprecated `@lokascript/domain-*` references repoint to `@lokascript/domains`.
+
+## [2.11.1] - 2026-08-25
+
+Patch release. Fixes the release workflow itself: the build now runs **after**
+the version sync, so a published tarball carries its own version rather than
+the previous one, and `version.ts` is committed (#925). Also lands the
+agent-loop A/B benchmark with an isolated generator (#924).
+
+## [2.11.0] - 2026-08-24
+
+Full notes: [GitHub Releases](https://github.com/codetalcott/hyperfixi/releases/tag/v2.11.0).
+
+### Changed
+
+- **The domain-DSL family moved out** (#909). Twelve packages left this
+  monorepo: the ten publishable `@lokascript/domain-*` packages (bdd,
+  behaviorspec, config, flow, jsx, learn, llm, sql, todo, voice) plus the
+  private `domain-toolkit` and `mcp-multilingual-intent`. They now ship as
+  **`@lokascript/domains`**, one subpath export per domain, from the
+  [lokascript-domains](https://github.com/codetalcott/lokascript-domains)
+  repository.
+
+  All ten published names are **deprecated on npm**, each naming its
+  replacement subpath (`@lokascript/domain-sql` → "import from
+  `@lokascript/domains/sql`"). Existing installs keep working — deprecation is
+  a warning, not a removal — they simply stop receiving updates. The
+  pre-deletion tree is preserved at the `moved/domain-family` tag.
+
+  _Migration:_ replace nine dependencies with one, and change
+  `from '@lokascript/domain-x'` to `from '@lokascript/domains/x'`. The root
+  entry absorbed `domain-config` and exports `createDomainRegistry`,
+  `registerAllDomains` and `DOMAIN_PRIORITY`.
+
+### Added
+
+- **"Show in My Language"** — LSP request plus VSCode command (#921).
+- **Verified-translation badge** in the compilation service (#920).
+- **`scoreFidelity` / `@lokascript/semantic/fidelity`** (#919).
+- **Inert-shape warnings** (#918) and surfaced unconsumed-input warnings (#916)
+  in the compilation service.
+- **The agent-loop benchmark** and the silent-failure finding it produced
+  (#915), and the MCP server's agent-era roadmap (#914).
+- **Element-collection write-back** for hypermedia tables, with gallery
+  examples and four system fixes (#904).
+- **The comprehensive Playwright tier now runs in CI** (#908) alongside
+  `quick`. It had never run: 122 specs, hiding four real bugs — a detached
+  `startViewTransition`, untracked reads of unset globals, concurrent effects
+  clobbering dependency capture, and `put` stringifying DocumentFragments
+  (#905).
+
+### Fixed
+
+- **The `hyperscript-adapter` review arc** (#895–#902): whole-string
+  translation first, taking canonical validity from 2849/3105 to **3105/3105**
+  (#899); a host-parser validity gate so an invalid render falls back to the
+  author's text (#900); the dead split-statement fallback deleted after
+  measuring 0 outputs (#901); and the slim divergence set burned 6 → 1 via
+  schema marker data, repairing 39 body-dropping corpus rows (#902).
+- **`qu take.recipient`** — the R1 tail's last `take` row (#910).
+- **The release workflows** no longer reference the moved-out domain family
+  (#922).
+
 ## [2.10.0] - 2026-08-01
 
 Full notes: [GitHub Releases](https://github.com/codetalcott/hyperfixi/releases/tag/v2.10.0).
@@ -394,7 +693,8 @@ _Synchronized version release. See git history for details._
 - npm access token stored in GitHub Secrets
 - 2FA recommended for npm organization
 
-[Unreleased]: https://github.com/codetalcott/hyperfixi/compare/v2.10.0...HEAD
+[Unreleased]: https://github.com/codetalcott/hyperfixi/compare/v3.1.0...HEAD
+[3.1.0]: https://github.com/codetalcott/hyperfixi/compare/v3.0.0...v3.1.0
 [2.10.0]: https://github.com/codetalcott/hyperfixi/compare/v2.9.0...v2.10.0
 [2.9.0]: https://github.com/codetalcott/hyperfixi/compare/v2.8.0...v2.9.0
 [2.8.0]: https://github.com/codetalcott/hyperfixi/compare/v2.5.1...v2.8.0

package/data/engine-verification.json

@@ -5,7 +5,7 @@
       "hyperscript": "upstream _hyperscript parse with zero recovered errors (extensions loaded: socket, worker, eventsource); parse-level only",
       "html": "HTML-markup patterns: each _= / script-tag snippet verified individually; hx-live/sse-*/ws-* attributes are hyperfixi-only and block the upstream claim"
     },
-    "lokascriptVersion": "2.9.4",
+    "lokascriptVersion": "2.11.1",
     "hyperscriptVersion": "0.9.93"
   },
   "engines": {
@@ -169,7 +169,7 @@
     "wait-for-event": "both",
     "wait-then": "both",
     "when-multiple-changes": "both",
-    "when-value-changes": "lokascript",
+    "when-value-changes": "both",
     "window-keydown": "both",
     "window-resize": "both",
     "window-scroll": "both",
@@ -847,8 +847,7 @@
     },
     "when-value-changes": {
       "lokascript": true,
-      "hyperscript": false,
-      "hyperscriptError": "Unexpected value: $"
+      "hyperscript": true
     },
     "window-keydown": {
       "lokascript": true,

package/dist/api/index.d.mts

@@ -1,5 +1,5 @@
-export { a as addLLMExample, b as buildFewShotContext, g as getAllPatterns, c as getAllTranslations, d as getExamplesByCommand, e as getHighConfidenceTranslations, f as getHighQualityExamples, h as getLLMExamples, i as getLLMStats, j as getMostUsedExamples, k as getPatternById, l as getPatternStats, m as getPatternsByCategory, n as getPatternsByCommand, o as getTranslation, p as getTranslationStats, q as getTranslationsByLanguage, r as getVerifiedTranslations, s as getWordOrder, t as searchPatterns, u as updateQualityScore, v as verifyTranslation } from '../llm-BfjUr-al.mjs';
-import { C as ConnectionOptions, P as PatternRole, a as Pattern, R as RoleType } from '../index-1ENU9L4F.mjs';
+export { a as addLLMExample, b as buildFewShotContext, g as getAllPatterns, c as getAllTranslations, d as getExamplesByCommand, e as getHighConfidenceTranslations, f as getHighQualityExamples, h as getLLMExamples, i as getLLMStats, j as getMostUsedExamples, k as getPatternById, l as getPatternStats, m as getPatternsByCategory, n as getPatternsByCommand, o as getTranslation, p as getTranslationStats, q as getTranslationsByLanguage, r as getVerifiedTranslations, s as getWordOrder, t as searchPatterns, u as updateQualityScore, v as verifyTranslation } from '../llm-B5nGz8V1.mjs';
+import { C as ConnectionOptions, P as PatternRole, a as Pattern, R as RoleType } from '../index-6GkHj5yJ.mjs';
 import { SemanticRole } from '@lokascript/semantic';
 
 /**

package/dist/api/index.d.ts

@@ -1,5 +1,5 @@
-export { a as addLLMExample, b as buildFewShotContext, g as getAllPatterns, c as getAllTranslations, d as getExamplesByCommand, e as getHighConfidenceTranslations, f as getHighQualityExamples, h as getLLMExamples, i as getLLMStats, j as getMostUsedExamples, k as getPatternById, l as getPatternStats, m as getPatternsByCategory, n as getPatternsByCommand, o as getTranslation, p as getTranslationStats, q as getTranslationsByLanguage, r as getVerifiedTranslations, s as getWordOrder, t as searchPatterns, u as updateQualityScore, v as verifyTranslation } from '../llm-FVj71n77.js';
-import { C as ConnectionOptions, P as PatternRole, a as Pattern, R as RoleType } from '../index-1ENU9L4F.js';
+export { a as addLLMExample, b as buildFewShotContext, g as getAllPatterns, c as getAllTranslations, d as getExamplesByCommand, e as getHighConfidenceTranslations, f as getHighQualityExamples, h as getLLMExamples, i as getLLMStats, j as getMostUsedExamples, k as getPatternById, l as getPatternStats, m as getPatternsByCategory, n as getPatternsByCommand, o as getTranslation, p as getTranslationStats, q as getTranslationsByLanguage, r as getVerifiedTranslations, s as getWordOrder, t as searchPatterns, u as updateQualityScore, v as verifyTranslation } from '../llm-rEdJQScF.js';
+import { C as ConnectionOptions, P as PatternRole, a as Pattern, R as RoleType } from '../index-6GkHj5yJ.js';
 import { SemanticRole } from '@lokascript/semantic';
 
 /**

package/dist/{index-1ENU9L4F.d.mts → index-6GkHj5yJ.d.mts}

@@ -36,7 +36,21 @@ interface ClassifiedPattern extends Pattern {
 }
 type ComplexityLevel = 'simple-command' | 'simple-event' | 'chained-command' | 'multi-line-event' | 'behavior-install' | 'behavior-def' | 'def-function' | 'complex';
 type WordOrder = 'SVO' | 'SOV' | 'VSO' | 'V2';
-type TranslationMethod = 'auto-generated' | 'hand-crafted' | 'verified';
+/**
+ * How a `pattern_translations` row was produced. The first three are the
+ * values `sync-translations` writes today; the rest are legacy/aspirational
+ * labels kept so existing rows and `getTranslationPriority` keep typing.
+ *
+ * - `semantic-render`: @lokascript/semantic `render(parse_en(en), L)` chosen by
+ *   the `best` writer (or the `semantic` writer).
+ * - `grammar-transform`: @lokascript/i18n GrammarTransformer — under the `best`
+ *   writer, a row i18n still renders better than semantic (the kept-row ratchet).
+ * - `grammar-transform-no-reference`: `best` only — the i18n row because semantic
+ *   cannot parse the ENGLISH source (a parser-coverage gap, not a render loss).
+ * - `keyword-substitute`: word-for-word fallback for a language with no grammar profile.
+ * - `original`: the English row; `non-translatable-identity`: markup rows copied verbatim.
+ */
+type TranslationMethod = 'semantic-render' | 'grammar-transform' | 'grammar-transform-no-reference' | 'keyword-substitute' | 'original' | 'non-translatable-identity' | 'auto-generated' | 'hand-crafted' | 'verified';
 /**
  * A translation of a pattern to a specific language.
  */

package/dist/{index-1ENU9L4F.d.ts → index-6GkHj5yJ.d.ts}

@@ -36,7 +36,21 @@ interface ClassifiedPattern extends Pattern {
 }
 type ComplexityLevel = 'simple-command' | 'simple-event' | 'chained-command' | 'multi-line-event' | 'behavior-install' | 'behavior-def' | 'def-function' | 'complex';
 type WordOrder = 'SVO' | 'SOV' | 'VSO' | 'V2';
-type TranslationMethod = 'auto-generated' | 'hand-crafted' | 'verified';
+/**
+ * How a `pattern_translations` row was produced. The first three are the
+ * values `sync-translations` writes today; the rest are legacy/aspirational
+ * labels kept so existing rows and `getTranslationPriority` keep typing.
+ *
+ * - `semantic-render`: @lokascript/semantic `render(parse_en(en), L)` chosen by
+ *   the `best` writer (or the `semantic` writer).
+ * - `grammar-transform`: @lokascript/i18n GrammarTransformer — under the `best`
+ *   writer, a row i18n still renders better than semantic (the kept-row ratchet).
+ * - `grammar-transform-no-reference`: `best` only — the i18n row because semantic
+ *   cannot parse the ENGLISH source (a parser-coverage gap, not a render loss).
+ * - `keyword-substitute`: word-for-word fallback for a language with no grammar profile.
+ * - `original`: the English row; `non-translatable-identity`: markup rows copied verbatim.
+ */
+type TranslationMethod = 'semantic-render' | 'grammar-transform' | 'grammar-transform-no-reference' | 'keyword-substitute' | 'original' | 'non-translatable-identity' | 'auto-generated' | 'hand-crafted' | 'verified';
 /**
  * A translation of a pattern to a specific language.
  */

package/dist/index.d.mts

@@ -1,7 +1,7 @@
-import { C as ConnectionOptions, b as Command, E as Expression, F as Feature, K as Keyword, S as SpecialSymbol, L as LanguageDocsStats, c as LanguageElementType, d as LanguageElement, e as LLMExample, f as PatternsReference } from './index-1ENU9L4F.mjs';
-export { g as ClassifiedPattern, h as ComplexityLevel, D as DiscoveryResult, i as EngineCompat, j as ExpressionOperator, a as Pattern, k as PatternStats, l as SearchOptions, m as SyncOptions, n as SyncResult, T as TestOptions, o as Translation, p as TranslationMethod, V as ValidationOptions, q as ValidationResult, r as VerificationResult, W as WordOrder } from './index-1ENU9L4F.mjs';
+import { C as ConnectionOptions, b as Command, E as Expression, F as Feature, K as Keyword, S as SpecialSymbol, L as LanguageDocsStats, c as LanguageElementType, d as LanguageElement, e as LLMExample, f as PatternsReference } from './index-6GkHj5yJ.mjs';
+export { g as ClassifiedPattern, h as ComplexityLevel, D as DiscoveryResult, i as EngineCompat, j as ExpressionOperator, a as Pattern, k as PatternStats, l as SearchOptions, m as SyncOptions, n as SyncResult, T as TestOptions, o as Translation, p as TranslationMethod, V as ValidationOptions, q as ValidationResult, r as VerificationResult, W as WordOrder } from './index-6GkHj5yJ.mjs';
 import Database from 'better-sqlite3';
-export { a as addLLMExample, b as buildFewShotContext, g as getAllPatterns, c as getAllTranslations, w as getBehaviorPatterns, d as getExamplesByCommand, e as getHighConfidenceTranslations, f as getHighQualityExamples, h as getLLMExamples, i as getLLMStats, j as getMostUsedExamples, k as getPatternById, l as getPatternStats, m as getPatternsByCategory, n as getPatternsByCommand, o as getTranslation, p as getTranslationStats, q as getTranslationsByLanguage, r as getVerifiedTranslations, s as getWordOrder, t as searchPatterns, u as updateQualityScore, v as verifyTranslation } from './llm-BfjUr-al.mjs';
+export { a as addLLMExample, b as buildFewShotContext, g as getAllPatterns, c as getAllTranslations, w as getBehaviorPatterns, d as getExamplesByCommand, e as getHighConfidenceTranslations, f as getHighQualityExamples, h as getLLMExamples, i as getLLMStats, j as getMostUsedExamples, k as getPatternById, l as getPatternStats, m as getPatternsByCategory, n as getPatternsByCommand, o as getTranslation, p as getTranslationStats, q as getTranslationsByLanguage, r as getVerifiedTranslations, s as getWordOrder, t as searchPatterns, u as updateQualityScore, v as verifyTranslation } from './llm-B5nGz8V1.mjs';
 export { DbStampStatus, checkDbStamp, computeDbInputHash, dbStampPath, discoverPatterns, seedLLMExamples, syncTranslations, validateAllTranslations, writeDbStamp } from './sync/index.mjs';
 import '@lokascript/semantic';
 

package/dist/index.d.ts

@@ -1,7 +1,7 @@
-import { C as ConnectionOptions, b as Command, E as Expression, F as Feature, K as Keyword, S as SpecialSymbol, L as LanguageDocsStats, c as LanguageElementType, d as LanguageElement, e as LLMExample, f as PatternsReference } from './index-1ENU9L4F.js';
-export { g as ClassifiedPattern, h as ComplexityLevel, D as DiscoveryResult, i as EngineCompat, j as ExpressionOperator, a as Pattern, k as PatternStats, l as SearchOptions, m as SyncOptions, n as SyncResult, T as TestOptions, o as Translation, p as TranslationMethod, V as ValidationOptions, q as ValidationResult, r as VerificationResult, W as WordOrder } from './index-1ENU9L4F.js';
+import { C as ConnectionOptions, b as Command, E as Expression, F as Feature, K as Keyword, S as SpecialSymbol, L as LanguageDocsStats, c as LanguageElementType, d as LanguageElement, e as LLMExample, f as PatternsReference } from './index-6GkHj5yJ.js';
+export { g as ClassifiedPattern, h as ComplexityLevel, D as DiscoveryResult, i as EngineCompat, j as ExpressionOperator, a as Pattern, k as PatternStats, l as SearchOptions, m as SyncOptions, n as SyncResult, T as TestOptions, o as Translation, p as TranslationMethod, V as ValidationOptions, q as ValidationResult, r as VerificationResult, W as WordOrder } from './index-6GkHj5yJ.js';
 import Database from 'better-sqlite3';
-export { a as addLLMExample, b as buildFewShotContext, g as getAllPatterns, c as getAllTranslations, w as getBehaviorPatterns, d as getExamplesByCommand, e as getHighConfidenceTranslations, f as getHighQualityExamples, h as getLLMExamples, i as getLLMStats, j as getMostUsedExamples, k as getPatternById, l as getPatternStats, m as getPatternsByCategory, n as getPatternsByCommand, o as getTranslation, p as getTranslationStats, q as getTranslationsByLanguage, r as getVerifiedTranslations, s as getWordOrder, t as searchPatterns, u as updateQualityScore, v as verifyTranslation } from './llm-FVj71n77.js';
+export { a as addLLMExample, b as buildFewShotContext, g as getAllPatterns, c as getAllTranslations, w as getBehaviorPatterns, d as getExamplesByCommand, e as getHighConfidenceTranslations, f as getHighQualityExamples, h as getLLMExamples, i as getLLMStats, j as getMostUsedExamples, k as getPatternById, l as getPatternStats, m as getPatternsByCategory, n as getPatternsByCommand, o as getTranslation, p as getTranslationStats, q as getTranslationsByLanguage, r as getVerifiedTranslations, s as getWordOrder, t as searchPatterns, u as updateQualityScore, v as verifyTranslation } from './llm-rEdJQScF.js';
 export { DbStampStatus, checkDbStamp, computeDbInputHash, dbStampPath, discoverPatterns, seedLLMExamples, syncTranslations, validateAllTranslations, writeDbStamp } from './sync/index.js';
 import '@lokascript/semantic';
 

package/dist/{llm-BfjUr-al.d.mts → llm-B5nGz8V1.d.mts}

@@ -1,4 +1,4 @@
-import { l as SearchOptions, C as ConnectionOptions, a as Pattern, k as PatternStats, o as Translation, W as WordOrder, r as VerificationResult, e as LLMExample } from './index-1ENU9L4F.mjs';
+import { l as SearchOptions, C as ConnectionOptions, a as Pattern, k as PatternStats, o as Translation, W as WordOrder, r as VerificationResult, e as LLMExample } from './index-6GkHj5yJ.mjs';
 
 /**
  * Pattern Queries API

package/dist/{llm-FVj71n77.d.ts → llm-rEdJQScF.d.ts}

@@ -1,4 +1,4 @@
-import { l as SearchOptions, C as ConnectionOptions, a as Pattern, k as PatternStats, o as Translation, W as WordOrder, r as VerificationResult, e as LLMExample } from './index-1ENU9L4F.js';
+import { l as SearchOptions, C as ConnectionOptions, a as Pattern, k as PatternStats, o as Translation, W as WordOrder, r as VerificationResult, e as LLMExample } from './index-6GkHj5yJ.js';
 
 /**
  * Pattern Queries API

package/dist/sync/index.d.mts

@@ -1,4 +1,4 @@
-import { D as DiscoveryResult, m as SyncOptions, n as SyncResult, V as ValidationOptions, q as ValidationResult } from '../index-1ENU9L4F.mjs';
+import { D as DiscoveryResult, m as SyncOptions, n as SyncResult, V as ValidationOptions, q as ValidationResult } from '../index-6GkHj5yJ.mjs';
 import '@lokascript/semantic';
 
 /**

package/dist/sync/index.d.ts

@@ -1,4 +1,4 @@
-import { D as DiscoveryResult, m as SyncOptions, n as SyncResult, V as ValidationOptions, q as ValidationResult } from '../index-1ENU9L4F.js';
+import { D as DiscoveryResult, m as SyncOptions, n as SyncResult, V as ValidationOptions, q as ValidationResult } from '../index-6GkHj5yJ.js';
 import '@lokascript/semantic';
 
 /**

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@hyperfixi/patterns-reference",
-  "version": "2.11.1",
+  "version": "3.1.0",
   "description": "Queryable patterns database for hyperscript with multilingual translations and LLM support",
   "main": "dist/index.js",
   "module": "dist/index.mjs",
@@ -66,13 +66,13 @@
   },
   "homepage": "https://github.com/codetalcott/hyperfixi/tree/main/packages/patterns-reference#readme",
   "dependencies": {
-    "@lokascript/semantic": "^2.11.1",
+    "@lokascript/semantic": "^3.1.0",
     "better-sqlite3": "^13.0.1"
   },
   "devDependencies": {
-    "@hyperfixi/core": "^2.11.1",
-    "@hyperfixi/reactivity": "^2.11.1",
-    "@hyperfixi/realtime": "^2.11.1",
+    "@hyperfixi/core": "^3.1.0",
+    "@hyperfixi/reactivity": "^3.1.0",
+    "@hyperfixi/realtime": "^3.1.0",
     "@types/better-sqlite3": "^7.6.0",
     "@types/node": "^26.1.2",
     "hyperscript.org": "0.9.93",

package/src/sync/db-stamp.ts

@@ -74,6 +74,10 @@ export function computeDbInputHash(dbPath: string): string {
   const root = repoRootFromDbPath(dbPath);
   const files = dbInputFiles(dbPath);
   const h = createHash('sha256');
+  // The renderer used to be hashed in here, because a DB written under
+  // PATTERNS_RENDERER=i18n|semantic|best was a DIFFERENT corpus from the same
+  // source and a gate expecting the default had to see it as stale. There is one
+  // renderer now, so the source files alone determine the corpus.
   h.update(`files:${files.length}\0`);
   for (const f of files) {
     h.update(relative(root, f).split(sep).join('/'));

package/src/sync/markup-attributes.ts

@@ -0,0 +1,111 @@
+/**
+ * Locate and replace the hyperscript bodies inside an HTML-markup corpus row.
+ *
+ * WHY A REGEX, HERE, DELIBERATELY
+ * -------------------------------
+ * `html-snippets.ts` warns against regex-scanning HTML — correctly, for its job
+ * (walking whole `.html` FILES, where a `<pre>` showing escaped source matches a
+ * live attribute). This module has the opposite job: it takes one known-markup
+ * corpus row and must splice a translated body back in **without disturbing a
+ * single other byte** — indentation, quote style, attribute order, inner text.
+ * A DOM round-trip (jsdom `innerHTML`) reserializes all of that, so it is the
+ * wrong tool: the corpus stores these rows verbatim and they are compared
+ * verbatim. Offsets + splice keeps everything outside the attribute value
+ * untouched by construction.
+ *
+ * Scope: the `_` attribute, double- or single-quoted, which is what the corpus's
+ * markup rows carry. A value may contain the OTHER quote character — hyperscript
+ * string literals inside a double-quoted attribute are single-quoted, which is
+ * how `_="set ^user to {name: 'Demo'}"` is written — so the value class excludes
+ * only its own delimiter. (`hx-live` bodies are hyperscript too, but every corpus row
+ * using one is flagged non-translatable — its attribute NAMES are resolved
+ * per-language by vocab modules, so translating only the body would be half a
+ * job.) An attribute whose value contains its own quote character cannot occur
+ * in well-formed HTML and is not handled.
+ */
+
+/** One `_="…"` attribute value found in a markup row. */
+export interface AttributeSpan {
+  /** Offset of the first character of the VALUE (inside the quotes). */
+  readonly start: number;
+  /** Offset one past the last character of the value. */
+  readonly end: number;
+  /** The raw attribute value — the hyperscript body. */
+  readonly body: string;
+}
+
+// `_` preceded by a tag-name/whitespace boundary so `data_x="…"` is not matched.
+// The value excludes only its own delimiter: `[^"]*` for a double-quoted value,
+// `[^']*` for a single-quoted one. (A single class excluding BOTH quotes was the
+// first version and it silently skipped every attribute containing a hyperscript
+// string literal — `component-with-conditional`, in all 23 languages.)
+const ATTRIBUTE = /(^|[\s"'])_\s*=\s*(?:"([^"]*)"|'([^']*)')/g;
+
+/** Every `_="…"` body in `markup`, in source order. */
+export function findHyperscriptAttributes(markup: string): AttributeSpan[] {
+  const spans: AttributeSpan[] = [];
+  ATTRIBUTE.lastIndex = 0;
+  let match: RegExpExecArray | null;
+  while ((match = ATTRIBUTE.exec(markup)) !== null) {
+    const body = match[2] ?? match[3];
+    // The value's offset: match start + everything before the body in the match.
+    const start = match.index + match[0].length - body.length - 1;
+    spans.push({ start, end: start + body.length, body });
+  }
+  return spans;
+}
+
+/**
+ * Replace each span's body with `replacements[i]`, leaving every other byte of
+ * `markup` untouched. Splices right-to-left so earlier offsets stay valid.
+ */
+export function spliceHyperscriptAttributes(
+  markup: string,
+  spans: readonly AttributeSpan[],
+  replacements: readonly string[]
+): string {
+  if (spans.length !== replacements.length) {
+    throw new Error(
+      `spliceHyperscriptAttributes: ${spans.length} spans but ${replacements.length} replacements`
+    );
+  }
+  let out = markup;
+  for (let i = spans.length - 1; i >= 0; i--) {
+    out = out.slice(0, spans[i].start) + replacements[i] + out.slice(spans[i].end);
+  }
+  return out;
+}
+
+/**
+ * Whether a corpus row is HTML markup rather than hyperscript text.
+ *
+ * Mirrors testing-framework's `isHtmlMarkupPattern` (top-level hyperscript never
+ * begins with `<`; a leading selector literal like `<button/>` is an expression,
+ * never a statement start). Kept as its own copy rather than a dependency: this
+ * package is upstream of the testing framework in the build order.
+ */
+export function isMarkupRow(code: string): boolean {
+  return /^\s*<[a-zA-Z!]/.test(code);
+}
+
+/**
+ * Whether a parse carried the whole body: every non-whitespace character of the
+ * source reappears in its own English re-render.
+ *
+ * This is the guard that keeps a TRUNCATING parse out of the corpus. Measured
+ * 2026-08-27: `set ^user to attrs.data as JSON` parses, scores "faithful"
+ * against itself in all 23 languages, and has silently dropped `as JSON` — the
+ * conversion is in no role, so every recall metric compares two equally
+ * truncated things. Translating such a body would ship the truncation into 23
+ * languages; leaving it English keeps the text intact until the parser is fixed.
+ *
+ * Whitespace is ignored because a faithful re-render legitimately re-spaces
+ * punctuation (`{name: 'Demo'}` → `{ name : 'Demo' }`). The known limitation:
+ * whitespace INSIDE a string literal is ignored too, so a body differing only
+ * by spaces inside quotes would be accepted — narrow, and it cannot drop
+ * content, only re-space it.
+ */
+export function reRenderPreservesContent(source: string, reRenderedEnglish: string): boolean {
+  const strip = (s: string) => s.replace(/\s+/g, '');
+  return strip(source) === strip(reRenderedEnglish);
+}

package/src/types/index.ts

@@ -55,7 +55,30 @@ export type ComplexityLevel =
 // =============================================================================
 
 export type WordOrder = 'SVO' | 'SOV' | 'VSO' | 'V2';
-export type TranslationMethod = 'auto-generated' | 'hand-crafted' | 'verified';
+/**
+ * How a `pattern_translations` row was produced. The first three are the
+ * values `sync-translations` writes today; the rest are legacy/aspirational
+ * labels kept so existing rows and `getTranslationPriority` keep typing.
+ *
+ * - `semantic-render`: @lokascript/semantic `render(parse_en(en), L)` chosen by
+ *   the `best` writer (or the `semantic` writer).
+ * - `grammar-transform`: @lokascript/i18n GrammarTransformer — under the `best`
+ *   writer, a row i18n still renders better than semantic (the kept-row ratchet).
+ * - `grammar-transform-no-reference`: `best` only — the i18n row because semantic
+ *   cannot parse the ENGLISH source (a parser-coverage gap, not a render loss).
+ * - `keyword-substitute`: word-for-word fallback for a language with no grammar profile.
+ * - `original`: the English row; `non-translatable-identity`: markup rows copied verbatim.
+ */
+export type TranslationMethod =
+  | 'semantic-render'
+  | 'grammar-transform'
+  | 'grammar-transform-no-reference'
+  | 'keyword-substitute'
+  | 'original'
+  | 'non-translatable-identity'
+  | 'auto-generated'
+  | 'hand-crafted'
+  | 'verified';
 
 /**
  * Pattern priority levels for translation selection.