docs(authority): keep README.md, GitBook EN/DE and DeepWiki continuously accurate and source-truth aligned

[qnbs]

Purpose

Establish one long-lived public-documentation authority and parity owner for WorldScript Studio.

This issue owns the continuous reconciliation of:

README.md
        ↕
GitBook — English
https://qnbs.gitbook.io/worldscript-studio/
        ↕
GitBook — Deutsch
https://qnbs.gitbook.io/worldscript-studio/de/
        ↕
DeepWiki repository view
https://deepwiki.com/qnbs/WorldScript-Studio
        ↕
CURRENT REPOSITORY / APP TRUTH

It is intentionally not a one-shot README rewrite issue. The initial comprehensive README/GitBook convergence is only the bootstrap. This issue should remain open as the durable owner for keeping the public documentation accurate while the application evolves.

Do not close it merely because one README/GitBook refresh PR lands. Close only if an equivalent or stronger automated/document-authority mechanism explicitly supersedes it.

---

Why a dedicated durable owner is needed

WorldScript Studio now has several public documentation surfaces with different roles:

• README.md — repository/product entry point;
• GitBook EN — polished reader-facing English documentation;
• GitBook DE — polished reader-facing German documentation;
• DeepWiki — code-derived repository exploration/documentation surface;
• repository docs / ADRs / current source / configuration / CI — engineering and implementation truth.

Without a durable reconciliation owner, these surfaces can independently drift after ordinary feature, security, AI, storage, export, PWA, desktop, native, accessibility, i18n or infrastructure changes.

Recent drift examples already found include:

• README describing GitHub Pages and Vercel as effectively identical despite materially different host capabilities;
• overly broad nothing leaves your device / everything works offline style claims;
• stale AI-mode semantics for Hybrid/Eco;
• feature-flag count/default drift (15 on / 8 off in current source versus older prose saying 16 / 7);
• evolving IDB encryption/rekey/recovery behavior;
• desktop filesystem at-rest-encryption limitations;
• Claude direct-vs-relay-vs-unavailable host differences;
• local AI / browser Ollama / local-server capability distinctions;
• Qt/GPUI roadmap prose that can easily outrun actual authority state;
• test/i18n metrics that must be source-synchronized rather than hand-maintained.

This issue owns the cross-surface documentation truth, not the underlying implementation work.

---

1. Documentation authority hierarchy

Use this hierarchy whenever documentation sources disagree:

CURRENT PRODUCTION SOURCE
+ CURRENT TESTS
+ CURRENT CONFIG / CI
+ CURRENT ADR / SECURITY ARCHITECTURE DOCS
        ↓
README.md
        ↓
GitBook EN / DE
        ↓
DeepWiki generated/code-derived presentation
        ↓
ROADMAP / TODO snapshots
        ↓
HISTORICAL sprint/audit/handoff material

Nuance:

• README and GitBook should be actively curated to match current source truth.
• DeepWiki is a useful code-derived public view, but generated interpretation must not override current source/ADRs/security authority.
• ROADMAP/TODO remain important planning sources but must not be treated as stronger implementation evidence than current code and admitted runtime behavior.

Cross-reference #559 for roadmap/native/Core planning-truth reconciliation and #549 for security/privacy truth.

---

2. Role of each public surface

README.md

README should remain the high-value product/repository entry point:

• concise but substantial overview;
• current product philosophy;
• local-first/offline-capable truth boundaries;
• primary feature tour;
• AI execution overview;
• privacy/security trust boundaries;
• platform capability summary;
• current stack;
• developer quick start;
• deployment/runtime distinctions;
• major known limitations;
• roadmap boundaries;
• prominent links into deeper documentation.

README should not become a duplicate of every deep GitBook page.

GitBook EN/DE

GitBook owns the reader-oriented deep documentation layer:

• user workflows;
• detailed product concepts;
• privacy/storage/encryption explanation;
• AI modes/providers;
• feature maturity;
• troubleshooting/recovery;
• architecture/developer pages;
• platform matrices;
• security model;
• English/German documentation parity.

The two language variants should remain semantically aligned while allowing idiomatic localized wording.

DeepWiki

DeepWiki is an intentional additional code-derived repository documentation / exploration surface.

The README must retain the established top-level badge/link:

[![Ask DeepWiki](https://deepwiki.com/badge.svg)](https://deepwiki.com/qnbs/WorldScript-Studio)

This badge is a deliberate long-lived integration and discoverability surface. Do not remove it during README redesigns without an explicit repository-level decision.

At documentation-convergence checkpoints, verify that the DeepWiki repository view remains reachable and reflects a reasonably current indexed repository state. Do not claim that the badge alone is the sole technical refresh mechanism unless that behavior is independently verified against current DeepWiki behavior.

---

3. Initial README convergence package

The next dedicated README docs PR should start from the already curated comprehensive candidate and requalify it against the then-current main rather than rewriting from scratch.

Required initial improvements include:

• preserve the DeepWiki badge at the top;
• prominently link GitBook EN and DE;
• describe GitBook as reader-facing and repo/current source as implementation authority;
• add a compact Voice & Dictation product section;
• add a compact Platform Capability Matrix;
• add a compact Security Model / trust-boundary hub;
• preserve ## 📚 Documentation Hub so existing README.md#-documentation-hub references do not break;
• retain local-first/offline-capable nuance;
• retain exact host-specific Claude routing truth;
• retain browser/PWA vs Tauri storage/encryption distinctions;
• retain explicit feature default-vs-maturity distinction;
• retain Qt/GPUI as roadmap/gated future work, not current implementation;
• keep rapidly changing model IDs out of long-lived prose where a canonical runtime catalogue is more appropriate.

The README PR should remain a standalone documentation PR, not be mixed into #719/#708-B/#713/#714 or another source-correctness slice.

---

4. README metric authority

Do not manually maintain generated metrics when repository automation owns them.

Current repository tooling includes:

pnpm run sync:readme

via scripts/sync-readme-metrics.mjs.

README rewrites must preserve the synchronizer's expected patterns or deliberately update the synchronizer with focused validation.

Metrics such as:

test count
test-file count
i18n key count
locale count

must be derived from current repository state immediately before the README PR is committed.

No hand-frozen count in this issue is authoritative.

---

5. Continuous documentation triggers

Re-audit README + GitBook whenever a merged change materially affects any existing public claim in these domains:

AI providers / model authority / execution modes
feature flags / defaults / maturity
privacy / security / trust boundaries
storage / encryption / recovery
project identity / async data-integrity semantics
PWA / service worker / offline behavior
deployment / hosting / CSP / proxy behavior
Tauri / desktop capabilities / native storage
Qt / GPUI / Rust Core roadmap authority
collaboration / E2E / room-key semantics
voice / local model downloads / permissions
DuckDB / analytics / OPFS
export formats / compile semantics / backups
plugins / extension model
accessibility
i18n / RTL / locale maturity
CI / validation / contribution workflow
toolchain requirements
release signing / updater / distribution
major user-visible features or removed features

A change does not require rewriting every page. It requires checking whether existing claims became stale.

---

6. Recommended cadence

Use event-driven reconciliation rather than noisy scheduled churn.

At minimum perform a parity pass:

1. after a major capability/security/architecture PR changes an existing public claim;
2. before a tagged/public release when documentation-affecting work has landed;
3. after canonical AI/provider/model authority changes (ai: requalify current model catalog, defaults and end-to-end AI functionality across all providers #704 lineage);
4. after feature-coverage/default authority changes (test(feature-flags): establish exhaustive coverage authority and prevent opt-in E2E blind spots #709 lineage);
5. after platform/runtime strategy changes (docs(roadmap): reconcile root v2 status with native roadmap and Core migration ledger #559/native roadmap lineage);
6. after security/privacy truth changes (security(truth): consolidate user-visible security/privacy claims across storage, providers, PWA, collaboration & supply chain #549 lineage);
7. after substantial GitBook restructuring;
8. when review/audit evidence identifies a stale claim.

Where deterministic drift can be derived from source, prefer automation. Do not try to mechanically generate nuanced privacy/security prose from code.

---

7. Required parity matrix

Maintain an explicit current disposition for at least:

Topic	README	GitBook EN	GitBook DE	Source owner
Product identity / local-first philosophy	required	required	required	current source/product architecture
Hosted surfaces / runtime differences	summary	detailed	detailed	deployment/config
AI providers/modes	summary	detailed	detailed	#704/current registry
Privacy/network boundaries	summary	detailed	detailed	#549/security architecture
IDB encryption lifecycle	summary	detailed	detailed	storage/security source
Desktop at-rest encryption limitation	required warning	detailed	detailed	native/storage source
Collaboration E2E + metadata/revocation	summary	detailed	detailed	collab source/issues
Feature flags/defaults/maturity	summary	detailed	detailed	#709/feature slice
Voice	summary	detailed	detailed	voice source
Export formats/backups	summary	detailed	detailed	#715/export source
PWA/Desktop/platform capability	matrix	detailed matrix	detailed matrix	deployment/native source
Native roadmap	bounded summary	developer detail	developer detail	#559/native roadmap
Developer setup/toolchain	required	detailed	detailed	package/CONTRIBUTING
Security model	bounded summary	detailed	detailed	#549/threat model
DeepWiki	badge + link	optional link	optional link	public integration

Do not force identical text. Require equivalent truth.

---

8. English/German GitBook parity

For every structural GitBook addition/removal, evaluate both language variants.

Maintain:

same major IA
same technical guarantees
same warnings/non-goals
same platform/provider/security semantics
idiomatic language-specific wording

Do not let DE become a stale translation of EN or vice versa.

If one language intentionally differs, document why.

---

9. Link and anchor integrity

Before README/GitBook publication:

• validate repository-relative README links;
• search for README anchors referenced elsewhere;
• preserve or migrate README.md#-documentation-hub deliberately;
• validate GitBook internal references in both variants;
• verify the EN/DE GitBook URLs;
• verify GitHub Pages/Vercel/repository/release/security links;
• verify the DeepWiki badge and repository link.

Do not silently break stable documentation entry points during a visual redesign.

---

10. Privacy/security absolute-language audit

Whenever README or GitBook is materially rewritten, inspect words such as:

all
always
never
nothing
fully
private
secure
encrypted
offline
local
E2E
zero tracking
no network

Keep an absolute only where the scope is actually proven.

Required enduring distinctions include:

local-first != no network
locally stored != encrypted
API-key encryption != project-data encryption
browser/PWA IDB protection != Tauri filesystem protection
collaboration content encryption != hidden signaling metadata
local AI != every application integration offline
updater signature != OS code signing

#549 remains the specialist security-truth owner; this issue makes sure those truths reach the public documentation surfaces.

---

11. Feature and AI authority dependencies

#704 AI provider/model authority

After canonical provider/model/default authority changes:

• re-read current provider registry/catalogue;
• remove stale exact model names from README/GitBook where they are not durable;
• ensure mode semantics match runtime;
• ensure Claude host behavior remains current;
• ensure local/cloud capability claims match admitted providers.

#709 feature coverage/default authority

After feature-flag/default/maturity changes:

• derive flag/default counts from current FeatureFlagsState;
• do not trust stale docs/FEATURE-PARITY.md counts over source;
• distinguish default ON from production qualification;
• update GitBook feature-maturity tables and README summary accordingly.

Current known drift to reconcile: source presently represents 23 flags with 15 default-on / 8 default-off, while older docs/FEATURE-PARITY.md prose has described 16 / 7.

---

12. Relationship to existing owners

This issue is an integration/documentation owner and must not absorb specialist implementation issues.

• security(truth): consolidate user-visible security/privacy claims across storage, providers, PWA, collaboration & supply chain #549 — security/privacy cross-surface truth.
• docs(roadmap): reconcile root v2 status with native roadmap and Core migration ledger #559 — roadmap/native/Core documentation truth.
• ai: requalify current model catalog, defaults and end-to-end AI functionality across all providers #704 — AI provider/model/default capability authority.
• test(feature-flags): establish exhaustive coverage authority and prevent opt-in E2E blind spots #709 — feature coverage/default/maturity authority.
• export(formats): establish cross-format parity and artifact-level qualification for PDF, EPUB and compile outputs #715 — export-format semantic parity/qualification.
• analytics(duckdb): requalify default-on DuckDB-WASM runtime, privacy, persistence and recovery semantics #716 — DuckDB runtime/privacy/persistence truth.
• perf(webgpu): reconcile enableComputeShaders with reachable production GPU paths and truthful fallback/coverage #717 — ComputeShaders reachability/fallback truth.
• security(collab): define room-key lifecycle, password rotation and participant revocation semantics #720 — collaboration room-key lifecycle/revocation semantics.
• ops(worker-bus): surface privacy-safe task health, dead letters and circuit-breaker recovery #721 — WorkerBus privacy-safe operational health.
• security(storage): define preserve-first recovery for IDB encryption recovery-required states #722 — IDB recovery-required remediation.
• security(process): establish an incident-response runbook and evidence-preserving emergency release protocol #723 — incident-response runbook.

When one of these changes public truth, reconcile README/GitBook here; do not duplicate the underlying implementation work.

---

13. Validation for documentation PRs

Re-read package.json at execution time and use only real repository scripts.

Expected relevant checks include, where applicable:

pnpm run sync:readme
pnpm run docs:check
pnpm run content:guard
pnpm run i18n:check
pnpm run lint
pnpm run typecheck
pnpm run ci:prepush

Also validate:

README anchor/link integrity
GitBook EN/DE structure parity
critical privacy/security/AI claims against current source
exact-head CI/review evidence for the README PR
GitBook published content after merge/change-request publication

---

14. Initial acceptance / bootstrap checklist

The first convergence pass under this issue should achieve:

• comprehensive README replacement is requalified against current main rather than copied blindly from an older candidate;
• DeepWiki badge remains at the top of README;
• GitBook EN and DE are prominently linked;
• README contains the compact Voice & Dictation section;
• README contains a compact Platform Capability Matrix;
• README contains a compact Security Model / trust-boundary section;
• Documentation Hub anchor compatibility is preserved;
• generated README metrics are synchronized from current source;
• current feature defaults/maturity are re-fetched after test(feature-flags): establish exhaustive coverage authority and prevent opt-in E2E blind spots #709-relevant work;
• AI model/provider/mode authority is re-fetched after ai: requalify current model catalog, defaults and end-to-end AI functionality across all providers #704-A-relevant work;
• live README no longer claims GitHub Pages/Vercel are operationally identical;
• live README no longer contains unbounded nothing leaves your device / everything works offline style claims;
• README and GitBook agree on desktop filesystem encryption limitations;
• README and GitBook agree on Claude host-specific routing;
• README and GitBook agree on Local as the strict no-cloud AI inference mode;
• README and GitBook agree that Eco is also a strict no-cloud AI inference mode (as of the fix(ai): close cloud-AI policy bypass in generateJson/generateImage/streamAiHelpResponse #706/fix(project): guard AI-generation results against a stale/switched active project #707 privacy-gate fix, assertCloudAiAllowed blocks Eco identically to Local — re-verify this against source at execution time rather than trusting this note indefinitely);
• README and GitBook distinguish feature default state from qualification maturity;
• README and GitBook describe Qt/GPUI as roadmap/gated work rather than current implementation;
• DeepWiki link/badge works after the rewrite;
• GitBook EN/DE remain structurally and semantically aligned;
• exact-head docs/CI/review gates pass for the README PR;
• resulting-main README rendering and public documentation links are checked after merge.

---

15. Long-lived operating contract

After the initial checklist is complete, keep this issue open.

Use comments/checkpoints here to record meaningful documentation-truth events, for example:

MERGED CHANGE / RELEASE
→ affected public claims
→ README disposition
→ GitBook EN disposition
→ GitBook DE disposition
→ DeepWiki/index check if relevant
→ source/evidence links

Avoid comment spam for changes that do not affect documentation truth.

Periodically prune obsolete checkpoint comments only if repository governance supports it; otherwise retain them as a useful public-doc truth history.

---

Non-goals

• using documentation edits to declare unresolved implementation issues complete;
• duplicating security(truth): consolidate user-visible security/privacy claims across storage, providers, PWA, collaboration & supply chain #549/docs(roadmap): reconcile root v2 status with native roadmap and Core migration ledger #559/ai: requalify current model catalog, defaults and end-to-end AI functionality across all providers #704/test(feature-flags): establish exhaustive coverage authority and prevent opt-in E2E blind spots #709 or other specialist acceptance criteria;
• copying the entire GitBook into README;
• making README and GitBook text byte-for-byte identical;
• manually editing generated README metrics forever;
• automatically changing production defaults based on public docs;
• claiming DeepWiki generated prose overrides current repository source;
• rewriting historical sprint/archive documents merely for stylistic uniformity;
• hiding known limitations to make the product sound more mature;
• starting Qt/GPUI implementation from a docs issue.

---

Priority / disposition

P2 — long-lived documentation authority / product-truth maintenance.

The initial README convergence should be executed at a clean handoff after the current P1 AI/data-integrity chain has stabilized the implementation truth it needs to describe. Afterwards this issue remains the durable parity owner throughout continued WorldScript Studio development.

[qnbs]

Documentation convergence activated

Trigger: #719 merged.

Resulting-main baseline:

• SHA: 36860f5c99f213151f514bd21247c2057988a6bc
• Resulting-main CI/CD: all checks green (Quality Gate Node 22/24, Build, Playwright E2E, E2E Deep Coverage, Browser Quality VRT+Lighthouse, Storybook, Security Audit, Verified Signatures, GitHub Pages deploy) — zero failures
• Resulting-main CodeQL: success

Execution:

• Standalone README/GitBook/DeepWiki convergence is now the immediate priority under this issue.
• The curated README baseline will be requalified against this exact main (not applied verbatim).
• ai(project): bind asynchronous image and interview side effects to originating project #708-B is intentionally deferred until this docs task reaches a clean boundary, per current scheduling priority.

Authority:

• current source/tests/config/CI/ADRs/security docs > README.md > GitBook EN/DE > DeepWiki generated view

Expected outputs:

• standalone signed docs PR referencing (not closing) docs(authority): keep README.md, GitBook EN/DE and DeepWiki continuously accurate and source-truth aligned #724
• GitBook EN/DE parity disposition
• DeepWiki badge/link verification
• synchronized README metrics via pnpm run sync:readme

[qnbs]

Bootstrap documentation PR opened

• PR: docs: comprehensively refresh README product truth and GitBook entry point #726
• Base SHA: 36860f5c99f213151f514bd21247c2057988a6bc
• Head SHA: ebe60147bcd241b87788c0b16119de8abdb7a5ee
• README baseline requalified: yes — corrected one stale claim (Eco is now a strict no-cloud mode, matching Local, per the merged fix(ai): close cloud-AI policy bypass in generateJson/generateImage/streamAiHelpResponse #706/fix(project): guard AI-generation results against a stale/switched active project #707 privacy-gate fix)
• README metrics sync: already in sync, no drift (pnpm run sync:readme)
• Documentation Hub anchor: preserved (README.md#-documentation-hub, verified against all cross-repo references)
• DeepWiki badge/link: preserved at top of README, unchanged target
• GitBook EN parity: not independently re-verified against the live site in this pass (no authenticated GitBook write access available) — no GitBook mutation claimed
• GitBook DE parity: same as above
• GitBook mutation required: unknown/not yet audited — recorded as a residual, not silently assumed clean
• Residual documentation truth: docs/FEATURE-PARITY.md's stale "16/7" flag-split summary paragraph was also corrected in this PR (small, directly related fix); no other residuals found in this pass

#724 remains open as the durable documentation authority.

[qnbs]

Bootstrap documentation convergence merged

• PR: docs: comprehensively refresh README product truth and GitBook entry point #726 — docs: comprehensively refresh README product truth and GitBook entry point #726
• Final PR head: 5e1e425ff3b335e64475c9c68576b99688ce49d2
• Merge commit / resulting-main SHA: 021b1bdc145f211cf4f4239db7ad26e44ca7fae2

Validation

• Exact-head CI/CD: green (all required + advisory checks passed on 5e1e425f)
• Exact-head CodeQL: success
• Resulting-main CI/CD: green (Quality Gate Node 22/24, Build, Playwright E2E, E2E Deep Coverage, Browser Quality VRT+Lighthouse, Storybook, Security Audit, Verified Signatures, GitHub Pages deploy, CI Success aggregate)
• Resulting-main CodeQL: success
• README/docs checks: pnpm run docs:check / sync:readme / content:guard / i18n:check — all green
• Metrics synchronization: no drift (7750+ tests / 606 files / 2942 keys / 19 locales — already in sync at merge time)

Public surfaces

• README: merged and live on main
• GitBook EN (space dzaWhxIfJfU66ttdNHqM): published via CR chore(deps-dev): bump jsdom from 29.0.2 to 29.1.1 #9 (id EA9ARQZG4YNtsICuqFmp), revision 48k3joVop9aqPfRostcU
• GitBook DE (space oeCpxXgZWxexpMiulvnL): published via CR chore(deps): bump actions/dependency-review-action from 4 to 5 #3 (id TERSl5ZuhNSoH2elxYTx), revision xIm0oHV7OxW0gpI034kr
• DeepWiki badge/link: preserved at top of README, target unchanged

Truth corrections

• Local/Eco AI-mode semantics: both README and GitBook (6 EN + 6 DE pages: Local vs cloud AI, FAQ, AI tools and Copilot, Privacy, Quick Start, AI providers and execution modes) now correctly state that Eco is blocked from cloud-provider AI calls by the same assertCloudAiAllowed gate as Local (confirmed against source post-fix(ai): close cloud-AI policy bypass in generateJson/generateImage/streamAiHelpResponse #706/fix(project): guard AI-generation results against a stale/switched active project #707), not merely a battery/cost optimization. Scope is explicitly bounded to AI-provider inference/routing, not a blanket no-network claim.
• docs/FEATURE-PARITY.md: removed the enableWebnnInference ghost row (not a real FeatureFlagsState field) and corrected a second stale "17 on / 6 off" drift-summary paragraph to match the current 23 flags / 15 default-on / 8 default-off split — found by CodeAnt/CodeRabbit/cubic independently on PR docs: comprehensively refresh README product truth and GitBook entry point #726, fixed and re-verified via pnpm exec tsx scripts/audit-feature-parity.ts (0 errors).
• GitBook stable internal links/card targets restored before publication; post-publish parity search confirmed the old "Eco is not the strict privacy mode" framing no longer appears anywhere in published EN/DE content.

Residuals

• docs/dependabot-triage-hardening and docs/local-repo-housekeeping local branches contain documentation content that was never merged via any PR — flagged for manual review, not evaluated as part of this docs convergence.
• No other known documentation-truth residuals from this pass.

Next trigger

Next technical source-correctness slice after budget reset / explicit continuation: #708-B (remaining #708 project-incarnation AI authority — #719 already closed the storage-namespace/ownership layer this depends on).

#724 remains OPEN as the durable documentation authority.

[qnbs]

Documentation-truth trigger checkpoint — 2026-09-16

The durable documentation authority remains OPEN. Several post-#726 changes now require explicit disposition at the next docs reconciliation; do not infer user-facing capability changes from architecture/planning PR titles alone.

Confirmed public-doc drift requiring correction

#762 — Node/toolchain support contract. After merged jsdom 30 (#754), current package.json and README still advertise Node >=22.19.0, while jsdom 30.0.1 requires ^22.22.2 || ^24.15.0 || >=26.0.0. #762 now owns the implementation/toolchain decision. Once its support policy is resolved, README/GitBook/developer setup must reflect the chosen exact Node contract. Until then, treat the current Node.js ≥22.19.0 public requirement as a known developer-doc truth defect, not authoritative compatibility evidence.

AI/runtime truth changes to consume

• ai: requalify current model catalog, defaults and end-to-end AI functionality across all providers #704 is terminal/closed historical umbrella. Current public AI truth should be derived from current source/catalogue plus active dedicated owners (ai(project): bind Writer, Copilot and deferred AI results to originating project incarnation #713/ai(lifecycle): unify Stop Generation, cancellation, supersession and duplicate-request semantics #714/ai(proforge): use native Claude structured outputs for schema-bound pipeline stages #729/ai(claude): right-size ProForge response budgets and implement real end-to-end streaming #731/ai(proforge): replace character-count token metrics with provider usage telemetry #732/release(ai): gate v1.28.7 on curated local-model and runtime qualification #743), not from treating ai: requalify current model catalog, defaults and end-to-end AI functionality across all providers #704 as ongoing work.
• test(ai): qualify Browser-Ollama admission boundary #744 added deterministic evidence that Browser-Ollama is network-inert while disabled and only contacts the local endpoint after explicit opt-in/user action. Public privacy/Settings docs may state that bounded contract; do not promote it into a claim that real Ollama inference/runtime qualification is complete.
• docs(proforge): plan real token accounting for the Claude path #727/docs(proforge): plan Claude max_tokens/timeout right-sizing #728/feat(proforge): plan Claude structured-outputs wiring #730 are merged planning/evidence documents. They must not cause README/GitBook to say that Claude structured outputs, larger ProForge budgets/real Claude SSE, or provider usage telemetry are implemented. Runtime owners ai(proforge): use native Claude structured outputs for schema-bound pipeline stages #729/ai(claude): right-size ProForge response budgets and implement real end-to-end streaming #731/ai(proforge): replace character-count token metrics with provider usage telemetry #732 remain open.
• refactor(ai): extract provider request lifecycle #748/refactor(ai): extract provider adapters #759/refactor(ai): extract connection-diagnostics and local-model-discovery #761 are behavior-preserving architecture decompositions. Do not turn those refactors into user-facing “new provider/runtime support” claims. Existing behavior may be re-described only when current source evidence actually changed the product contract.
• fix(ai): project-qualify image storage and use the real delete API #719/fix(project): fence async image authority by incarnation #734 replaced older ai(project): bind asynchronous image and interview side effects to originating project #708/fix(ai): reject stale AI image writes and project-qualify image storage #718 implementation assumptions for direct project/image/interview authority. Security/data-integrity prose that discusses stale cross-project AI effects should use the current owner map; ai(project): bind Writer, Copilot and deferred AI results to originating project incarnation #713 remains the deferred/apply-later residual owner and ai(lifecycle): unify Stop Generation, cancellation, supersession and duplicate-request semantics #714 the lifecycle owner.

Release-facing documentation rule

#743 remains OPEN / NOT PASS. Do not publish v1.28.7-facing “fully qualified local AI/runtime” claims before the gate reaches explicit PASS. Mocks/planning/structural refactors are evidence at their own layer only.

No immediate broad README/GitBook rewrite is required solely by this comment. The next event-driven #724 parity pass should consume these deltas, with #762's toolchain correction treated as the one already-confirmed current public-doc mismatch.

[qnbs]

Public-doc truth checkpoint — current post-v1.28.6 programme/toolchain drift — 2026-09-16

Current source authority at main 4137286887275ca03aa5b75906e6b529c68c786c has moved far enough that the next README/GitBook parity pass should consume the following durable owners rather than older snapshot prose.

Release/programme truth

• Latest shipped release may correctly remain v1.28.6 until a new tag exists.
• Current active release programme is release(ai): gate v1.28.7 on curated local-model and runtime qualification #743 for v1.28.7, not the completed v1.28.6 sprint still shown at the top of TODO.md.
• ai: requalify current model catalog, defaults and end-to-end AI functionality across all providers #704 is terminal historical umbrella evidence; current implementation owners include ai(project): bind Writer, Copilot and deferred AI results to originating project incarnation #713/ai(lifecycle): unify Stop Generation, cancellation, supersession and duplicate-request semantics #714/ai(proforge): use native Claude structured outputs for schema-bound pipeline stages #729/ai(claude): right-size ProForge response budgets and implement real end-to-end streaming #731/ai(proforge): replace character-count token metrics with provider usage telemetry #732 and the relevant test/admission/runtime owners.

Do not expose internal issue-number choreography prominently in user-facing docs; use these owners to re-fetch actual product truth before publication.

Developer/toolchain truth

New #762 tracks a concrete mismatch in the current advertised Node support contract:

WorldScript package/README: Node >=22.19.0
jsdom 30.0.1: ^22.22.2 || ^24.15.0 || >=26

and the current global undici <8 override forces jsdom 30 outside its declared dependency line. Until #762 resolves the intended contract, do not harden another public Node-floor statement elsewhere by copying the current README value.

#571 also confirms that manifest Playwright versions do not currently equal the effective browser/core line because repository overrides pin playwright and playwright-core to 1.61.0. Developer/test-tooling docs should describe the supported/effective line only after that owner converges.

#767 now owns the dedicated Vitest 5 breaking migration after the mixed #749 Dependabot PR proved incompatible with current test typings.

Security/dependency truth

#763 supersedes the current “no patched adm-zip release exists” narrative. adm-zip 0.6.1 now exists, but its security-fix status still needs source/repro qualification because advisory metadata is stale. Public/security docs should therefore avoid both stale absolutes:

NO FIX EXISTS        ← no longer safe to state
0.6.1 DEFINITELY FIXES IT ← not yet proven in-repo

Use #763's terminal evidence when available.

AI/runtime truth

#714 now carries a current P1 streaming-integrity finding: partial output from one provider can be followed by a full fallback answer from another provider in the same logical stream. Until corrected, avoid language implying fallback/streaming behavior is fully qualified across all cloud paths.

#729/#731/#732 remain runtime implementation owners even though their planning PRs #730/#728/#727 are merged.

Documentation operation

This checkpoint does not require an immediate giant README/GitBook rewrite while P1 runtime work is active. It updates the source-of-truth inputs for the next event-driven parity pass under #724. Prefer short current claims plus links to stable product docs over copying volatile model/tool/dependency details into multiple narrative surfaces.