Factory docs, home
Page navigation

Pre-release documentation. Proposed designs are labelled in the page; acceptance evidence is maintained in the build review. Local paths appear as placeholders. Current state: Build review.

This is the fixed 23-family scope of E28/E29 and the coverage input for E35. All rows remain gap found: draft artefact proofs have passed where stated, and the drafts were merged on 1 October 2026 after an Opus 5.5 review and accepted locally the same day by the owner-designated acceptance reviewer after Fable 5.1 reviews, but the complete adoption gate remains open. Epics and estimates own outcomes; programme #2 owns current PR status. The inventory is not a release or universal-portability claim.

The exact source/helper ownership is MODULE_INTELLIGENCE in Intelligence; structural graph checks enforce every runtime and type-only edge, including transitive capability and pinned-resource closures. Each linked Module reference owns its detailed dependency, error, bound and lifecycle contract. Helpers stay with their Module; extracting a shared type does not invent a new product.

Public entries below use the software-factory private package. The verified draft environment is Node.js 26.8.1 on macOS 27 arm64; individual contracts retain stricter platform and filesystem requirements. There is no registry publication or licensing decision. Other platforms need their own evidence. Importing any offered entry must remain passive.

ModuleEpicPublic entry or current gapDependencies and remaining proof
Command journalE28./command-journal — PR #11Caller-owned SQLite; replay, corruption and process restart proved in the private artefact.
ExecutionE28./execution — PR #12Journal dependency; caller enforces truthful cleanup and timeout ownership.
Fixture workerE28./fixture-worker, ./fixture-manifest — draft proofSeven actual emitted resources have separate versioned identity; launched JS, drift/link/missing refusal and old source history proved.
Fence and workspaceE28./attempt-fence, ./attempt-workspace — draft proofCaller-owned SQLite fence and workspace; lost ownership, cancellation and fenced restart recovery proved.
Guarded verification and recoveryE28./guarded-verification — draft proofCaller context supplies Journal, Execution, Evidence and paths; old recipe history, pin refusal and bounded fresh epochs preserved.
Confined executorE28./confined-executor — draft proofmacOS arm64, pinned runtime/launch tools and caller snapshots; external denial canaries, bounds and namespace quarantine proved. TypeScript consumers supply Node declarations; controller-loss and storage-race limits remain.
Evidence and AssuranceE28./evidence, ./assurance — draft proofCaller POSIX store; genuine read branding, wrong-pin rejection, missing/corrupt Evidence and historical Verdict preservation proved.
RepositoryE28./repository — draft proofCaller-owned bare Git; exact Candidate export, cross-product refusal, retained results and unknown integration outcomes proved.
PortfolioE29./portfolio — draft proofCaller journal, explicit Git and primary-checkout identity; external registration/import/restart and historical rows proved. Private package consumer and independent review passed; accepted locally on 1 October 2026.
DashboardE29./dashboard, ./dashboard-contract — draft proofLoopback server over a caller-owned synchronous backend; packaged browser assets fail closed when missing, stale or linked. Installed save/reload, stale save, lost acknowledgement and layouts proved; independent review passed; accepted locally on 1 October 2026.
Project guidanceE29./project-guidance — PR #13Caller identity registry replaces required Portfolio setup; Factory adapter preserves registration, revisions and pins. E34 v2 eight-value writes passed compatibility and browser proof and were accepted locally on 1 October 2026; live migration remains open.
AttentionE29./attention — PR #14Neutral input/output contract and pure projection; host validates identities, provenance and bounds. Dashboard adapter owns its envelope.
ProducersE28./producer, ./claude-producer, ./openai-producer — draft proofNeutral proposal/limit boundary; existing transports preserved. Deterministic CLI and loopback HTTP prove refusal, bounds, cancellation and unknown spend; no live provider qualification. The Claude adapter pins one executable path and hash on the Factory's own host, so real Claude use works only there; elsewhere it refuses with executable-missing.
Repository AssuranceE28./repository-assurance — draft proofCaller-supplied canonical Candidate/base and genuine store Evidence; wrong/missing/copied records, historical proof and exact protocol pairs proved.
Product gate and delivery recordsE28./product-gate and ./delivery-records — draft proofCaller-owned journal, fence and genuine Evidence; revocation, frozen Slice records, cross-product isolation and unknown-effect blocking proved.
Repository collectorE28./repository-collector — draft proofEight actual JavaScript resources with explicit compiled identity; launched process, drift/link/ownership refusal, cancellation and historical rejudgement proved.
IntelligenceE28./intelligence — PR #12Pure policy/registry contract: ordered priority lists per task and caller-declared routes and providers (selection only: nothing qualifies or admits a provider); current bundled tasks refuse, the Claude routes being unavailable for the Factory's model roles. Caller registry data grants no new authority.
Lifecycle scheduleE28./lifecycle-schedule — PR #12Caller commission, clock and account statements; fixed allowance and unknown-call non-retry. Import starts no scheduler.
Discovery sourcesE28./discovery-sources — draft proofCaller cache root, transport and clock; original journal/cache, expiry, drift, robots, request/byte bounds and interrupted claims proved.
Service accessE28./service-access, ./service-discovery, ./service-kinds — draft proofNeutral reporting types and caller names-only trees/listings; secret exclusion, partial/unknown coverage and substitute catalogue proved. No authentication or live telemetry connector.
Owner policyE29./owner-policy — draft proofCaller journal/clock and neutral calendar/reporting; historical periods, exact errors, replay and unknown/partial counts proved. Host ICU and Journal filesystem requirements remain. Gap: the owner's rules (src/owner-rules.ts, sf-owner-rules/1) are not in the ./owner-policy entry and have no package or consumer proof; one rules aggregate per journal.
Branch healthE29./branch-health — draft proofCaller reader/watch/clock and optional neutral mirror; Factory Portfolio wiring stays in the facade. Exact old journal, lost acknowledgement, restart, passive imports and public types proved.
Owner digestE29./owner-digest — PR #14Caller fact readers replace concrete Portfolio/Branch health. Fresh owner coverage must match the requested instant; old recorded digests retain their meaning.
Factory workflowE35./factory-workflow, ./factory-workflow-contract — candidate contractStateless prepared prefixes over native original owners, current support and guarded local integration recovery. Scoped installed individual/concert evidence retains its original pins. Historical source 7cf4b26e passed the 3,286/3,286 full checks and all 36 installed families. Independent bounded local evidence acceptance passed; exact-final-head CI is recorded separately from this frozen local proof. No release or customer-benefit claim.
Implementation WorkflowW30./implementation-workflow, ./implementation-workflow-contract, ./implementation-dispatch-gate, ./producer-workflow-contract — author candidateCaller ready Slice/store/Journal/private resources; local worker and separate non-invoking Producer gate compatibility. Historical f70 installed Implementation passed 8/8; historical 7cf Core22/22 and Dashboard26/26 passed. Historical source 7cf4b26e passed the 3,286/3,286 full checks and all 36 installed families. Independent bounded local evidence acceptance passed; exact-final-head CI is recorded separately from this frozen local proof.

| Verification Workflow | W31 | ./verification-workflow, ./verification-workflow-contract — author candidate | Exact caller-owned external seal in the existing Delivery owner; original proof witness and native re-derivation. Source/native and scoped installed source048 evidence remain historical. Historical source 7cf4b26e passed the 3,286/3,286 full checks and all 36 installed families. Independent bounded local evidence acceptance passed; exact-final-head CI is recorded separately from this frozen local proof. |

Current catalogue: 70 public entries / 36 installed families. The joint qualification record owns current combined evidence; historical records retain their exact scopes. Historical source 7cf4b26e passed the 3,286/3,286 full checks and all 36 installed families. Independent bounded local evidence acceptance passed; exact-final-head CI is recorded separately from this frozen local proof.

Additive sender coverage candidate

The historical 23-family scope above remains intact. The Provider event delivery candidate adds the missing separate installed sender family to the current catalogue, bringing its authored count from 35 to 36; public exports remain 70. The earlier EV1 family proves projection. The new sender driver exercised actual native ownership/private stores and controlled TLS admission, refusal and replay in its passing source048 family within an overall failed focused run. Historical source 7cf4b26e passed the 3,286/3,286 full checks and all 36 installed families. Independent bounded local evidence acceptance passed; exact-final-head CI is recorded separately from this frozen local proof. Historical 7cf bounded installed execution, full checks and independent local evidence acceptance passed. Exact-final-head CI remains pending. No live connector or domain effect is claimed.

Estimate after the first extractions

The revised total is 24–38 reviewed increments, including work already drafted: 13–18 existing-code refactoring, 9–17 new Workflow implementation, and 2–3 Factory composition proof. Confidence remains low. No epic is closed, and draft implementation is not subtracted as accepted completion. These units include implementation, external proof, fault/compatibility cases, independent review and docs; they are not days or PR counts.

E28 is now 7–10 increments: Journal (1); Execution/Schedule/Intelligence (1); confined execution (1); Evidence/Repository/Assurance/gate records (1–2); guarded worker/fence/recovery resources (1–2); repository collector resources (1); and Producers/Discovery/Service access consumers (1–2). The last group reuses existing injection seams and grows only for observed barriers.

E29 is now 5–6 increments: guidance identity (1); Attention/digest facts (1); Owner policy calendar/reporting (1); Branch health mirror (1); and Portfolio/Dashboard distribution (1–2). E34 stays 1–2 for compatible eight-value guidance. The four currently unbuilt Workflows retain their 9–17 estimate; these Module proofs do not implement their complete jobs.

The original 17–31 source-audit estimate bundled more families together. Actual tarball consumers now exist; independent review exposed a stale owner-coverage gap, and the resource audit confirmed that TypeScript sidecars cannot stand in for hashing executed JavaScript. Separate worker and collector versioned proofs are necessary. Re-estimate after those proofs and the first complete Workflow consumer. Preserve old source manifests and receipts; never silently repin them to make a package run.

Later Module and Workflow scopes

The fixed 23-family E35 denominator above records its audit baseline. It does not omit later work from the current catalogue. The module map separately indexes Calendar, Temporal Planning, Capacity Planning, Material Timing, Operational Signals, Work items, Proposal ownership, Guidance acceptance and provider tools, Milestone objectives/Research/work briefs, Research jobs, Delivery, Checkout Git, Marketing and shared-value libraries. Their references define contracts; build review defines accepted scope and remaining gates. New entries do not retroactively change the fixed audit or imply complete Workflow adoption.

Source: docs/adoption-inventory.md