@adia-ai/a2ui-compose 0.5.5 → 0.5.6

package/CHANGELOG.md

@@ -7,11 +7,46 @@ transpiler, evals. Retrieval + validation now ship as sibling
 packages (`@adia-ai/a2ui-retrieval`, `@adia-ai/a2ui-validator`) so
 non-compose consumers can depend on them without pulling the
 generator graph.
-
 ## [Unreleased]
 
 _No pending changes._
-
+## [0.5.6] - 2026-05-14
+
+### Removed — §195 (v0.5.6) — retired fragment-graph iteration codepath
+
+Deleted `packages/a2ui/compose/strategies/zettel/synthesizer.js`. The
+file shipped the `synthesizeComposition` function as a throwing stub
+since §37 (v0.4.7, 2026-05-12, fragment retirement). The dangling
+caller in `generator-adapter.js` (iteration branch, turn ≥ 2 + LLM
+present) caught the throw and auto-fired an `iteration-synthesis-failure`
+issue ticket on every multi-turn refinement — surfaced as an auto-fired
+bug at `.brain/audit-history/issues/2026-05-14-iteration-synthesis-failed-on-turn-3-for-4bfb.json`.
+
+### Changed — §195 (v0.5.6) — weak-retrieval branch bridges to chunk-zettel
+
+Replaced the `synthesizeComposition()` call in `generateZettel`'s
+weak-retrieval-with-LLM branch with a bridge to
+`chunk-synthesizer.js::composeFromIntent` (the §37 successor codepath).
+Return shape is preserved (`messages`, `validation`, `strategy:
+'composition-synthesized'`, `fragments_used: []`, `synthesized_template`,
+`synthesis`). Fragment instances are no longer tracked on this path —
+the chunk-bridge emits a single html-bearing component. Consumers that
+need fragment-level resolution should wire via `engine: 'chunk-zettel'`
+directly.
+
+The `composition-iterated` strategy is retired — turn ≥ 2 on
+`engine: 'zettel'` now takes the same path as turn 1. True
+history-aware iteration lives in `chunk-refiner.js::refineFromIntent`
+behind `engine: 'chunk-zettel'`.
+
+### Added — §195 (v0.5.6) — smoke probe
+
+NEW `scripts/smoke-iteration-synthesis.mjs` (npm run
+`smoke:iteration-synthesis`) verifies the §195 invariants at three
+layers: filesystem (`synthesizer.js` deleted), source
+(`generator-adapter.js` has no surviving reference to the retired
+identifiers), and runtime (4 sequential turns against the same
+sessionId neither throw nor emit `composition-iterated`).
 ## [0.5.5] - 2026-05-14
 
 ### Changed — §179 (v0.5.5) — principled `deprecated:` yaml field
@@ -27,7 +62,6 @@ Build pipeline (`scripts/build/components.mjs::compileProp`) collects deprecatio
 Demonstrated on `Avatar.name`. Verified via `buildSystemPrompt`: `name:String[] (deprecated, do not use)` surfaces in CORPUS CONTEXT.
 
 Commit `0379dd498`. See root CHANGELOG and journal §179 for full context.
-
 ## [0.5.4] - 2026-05-14
 
 ### Fixed — §163 transpiler prop-fidelity; catalog-driven extractor closes the harvester drop class (v0.5.4)
@@ -201,7 +235,6 @@ No compose source change in §173/§175 — both detect future drift in compose'
 catalog-vs-prompt + registry-vs-catalog invariants.
 
 ### Changed — `version`: `0.5.3` → `0.5.4`.
-
 ## [0.5.3] - 2026-05-14
 
 ### Fixed — Deterministic chunk-loading order in zettel composition library (§160, v0.5.3)
@@ -224,7 +257,6 @@ Finalizes the v0.6.0 deprecation schedule for the last two `_debug.*` fields tha
 **Scheduled removal**: v0.6.0 drops the `_debug` block entirely from the free-form-composed result shape. The dialog-recorder will read first-class fields directly. v0.5.3 is the **migration window** — any external consumers reading `_debug.attempts` / `_debug.warnings` should switch to the first-class fields before v0.6.0 ships.
 
 **Internal verification**: zero live in-repo consumers (`grep -rn '_debug\?\.attempts\|_debug\.attempts\|_debug\?\.warnings\|_debug\.warnings' apps/ playgrounds/ catalog/ packages/` returns only the deprecation comment itself).
-
 ## [0.5.2] - 2026-05-13
 
 ### Changed — `plan` graduates from `_debug.*` to first-class on free-form-composed result (§107a infra, v0.5.2)
@@ -259,7 +291,6 @@ Expected impact: lifts substitution ratio 27.4% → ~35-40% (target ≥30%); F1
 `strategies/registry.js` reads model priority chain at call-time (not module-load): `ctx.model` > `process.env.FREE_FORM_MODEL_OVERRIDE` > `FREE_FORM_MODEL_DEFAULT` (Haiku 4.5). Enables `eval-diff.mjs --model <id>` to run the Haiku-vs-Opus A/B for §127 without env-dance or static-import-ordering hazards. `FREE_FORM_MODEL` constant renamed to `FREE_FORM_MODEL_DEFAULT` to reflect the new priority chain.
 
 Default behavior unchanged when no override set — the v0.5.1 §108 Haiku pin holds for every consumer that doesn't explicitly override.
-
 ## [0.5.1] - 2026-05-13
 
 ### Added — Free-form composer INTENT-PARAPHRASE block + paraphrase-retry (§106, v0.5.1)
@@ -281,7 +312,6 @@ Consumers passing `ctx.llmAdapter` keep getting their adapter via the mint-failu
 ### Coverage at v0.5.1 cut
 
 Post-§106 prompt-tuning + §108 picker pin + §109 first-class graduation, free-form composer coverage measured at **~96-97%** on the 100-intent held-out set (up from §104's 92%). New AGENTS.md regression threshold floor: `cov≥96%, avg≥85, F1≥0.60` (§115 trip-wire baseline).
-
 ## [0.5.0] - 2026-05-13
 
 ### Added — Free-form composer auto-grouping (§103, v0.5.0)
@@ -299,17 +329,14 @@ Schema validation rejects unknown target keys; downgrade warnings fire when a ta
 ### Coverage at v0.5.0 cut
 
 After §93 (layout regrowth) + §94 (forms regrowth) + §103 (auto-grouping) + §104 (structural substitutions + v0.5.1 deferred regrowth fold-in), free-form composer coverage measured at **92% on the 100-intent held-out set** (up from 21% at §92 baseline). New AGENTS.md regression threshold floor: `cov≥80%, avg≥85, F1≥0.45` (§97 rebaseline).
-
 ## [0.4.9] - 2026-05-13
 
 _No pending changes._
-
 ## [0.4.8] - 2026-05-12
 
 ### Fixed — `strategies/zettel/generator-adapter.js` `ensureBooted()` race (§87a, v0.4.8)
 
 The zettel composition library's lazy boot had a race condition where concurrent `ensureBooted()` calls could each kick off a separate boot promise. Under contention (multiple requests landing simultaneously at server cold-start), the composition map could be partially populated when consumers reached the synchronous getters. Fix: memoize the boot promise on first call so all subsequent callers await the same promise. Pairs with the v0.4.8 §87 honest-floor eval threshold rebaseline — without the boot race fix, the 1% rebaseline was actually undercounting due to occasional empty-map reads.
-
 ## [0.4.7] - 2026-05-12
 
 ### Changed — `strategies/monolithic/_shared.js` `getComponentCatalog()` reads canonical catalog (§72)
@@ -317,7 +344,6 @@ The zettel composition library's lazy boot had a race condition where concurrent
 The legacy reader of `@adia-ai/a2ui-corpus/patterns/_components.json` has been migrated to read `@adia-ai/a2ui-corpus/catalog-a2ui_0_9.json` (the canonical v0.9 catalog, already the package root export). Per-component aliases are now lifted from `components[name].x-adiaui.synonyms.tags`. Same legacy output shape (`{ <Name>: { aliases: string[] } }`); zero behavior change for downstream callers.
 
 Closes the §65 carry-over from v0.4.6 — `@adia-ai/a2ui-corpus` `patterns/` + `compositions/` are now deleted from disk + tarball.
-
 ## [0.4.6] - 2026-05-12
 
 ### Changed — `core/` retirement follow-through (§64, v0.4.6)
@@ -353,7 +379,6 @@ Closes the catalog-drift surface that §56 explicitly deferred ("different schem
 5. `const` fields (e.g. `{const: 'Button'}`) are discriminator-only. Skip.
 
 Source: `packages/a2ui/compose/strategies/monolithic/_shared.js` (+91, -2 lines, commit `afda98f5`).
-
 ## [0.4.5] - 2026-05-12
 
 ### Changed — GenUI overhaul prompt-engineering (§56, v0.4.5)
@@ -369,7 +394,6 @@ Source: `packages/a2ui/compose/strategies/monolithic/_shared.js` (+91, -2 lines,
 - **`strategies/monolithic/generate-pro.js` — STRUCTURAL REFERENCE prose enriched.** Prior copy: "a real production block from the codebase matched this intent." New copy: "this chunk was retrieved from the AdiaUI training corpus (annotated production HTML, harvested from real app pages). It matched your intent on keyword/domain ranking." Threads chunk `metadata.domain`, `metadata.description`, `metadata.keywords` into the prompt when present. Reframes "do not copy the HTML" as "the chunk represents the SHAPE the user wants; instantiate it with their content" — more actionable framing.
 
 See root [CHANGELOG.md `[Unreleased]`](../../../CHANGELOG.md) for the v0.4.5 overhaul arc + apps/genui/CHANGELOG.md `[Unreleased]` for the per-§ rollup.
-
 ## [0.4.4] - 2026-05-12
 
 ### Changed
@@ -386,7 +410,6 @@ See root [CHANGELOG.md `[Unreleased]`](../../../CHANGELOG.md) for the v0.4.5 ove
 - **MCP server `mcp/server.js` (§37).** Removed `get_fragment` tool, simplified `zettel_stats`, updated imports + boot logs + engine description to reflect the post-fragment world.
 
 See root [CHANGELOG.md `[Unreleased]`](../../../CHANGELOG.md) for the cross-cutting arc narrative + [docs/journal/2026/05/2026-05-12.md](../../../docs/journal/2026/05/2026-05-12.md) §§ 37 / 38 / 41 / 47 for per-§ details.
-
 ## [0.4.3] - 2026-05-11
 
 ### Fixed
@@ -396,7 +419,6 @@ See root [CHANGELOG.md `[Unreleased]`](../../../CHANGELOG.md) for the cross-cutt
 ### Lockstep
 
 9-package coordinated PATCH cut to v0.4.3 (per [`docs/specs/package-architecture.md` § 15](../../../docs/specs/package-architecture.md#15-versioning-policy)). Internal `@adia-ai/*` dep ranges stay at `^0.4.0` (patch-cut asymmetry — `^0.4.0` covers `0.4.x` under semver). Source change scoped to `core/generator.js` + `strategies/zettel/state-cache.js`. Rides alongside `@adia-ai/web-components` v0.4.3 (input-ui locale + thousands grouping + hold-to-repeat). See root [CHANGELOG.md `## [0.4.3]`](../../../CHANGELOG.md) for the cut narrative.
-
 ## [0.4.2] - 2026-05-11
 
 ### Ride-along (no source changes)
@@ -404,7 +426,6 @@ See root [CHANGELOG.md `[Unreleased]`](../../../CHANGELOG.md) for the cross-cutt
 Lockstep PATCH cut alongside `@adia-ai/web-components@0.4.2` (`<input-ui type="number">` rewrite drops native `<input type=number>` wrapping) + `@adia-ai/web-modules@0.4.2` (`<editor-sidebar>` grid-track width-mirror fix). Source byte-identical to v0.4.1.
 
 Internal `@adia-ai/*` dep ranges stay at `^0.4.0` (patch-cut asymmetry — `^0.4.0` covers `0.4.x` under semver). See root [CHANGELOG.md `## [0.4.2]`](../../../CHANGELOG.md) for the cut narrative.
-
 ## [0.4.1] - 2026-05-10
 
 ### Ride-along (no source changes)
@@ -412,7 +433,6 @@ Internal `@adia-ai/*` dep ranges stay at `^0.4.0` (patch-cut asymmetry — `^0.4
 Lockstep PATCH cut alongside `@adia-ai/web-modules@0.4.1` (simple cluster) + `@adia-ai/a2ui-validator@0.4.1` (Phase 3 foundation) + `@adia-ai/a2ui-corpus@0.4.1` (fragment metrics reconciliation). Source byte-identical to v0.4.0.
 
 Internal `@adia-ai/*` dep ranges bumped from `^0.4.0` to `^0.4.1`. See root [CHANGELOG.md `## [0.4.1]`](../../../CHANGELOG.md) for the cut narrative.
-
 ## [0.4.0] - 2026-05-10
 
 ### Ride-along (no source changes)
@@ -420,25 +440,21 @@ Internal `@adia-ai/*` dep ranges bumped from `^0.4.0` to `^0.4.1`. See root [CHA
 Lockstep MINOR cut alongside `@adia-ai/web-modules@0.4.0` (ADR-0024 legacy shell shapes retired). Source byte-identical to v0.3.6.
 
 Internal `@adia-ai/*` dep ranges bumped from `^0.3.0` to `^0.4.0`. See root [CHANGELOG.md `## [0.4.0]`](../../../CHANGELOG.md) for the cut narrative.
-
 ## [0.3.6] - 2026-05-10
 
 ### Ride-along (no source changes)
 
 Lockstep version bump only — source byte-identical to v0.3.5. Internal `@adia-ai/*` dep ranges remain at `^0.3.0`. See root [CHANGELOG.md `## [0.3.6]`](../../../CHANGELOG.md) for the cut narrative.
-
 ## [0.3.5] - 2026-05-07
 
 ### Ride-along (no source changes)
 
 Lockstep version bump only — source byte-identical to v0.3.4. Internal `@adia-ai/*` dep ranges remain at `^0.3.0`. See root [CHANGELOG.md `## [0.3.5]`](../../../CHANGELOG.md) for the cut narrative.
-
 ## [0.3.4] - 2026-05-07
 
 ### Ride-along (no source changes)
 
 Lockstep version bump only — source byte-identical to v0.3.3. Internal `@adia-ai/*` dep ranges remain at `^0.3.0`. See root [CHANGELOG.md `## [0.3.4]`](../../../CHANGELOG.md) for the cut narrative.
-
 ## [0.3.3] - 2026-05-07
 
 **Lockstep cut.** All 9 published `@adia-ai/*` packages now share version `0.3.3`, governed by [`docs/specs/package-architecture.md` § 15](../../../docs/specs/package-architecture.md#15-versioning-policy). Internal `@adia-ai/*` ranges stay at `^0.3.0` (patch-cut asymmetry — caret floats `0.3.x`).
@@ -483,7 +499,6 @@ Lockstep version bump only — source byte-identical to v0.3.3. Internal `@adia-
   for the lifetime of the process. Trade-offs: good for long-running
   MCP, bad for tests/hot-reload. To force a reload, call `loadAll()`
   directly. (closes backlog #100)
-
 ## [0.3.2] - 2026-05-06
 
 **9-package lockstep patch cut to v0.3.2.** All lockstep members share
@@ -510,7 +525,6 @@ Internal `@adia-ai/*` dep ranges unchanged at `^0.3.0`.
 ### Changed
 
 - `version`: `0.3.1` → `0.3.2`.
-
 ## [0.3.1] - 2026-05-06
 
 **9-package lockstep patch cut.** All 9 published `@adia-ai/*` packages bump 0.3.0 → 0.3.1 per [`docs/specs/package-architecture.md` § 15](../../../docs/specs/package-architecture.md#15-versioning-policy). Internal `@adia-ai/*` dep ranges remain at `^0.3.0` (covers `0.3.1` under semver — patch-cut asymmetry).
@@ -521,7 +535,6 @@ This package itself ships **no source changes** in v0.3.1. The cut bumps version
 
 - `version`: `0.3.0` → `0.3.1`.
 - Internal `@adia-ai/*` dep ranges: unchanged at `^0.3.0` (covers `0.3.1` under semver — patch-cut asymmetry).
-
 ## [0.3.0] - 2026-05-05
 
 **9-package lockstep cut + LLM subpath dropped.** All 9 published `@adia-ai/*` packages bump 0.2.5 → 0.3.0 per [`docs/specs/package-architecture.md` § 15](../../../docs/specs/package-architecture.md#15-versioning-policy). Internal `@adia-ai/*` dep ranges bump `^0.2.0` → `^0.3.0`.
@@ -567,7 +580,6 @@ This is a **minor cut on top of v0.2.5 with one BREAKING change**: the `./llm` s
 ```
 
 Plus rewrite imports as shown above. The change is mechanical.
-
 ## [0.2.5] - 2026-05-04
 
 **8-package lockstep cut.** All 8 published `@adia-ai/*` packages bump 0.2.4 → 0.2.5 per [`docs/specs/package-architecture.md` § 15](../../../docs/specs/package-architecture.md#15-versioning-policy). Internal `@adia-ai/*` dep ranges remain at `^0.2.0` (covers 0.2.5 under semver — patch-cut asymmetry).
@@ -578,7 +590,6 @@ This is a **patch cut on top of v0.2.4, no BREAKING changes.** Substantive conte
 
 - `version`: `0.2.4` → `0.2.5`.
 - Internal `@adia-ai/*` dep ranges: unchanged at `^0.2.0` (covers `0.2.5` under semver — patch-cut asymmetry).
-
 ## [0.2.4] - 2026-05-04
 
 **8-package lockstep cut.** All 8 published `@adia-ai/*` packages bump 0.2.3 → 0.2.4 per [`docs/specs/package-architecture.md` § 15](../../../docs/specs/package-architecture.md#15-versioning-policy). Internal `@adia-ai/*` dep ranges remain at `^0.2.0` (covers 0.2.4 under semver — patch-cut asymmetry).
@@ -597,7 +608,6 @@ _Nothing yet._
 ---
 
 ---
-
 ## [0.2.3] - 2026-05-04
 
 **Lockstep cut.** All 8 published `@adia-ai/*` packages bump
@@ -615,7 +625,6 @@ Patch cut — no breaking changes.
 ### No source changes
 
 `@adia-ai/a2ui-compose` source is byte-identical to `0.2.2`. The cut bumps version + the internal dep range entries only.
-
 ## [0.2.2] - 2026-05-02
 
 **Lockstep cut + reasoning-panel emission fixes.** All 8 published `@adia-ai/*` packages bump 0.2.1 → 0.2.2 per [`docs/specs/package-architecture.md` § 15](../../../../docs/specs/package-architecture.md#15-versioning-policy). Patch cut — no breaking changes.
@@ -639,7 +648,6 @@ Patch cut — no breaking changes.
 Companion fix in `@adia-ai/a2ui-retrieval` Unreleased (web-research.js EXPLICIT/IMPLICIT pattern split). Both surfaces feed the same reasoning-panel deception class — five distinct bugs caught + fixed in commit `9986c71e`.
 
 ---
-
 ## [0.2.1] - 2026-05-02
 
 **Lockstep cut + scope-drift gate at composer time + skeleton harvest + Tier-1 block filter + high-resolution session ticket trace.** All 8 published `@adia-ai/*` packages bump 0.2.0 → 0.2.1 per [`docs/specs/package-architecture.md` § 15](../../../../docs/specs/package-architecture.md#15-versioning-policy). Patch cut — no breaking changes.
@@ -678,7 +686,6 @@ The synthesizer now captures a `retrievalTrace` per attempt: Tier-1 hits with sc
 `issue-reporter.js` renders this as a sibling Markdown ticket alongside the JSON; sections cover header, description, reproduction, component count, retrieval log table, LLM attempts (raw responses), user prompt, composer plan, generated HTML preview, warnings, ops history, environment. A maintainer can replay any flagged session from the ticket alone.
 
 ---
-
 ## [0.2.0] - 2026-05-02
 
 **Lockstep cut + boundary cleanup.** All 8 published `@adia-ai/*`
@@ -733,7 +740,6 @@ cycle, but should update on next touch:
 ```
 
 ---
-
 ## [0.1.0] - 2026-04-28
 
 **Multi-turn refinement engine (Phase A).** Adds three new modules to the
@@ -820,7 +826,6 @@ Additive surface; no breaking changes. Existing consumers calling
 `composeFromIntent` continue to work unchanged.
 
 ---
-
 ## [0.0.1] - 2026-04-24
 
 First public release. Framework-agnostic compose engine for the
@@ -879,7 +884,6 @@ behind a plug-in engine registry.
 - Stale internal-identifier references in comments + doc strings swept to the current naming.
 
 ---
-
 ## [0.1.0] — internal baseline (unreleased)
 
 Initial version at the time the monorepo was established. Contains:

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@adia-ai/a2ui-compose",
-  "version": "0.5.5",
+  "version": "0.5.6",
   "description": "AdiaUI A2UI compose engine \u2014 framework-agnostic. Takes natural-language intents + a catalog and produces A2UI protocol messages. Pairs with `@adia-ai/a2ui-retrieval` (intent classification, catalog lookup) and `@adia-ai/a2ui-validator` (schema + semantic checks).",
   "type": "module",
   "exports": {

package/strategies/monolithic/_shared.js

@@ -903,6 +903,15 @@ export function mergeCanvasDiff(priorComponents, diffComponents) {
   const modifiedIds = new Set();
   const newComponents = [];
 
+  // §190b (v0.5.6) — Aggregate blocked-layout-type-change warnings.
+  // The for-loop below can fire `console.warn` once per blocked component,
+  // producing 5-10 line bursts when the LLM tries to restructure a form's
+  // fields (typical case: 5 Field→Column blocks for a 5-field form).
+  // Collect blocks into an array; emit a single aggregated warning after
+  // the loop. Closes the 2026-05-14 user-reported "buglabels" intent
+  // emitting 6 consecutive blocks.
+  const blockedLayoutChanges = [];
+
   for (const dc of diffComponents) {
     if (!dc || !dc.id) continue;
 
@@ -920,7 +929,7 @@ export function mergeCanvasDiff(priorComponents, diffComponents) {
       const LAYOUT_CONTAINERS = new Set(['Grid', 'Row', 'Column', 'Stack']);
       if (dc.component && dc.component !== existing.component &&
           (LAYOUT_CONTAINERS.has(existing.component) || LAYOUT_CONTAINERS.has(dc.component))) {
-        console.warn(`[mergeCanvasDiff] Blocked layout type change: ${existing.component}→${dc.component} on id="${dc.id}". Preserving original type.`);
+        blockedLayoutChanges.push({ id: dc.id, from: existing.component, to: dc.component });
         delete dc.component;
       }
       const merged = { ...existing, ...dc };
@@ -934,6 +943,14 @@ export function mergeCanvasDiff(priorComponents, diffComponents) {
     }
   }
 
+  // §190b — single aggregated warning for blocked layout changes.
+  if (blockedLayoutChanges.length > 0) {
+    const summary = blockedLayoutChanges
+      .map(b => `${b.from}→${b.to} on id="${b.id}"`)
+      .join(', ');
+    console.warn(`[mergeCanvasDiff] Blocked ${blockedLayoutChanges.length} layout type change${blockedLayoutChanges.length === 1 ? '' : 's'}: ${summary}. Preserving original types.`);
+  }
+
   // Build result: prior components (minus deleted) + new components
   const result = [];
   for (const c of priorComponents) {

package/strategies/monolithic/merge-canvas-diff.test.js

@@ -197,8 +197,12 @@ describe('§168 mergeCanvasDiff — iteration handoff invariants', () => {
       const result = mergeCanvasDiff(prior, llmOutput);
       // Grid type preserved despite LLM's attempted change
       expect(result.find(c => c.id === 'root').component).toBe('Grid');
+      // §190b — warning is aggregated: "Blocked N layout type change(s): …"
       expect(warn).toHaveBeenCalledWith(
-        expect.stringContaining('Blocked layout type change')
+        expect.stringContaining('Blocked 1 layout type change')
+      );
+      expect(warn).toHaveBeenCalledWith(
+        expect.stringContaining('Grid→Column on id="root"')
       );
       warn.mockRestore();
     });

package/strategies/zettel/generator-adapter.js

@@ -3,35 +3,66 @@
  *
  *   generate({ intent, mode, llmAdapter, sessionId }) -> { messages, validation, strategy, ...extra }
  *
- * Three reasoning layers:
+ * Two reasoning layers (post-§195, v0.5.6):
  *   1. Retrieval (always available) — keyword-rank the corpus, resolve top composition.
- *   2. LLM synthesis (when llmAdapter provided AND retrieval weak) — compose a new
- *      composition from fragments, have the composer resolve it into A2UI messages.
- *   3. Session-aware iteration (when sessionId provided AND prior turns exist) —
- *      pass prior-turn history to the LLM so follow-ups ("add a button", "hydrate
- *      with real images") modify the existing canvas instead of regenerating.
+ *   2. LLM synthesis (when llmAdapter provided AND retrieval weak) — bridge to
+ *      chunk-zettel's `composeFromIntent` (the §37 successor codepath) which
+ *      produces a single-shot html composition from the chunk corpus.
+ *
+ * Session iteration is currently NOT history-aware in this engine. Prior turns
+ * are recorded for analytics + drift tracking, but turn≥2 takes the same code
+ * path as turn 1 (fresh retrieval → fresh synthesis). Full history-aware
+ * iteration (multi-turn refinement modifying an existing canvas) lives in the
+ * `chunk-zettel` engine via `chunk-refiner.js::refineFromIntent` — wire it via
+ * `engine: 'chunk-zettel'` rather than `engine: 'zettel'`.
+ *
+ * The prior fragment-graph iteration codepath (`synthesizeComposition` with
+ * `historySummary`) was retired in §37 (2026-05-12) when fragments retired.
+ * §195 (v0.5.6) cleaned up the dangling caller — turn≥2 no longer throws +
+ * auto-fires a `iteration-synthesis-failure` issue on every multi-turn turn.
  *
  * Strategy labels in the return:
  *   - composition-match          — fresh retrieval, strong match, emitted verbatim
- *   - composition-synthesized    — fresh LLM composition (no prior turns)
- *   - composition-iterated       — LLM modified prior turn's template
+ *   - composition-synthesized    — chunk-zettel single-shot synthesis
  *   - fragment-candidates        — retrieval weak + no LLM, returning atoms only
- *   - synthesis-failed           — LLM tried and failed validation
+ *   - synthesis-failed           — chunk-zettel tried and failed validation
  */
 import {
   getComposition,
   searchAll,
 } from './composition-library.js';
 import { resolveComposition, templateToMessages } from './composer.js';
-import { synthesizeComposition } from './synthesizer.js';
 import {
   recordTurn,
   getTurns,
-  buildHistorySummary,
 } from './session-store.js';
-import { autoReport } from './issue-reporter.js';
 import { validateSchema } from '../../../validator/validator.js';
 
+// Lazy-load the chunk-synthesizer bridge. It imports chunk-corpus data
+// via composition-library's dual-mode loader, which is already top-level
+// awaited; importing it here statically would not break anything but the
+// async-import keeps the cold-start surface tight for the strong-match
+// (no-LLM) path that doesn't need synthesis at all.
+async function bridgeToChunkSynthesis({ intent, llmAdapter }) {
+  const { composeFromIntent } = await import('./chunk-synthesizer.js');
+  const result = await composeFromIntent({ intent, llmAdapter, maxAttempts: 2 });
+  // Shape-adapt chunk-synthesizer's `{ html, plan, source, ... }` to the
+  // synthesis-branch's prior `{ messages, template, synthesis }` contract.
+  const messages = result.html
+    ? [{ type: 'updateComponents', components: [{ id: 'chunk-root', component: 'article', html: result.html }] }]
+    : [];
+  return {
+    messages,
+    template: result.plan ? [{ $chunk: 'composeFromIntent', plan: result.plan }] : [],
+    synthesis: {
+      source: result.source,
+      score: result.score,
+      warnings: result.warnings || [],
+      scopeDrift: result.scopeDrift || null,
+    },
+  };
+}
+
 // Composition library auto-loads at module import time via top-level `await
 // loadAll()` in composition-library.js (§72 dual-mode loader, commit
 // `76dbcff2`). The previous `ensureBooted()` here called `loadAll()` WITHOUT
@@ -72,62 +103,17 @@ function toUpdateComponentsMessages(template) {
 }
 
 export async function generateZettel({ intent, mode = 'instant', llmAdapter = null, sessionId = null } = {}) {
-  // ── Session-aware iteration (turn > 1) ──
-  // If we have prior turns AND an LLM, the user is almost certainly modifying
-  // the existing canvas — NEVER pick a fresh retrieved composition. Go straight
-  // to synthesis with history context. This is what makes follow-ups work.
+  // ── Session iteration note (post-§195, v0.5.6) ──
+  // Fragment-graph history-aware iteration (`synthesizeComposition` with
+  // `historySummary`) was retired in §37 when fragments retired. Turn≥2 now
+  // takes the same path as turn 1 — fresh retrieval → fresh chunk-synthesis.
+  // For true history-aware iteration (modify-an-existing-canvas), wire the
+  // request via `engine: 'chunk-zettel'` which uses `chunk-refiner.js`.
+  //
+  // Prior turns are still recorded (analytics + drift tracking via `getDrift`)
+  // but the iteration BRANCH is gone — no more spurious
+  // `iteration-synthesis-failure` auto-fires on every multi-turn request.
   const priorTurns = sessionId ? getTurns(sessionId) : [];
-  const hasHistory = priorTurns.length > 0;
-
-  if (hasHistory && llmAdapter) {
-    try {
-      const historySummary = buildHistorySummary(sessionId, 3);
-      const synth = await synthesizeComposition({ intent, llmAdapter, historySummary });
-      const validation = validateSchema(synth.messages, { intent });
-      const fragments = (synth.template || []).filter((n) => n.$fragment).map((n) => n.$fragment);
-      recordTurn(sessionId, {
-        intent,
-        messages: synth.messages,
-        template: synth.template,
-        composition: null,
-        strategy: 'composition-iterated',
-        fragments,
-        validation,
-      });
-      return {
-        messages: synth.messages,
-        validation,
-        strategy: 'composition-iterated',
-        retrieval: { hit: false, rank: null, candidate: null, reason: `iteration on turn ${priorTurns.length + 1}` },
-        composition: null,
-        fragments_used: fragments,
-        synthesized_template: synth.template,
-        synthesis: synth.synthesis,
-        sessionTurns: priorTurns.length + 1,
-      };
-    } catch (err) {
-      // If iteration synthesis fails, fall through to the normal path. Record
-      // the failure so the next turn can see we tried, and auto-fire an issue
-      // ticket so synthesis-owners can investigate the failure post-hoc.
-      console.error('[zettel] iteration synthesis failed:', err.message);
-      try {
-        await autoReport(
-          'iteration-synthesis-failure',
-          {
-            intent,
-            turn: priorTurns.length + 1,
-            state_id: sessionId,
-            body: `Auto-fired by generator-adapter. Iteration synthesis threw on turn ${priorTurns.length + 1}.\n\nError: \`${err.message}\``,
-            tags: ['generator-adapter', `turn-${priorTurns.length + 1}`],
-          },
-          { evalMode: mode === 'eval' }
-        );
-      } catch (reportErr) {
-        // Never let issue-reporting crash the request path.
-        console.error('[zettel] autoReport failed:', reportErr.message);
-      }
-    }
-  }
 
   const hits = searchAll(intent, { limit: 5 });
   const composition = hits.find((h) => h.type === 'composition');
@@ -163,19 +149,22 @@ export async function generateZettel({ intent, mode = 'instant', llmAdapter = nu
     };
   }
 
-  // ── Weak/no retrieval: try LLM synthesis if available (fresh, no history) ──
+  // ── Weak/no retrieval: bridge to chunk-zettel synthesis (the §37 successor) ──
   if (llmAdapter && mode !== 'instant-only') {
     try {
-      const synth = await synthesizeComposition({ intent, llmAdapter });
+      const synth = await bridgeToChunkSynthesis({ intent, llmAdapter });
       const validation = validateSchema(synth.messages, { intent });
-      const fragments = (synth.template || []).filter((n) => n.$fragment).map((n) => n.$fragment);
+      // Chunk-bridge produces an html-bearing single component, not
+      // resolved fragment instances. `fragments_used` is empty by design;
+      // consumers tracking fragment usage should switch to the
+      // chunk-zettel engine which surfaces `_debug.plan` directly.
       recordTurn(sessionId, {
         intent,
         messages: synth.messages,
         template: synth.template,
         composition: null,
         strategy: 'composition-synthesized',
-        fragments,
+        fragments: [],
         validation,
       });
       return {
@@ -189,7 +178,7 @@ export async function generateZettel({ intent, mode = 'instant', llmAdapter = nu
           reason: composition ? `top score ${composition.score} below threshold ${STRONG_MATCH_THRESHOLD}` : 'no composition retrieved',
         },
         composition: null,
-        fragments_used: fragments,
+        fragments_used: [],
         synthesized_template: synth.template,
         synthesis: synth.synthesis,
         candidates: hits,

package/strategies/zettel/synthesizer.js

@@ -1,22 +0,0 @@
-/**
- * Composition synthesizer — RETIRED (§37, 2026-05-12).
- *
- * Previously: when zettel retrieval was weak, this called the LLM to
- * assemble a NEW composition from the fragment catalog (technique B/C
- * fragment-graph synthesis). Fragments are retired and there's no
- * fragment catalog to draw from anymore. The deterministic path
- * (`resolveComposition` on a retrieved composition) still works.
- *
- * Until a chunk-based synthesis fallback lands, this export throws on
- * call. Callers should catch and fall through to monolithic-pro's
- * generation path. generator-adapter.js handles this gracefully.
- */
-
-export async function synthesizeComposition() {
-  const err = new Error(
-    'synthesizeComposition is retired (§37 fragment retirement). ' +
-    'Use chunk-zettel or monolithic-pro for LLM-driven composition.'
-  );
-  err.code = 'SYNTHESIZER_RETIRED';
-  throw err;
-}