The owner decision of 30 September 2026 is Module → Workflow → Factory. A consumer should be able to take one building block, adopt a complete job, or use the full lifecycle. This applies to existing and planned parts. Design owns the definitions; Core values makes independent composition a decision rule.
What can be adopted today?
The repository has bounded source modules and local behavioural evidence. The first E28 draft adds a private packed distribution with the public entry point software-factory/command-journal; the adoption guide defines its dependencies, contract and external temporary consumer. A following E28 foundations draft packages Execution, Lifecycle schedule and Intelligence without changing their source behaviour. The E29 draft adds project guidance through caller-owned identities and a second external package consumer. The guidance draft evidence records its exact artefact and failure checks. These bounded distributions were accepted locally on 1 October 2026 (owner-designated acceptance reviewer, after independent Opus 5.5 and Fable 5.1 reviews). Other Module and complete Workflow adoption gates remain open. Source tests alone prove their stated local scope, not adoption by another system.
Some parts already accept collaborators and configuration; others directly depend on Factory concerns. Preserve proven seams and refactor only observed barriers. An independent consumer may still require declared dependencies such as Node.js, SQLite, Git or macOS. Independent does not mean dependency-free or safe on every platform.
Independent adoption gate
Before offering a Module or Workflow as independently usable, record:
- Contract. A documented entry point, typed inputs and outputs, errors, stable IDs, version/compatibility policy, bounds and lifecycle. List runtime, platform, storage, provider and permission dependencies, including transitive dependencies. Importing the entry point starts no server, scheduler, sign-in or unrelated work.
- External consumer. A minimal application outside Factory consumes the intended distribution using only that contract and declared dependencies. It completes a useful example without importing private source paths, starting the Factory coordinator or configuring unrelated Portfolio/Dashboard state. Record the exact artefact, environment and result; a copied in-repository test is insufficient.
- Substitution. Where a collaborator exists, use an alternate conforming implementation without editing the consumer. A pure Module records that none applies. A supplied authority or evidence adapter must enforce the same trust boundary; arbitrary caller assertions cannot become authenticated Verdicts.
- Failure and state. Independently exercise applicable duplicate, invalid-input, cancellation, restart, corruption, revocation and unknown-effect cases. Keep IDs, data isolation, provenance, Budgets and recovery semantics. Prove compatibility or migration for persisted state and active pins; never silently adopt or recreate a consumer's data.
- Composition. Reuse the same contract in Factory, prove the affected integration and enforce dependency direction. Modules never import Factory orchestration. Workflows use public contracts of their declared Modules and any nested Workflows; dependencies are acyclic. Product data and authority stay separate.
Monorepo layout, in-process calls and a single deployable application remain valid. Package boundaries should follow demonstrated consumer needs; this decision does not require microservices, one package per file, a plugin framework, a new workflow engine or public publication. Replacing a platform-specific safety boundary needs equivalent independent evidence before use.
E29 draft also exposes neutral Attention and Owner digest contracts in the private tarball. A separate consumer supplies its own facts and SQLite file with no Portfolio, Git or provider setup; passive/typed imports, alternate readers, refusal and stored-record compatibility are checked. Draft evidence records the bounded implementation proof; the contracts were accepted locally on 1 October 2026, and the remaining E29 families stay open.
Workflows to expose
The composition map draws these Workflows with their capabilities and the Modules each shares or uses alone. These are required adoption targets, not accepted implementations. A Workflow may offer several versioned Recipes; a Run executes a Recipe. Decision domains retain their owners when Workflows are composed.
| Workflow | Host provides | Accountable result | Composition boundary |
|---|---|---|---|
| Research | Question/Objective, commissioned Sources, evidence and model adapters, storage, time and Budget | Attributed findings and proposals, with unknowns and admission status | Can run on demand with no Factory schedule; selecting a Slice is a separate decision |
| Implementation | Ready Slice, source snapshot, workspace, permitted worker and Budget | Immutable Candidate and execution Evidence | Does not issue its own independent Verdict or require a Factory planner |
| Verification | Candidate, independently owned acceptance, collectors and trust configuration | Scoped Verified, Failed or Inconclusive Verdict and Evidence | Can verify a Candidate produced outside Factory; implementation cannot choose itself as the independent authority |
| Marketing | Confirmed release/benefit facts, audience/channel brief, provenance and optional permitted publishing adapter | Fact-linked material and, only when authorised, publication Receipts | Can use releases produced outside Factory; drafting needs no Factory delivery loop or permission to publish |
The W31 author candidate preserves Delivery's existing verification owner. It admits an exact external seal in the bound owned store and recovers only the original durable witness through native proof re-derivation. Orphan Verdicts cannot authorise another dispatch. Focused source/native evidence passed; installed adoption, independent final review and combined acceptance remain open.
Audit and refactoring queue
Tracked adoption programme. The canonical epic plan owns outcomes, acceptance and estimates; linked issues own current implementation status. This page owns the requirement and audit baseline.
Audit baseline: 291056f, 30 September 2026. The table records that source/dependency audit: at that baseline the private root package had no exports/distribution contract. The first draft journal distribution addresses only that family's packaging gap; it does not accept other families or close E28. All adoption gates remain open pending their own evidence and independent review. Existing build-review evidence and the original 27 epic IDs, outcomes, dependencies and acceptance remain intact. E28–E35 add explicit adoption work. New and changed parts inherit the gate through the common epic contract.
| ID | Observed gap and bounded next change | Owning epics |
|---|---|---|
| C01 | The private package has explicit exports; complete Workflow contracts remain open. Worker resources, fixture manifests and collector integrity checks depend on source placement; the Dashboard has a repository-relative asset default. Inventory every accepted and planned Module; define only the needed entry points, declared dependencies and external-consumer fixtures. Move host defaults into Factory wiring where needed: Discovery sources already accepts cache/transport/clock configuration, but defaults to a machine-specific volume. Retain reviewed dependency restrictions until replacement edges are independently proved. | E28; related E17, E18, E20, E25 |
| C02 | At the audit baseline ProductGuidance created Portfolio; the E29 draft now supplies caller-owned identities through a packaged core and compatible Factory adapter. Attention exposes Dashboard DTOs; Owner digest reads concrete Portfolio/Branch health sources. Owner policy imports schedule helpers and Service discovery imports Portfolio types. Extract minimal identity/read-model/time contracts and keep compatible Factory adapters. Preserve journal replay, registration checks, provenance and honest unknowns. | E29; related E18, E20, E23, E25 |
| C03 | Schedule and Discovery sources exist locally; the complete Research Workflow does not. Expose an on-demand job contract, then make scheduling an optional caller. Planned intake, synthesis and admission use declared collaborators and keep source trust and Budget rules. | E32; related E14, E24 |
| C04 | Repository/Execution/Assurance pieces exist, but independently consumable Implementation and Verification Workflows are unproved. Extract separate job contracts with external Candidate and result boundaries; retain current delivery work and immutable evidence trust. | E30 (Implementation), E31 (Verification); related E02, E06, E08, E17, E18 |
| C05 | Growth drafting, release facts and publishing are planned. Build the Marketing Workflow from externally supplied, validated facts; keep draft generation separate from authorised publication and observation. Apply the same gate to later validation, A/B, cleanup and retirement Workflows. | E33; related E21, E22, E26, E27 |
| C06 | Project guidance v2 permits eight values; historical v1 retains its six-value parser. Local compatibility proof is recorded in the build review; commissioned migration of real stored guidance remains separate. Preserve old revisions and pins, refuse truncation and update editor/API/docs together. | E34; related E23, E25 |
Coverage. E28/C01 inventories the full current module catalogue: Journal, Execution, fixture worker, fence/workspace, guarded verification/recovery, confined executor, Evidence/Assurance, Repository, Portfolio, Dashboard, Project guidance, Attention, Producers, Repository Assurance, Product gate/delivery records, Repository collector, Intelligence, Lifecycle schedule, Discovery sources, Service access, Owner policy, Branch health and Owner digest. Related helper files belong to their owning Module, not automatic new products. E28 owns foundational and executable-resource adoption; E29/C02 owns the seven project/operations families (Portfolio, Dashboard, Project guidance, Attention, Owner policy, Branch health and Owner digest). Inventorying these in E28 does not require finishing E29 before it can start. C03–C05 become the four separate Workflow epics E30–E33. E35 proves coverage of all 23 current families and the four Workflows, plus Factory use of the same contracts. It does not wait for every future lifecycle capability.
The remaining planned Workflow catalogue includes planning/prioritisation, release/recovery, service-access operations, outcome validation, simplification/retirement and knowledge capture/retrieval. For planned parts, apply C01 before the first adoption claim and the relevant Workflow issue before composition. The separate Knowledge draft remains subject to independent design review under the current working agreement; this requirement neither accepts that draft nor authorises its implementation. Future service-access/vault, action-gateway, measurement and recovery work retains its existing authority and independent verification gates.
How much work?
The original E28–E35 provisional estimate is 13–18 reviewed increments to refactor existing code, 9–17 to build the four missing complete Workflows, and 2–3 to prove their Factory composition: 24–38 overall. These are bounded units of implementation with external-consumer evidence, fault/compatibility checks, independent review and docs included; they are not equal-sized days or weeks. Confidence remains low. The first external consumers now work in draft PRs; the Module inventory records the observed distribution and coupling work. The adoption-stack families were accepted locally on 1 October 2026 (owner-designated acceptance reviewer, after independent Opus 5.5 and Fable 5.1 reviews); no epic closes. Shared work counts once even when it supports an original customer-value epic.
The original eight-epic breakdown is the canonical estimate, with dependencies and exclusions. The audit covered 23 Module families across 46 source files (43,627 lines at 9927368); that is inventory size, not rewrite scope. Existing injection and pure-function seams should be reused. Packaging/resource identity and evidence trust, followed by project/operations coupling, are the main refactoring risks. Building complete Workflows is additional product work.
The first E28 increment now has a privately packed CommandJournal consumer with replay, conflict, restart and corruption proof. A second packages Execution, Lifecycle schedule and Intelligence unchanged. These drafts are included in the total estimate, not subtracted as accepted completion. The initial source-audit range was 17–31; separate executed-resource proofs and the remaining operations families raise it to 24–38. The inventory gives the bounded work breakdown. Research and Marketing accept external inputs and need neither completed Implementation nor completed release Workflows. Public publication, new platform sandboxes, provider qualification, live customer/marketing effects and the separate Knowledge implementation are outside this estimate.
Sequence: inventory and smallest useful external consumer first; then fix the dependencies it exposes and prove each Workflow with real consumers. Do not pause useful delivery for a wholesale extraction. Record each result as not assessed, gap found, or adoption verified, with evidence and the exact supported environment; these are adoption review labels, not new runtime states. Owner: Factory coordinator; independent reviewer: Astra. No deadline is commissioned.
The E29 Owner policy draft accepts a caller-owned journal and clock with neutral calendar and Count dependencies. Existing state and error compatibility remain, and the package consumer needs no scheduling runtime or Portfolio setup. This is bounded draft evidence, accepted locally on 1 October 2026, with the broader adoption gate still open.
A further E28 draft packages guarded fixture resources with distinct compiled manifest/collector/recipe identities and actual emitted-JavaScript hashes. External proof covers worker ownership, Evidence and guarded recovery through caller-owned storage. Source records retain their original grammar and pins. Repository-collector resources remain separate; it was accepted locally on 1 October 2026, and all complete Workflow gates remain open.
The repository collector draft addresses another observed packaging gap: eight emitted runtime resources receive distinct recipe/collector identities, while neutral Candidate and snapshot types no longer pull Repository or delivery storage into the package. Its standalone consumer proves collection and historical rejudgement, not Git ownership, integration or a complete Workflow. It was accepted locally on 1 October 2026; draft evidence does not close E28.
The E29 Branch health draft separates the existing Factory Portfolio facade from a core with supplied identities, reader, clock and optional neutral mirror. Its stored protocol, episode identities and polling decisions remain unchanged. Actual package consumers exercise recovery and replay without registration or Git. Independent Astra review passed; the draft was merged on 1 October 2026 after an Opus 5.5 review and accepted locally the same day by the owner-designated acceptance reviewer after a Fable 5.1 review.
Discovery and Service draft
The E28/E29 Discovery/Service draft supplies caller-required cache storage through a shared Discovery core and preserves the Factory default in its facade. Service discovery owns only neutral reporting types; public Access/discovery/catalogue entry points remain one Module family. Offline tarball proof and original record compatibility are local draft evidence. Independent Astra review passed; the draft was merged on 1 October 2026 after an Opus 5.5 review and accepted locally the same day by the owner-designated acceptance reviewer after a Fable 5.1 review. No scheduling, Research, provider workflow or Knowledge is implemented.
The Repository/gate/records adoption draft adds three explicit E28 entries around unchanged runtime behaviour. The external consumer seals and verifies an actual local Candidate, obtains Product gate admission, integrates once and reopens retained receipts; missing authority and unknown outcomes never become permission. Draft evidence preserves a real pre-edit journal and original pins. This advances the Repository and Product gate/delivery-records families in the inventory above; It was accepted locally on 1 October 2026; E28 acceptance, the four Workflows and E35 remain open.
The Producers draft removes the concrete Repository import from the shared Producer contract and packages the existing adapters. Neutral proposal types retain their previous names and shapes. Installed consumers prove deterministic local transport behaviour, explicit caller admission, refusals and unknown accounting; they do not qualify a live provider or supply a durable recovery controller. Draft evidence records an independent Astra pass; the draft was merged on 1 October 2026 after an Opus 5.5 review and accepted locally the same day by the owner-designated acceptance reviewer after a Fable 5.1 review.
Additional future planning scope
E36 Temporal Planning adds a future Module outside the fixed 23-family audit: date relationships, capacity forecasts, material windows and typed stage-trigger proposals. The same independent adoption gate applies. Its estimate is 5–8 new-module increments, separate from E28–E35's 24–38; the E28–E36 subtotal is 29–46, with low confidence and no calendar estimate. E35 retains its existing scope and dependencies. Research and Marketing may consume planning inputs but do not require this Module. No planner, dispatcher or publishing authority is established by the draft.
Additional operational scope
E37/E38 add Operational Signal Intake and a Repair Workflow outside the fixed 23-family/four-Workflow baseline. Their 8–13 increments are additional to E28–E35's 24–38 and E36's 5–8, giving 37–59 combined with low confidence. Shared generic delivery work is counted once in E36; planning remains an optional port. Existing customer-value contracts and E35 dependencies are unchanged. Independent external adoption, authorised production evidence and source writeback are separate gates; this design implements none of them.
T36.5a adds the candidate guidance-acceptance entry: caller-owned actual Journal, registry, offered ownership and common physical authority fence. Actual original Guidance commit is acceptance; no inbox or offer release. Focused source, installed package and independent component verification precede coordinator integration; no live service is qualified.
M01 adds candidate milestone-objectives: one original caller Journal/commission and qualified current authority fence, immutable intent/plan history and passive native projections. Research/ticket/scheduling and a resource-content forecast bridge remain unavailable. Independent source/package adoption and coordinator integration are required before acceptance; the owner form is a separate bounded increment.
Optional provider diagnostics source candidate
Provider diagnostics is one pure optional projection of retained public result/receipt facts. It preserves original outcomes and uncertainty, refuses contradictory inputs and adds no provider, retry or policy authority. Source and installed-package acceptance are pending independently; existing workflow and live qualification gates are unchanged.
Marketing standalone package candidate
W33 now offers four private Marketing entries around its unchanged pure pack and caller-Journal job. The installed consumer supplies external facts and storage, preserves retained v1 history, revalidates changed sources and reopens original job replay in another process. Fixture drafter substitution, conflicts, fixed allowances, cancellation and interrupted unknown non-retry pass. Independent final review and acceptance remain open. No live drafter, publishing adapter, effectiveness evidence or complete Factory is qualified.
A separate Research semantics candidate exposes caller-owned native evidence readers, an explicit qualified semantic fence and substituted structured triage/synthesis ports. Its original shared allowance and first-offer records survive job aliases and unknown outcomes. Existing v1 collections retain zero model calls. Source/package qualification and independent final acceptance remain open; no live provider, default shadow threshold or feature admission is supplied.
Guidance v3 is an authored atomic values/rationale candidate behind the existing public entry. It requires an actual caller-qualified owner-decision fence and preserves old digests/receipts; existing remote /2 editing adapters are read/inspect/replay-only after explicit migration. Contract and writer inventory. Independent source, installed and integrated acceptance remain pending.
The Factory workflow concert candidate exposes a stateless prepared-prefix coordinator over existing native owners. It preserves individually callable stages and original identities, performs separate current support checks before new progression, and stops for missing inputs, refusals or uncertainty. Its public package and concert proof remain under qualification; this adds no live provider, scheduler, publication or customer authority.
The Provider event delivery installed candidate closes an authored coverage gap: its existing public entry previously had source-native controls and compiled/export coverage, while the EV1 installed family exercised only projection. A separate actual native-store/TLS sender driver is now registered in the 36-family catalogue. It is authored but unrun; independent installed review, complete matrix/full checks and exact-main CI remain open. Runtime contracts, original ownership and private caller authority are unchanged.
Current family evidence and remaining coupling are maintained in the Module inventory; historical draft records remain immutable.
