Use the implementation plan index first. The owner commissioned bounded work; packet descriptions do not claim implemented APIs. Drafts #11–#54 were merged on 1 October 2026 after an Opus 5.5 review on the owner's instruction; Fable acceptance remains pending for each. Each packet inherits the index's common contract: exact base, caller-owned dependencies, actual tarball consumer, strict types/passive imports, meaningful fault/replay proof, immutable historical records, independent review and additive evidence.
D01 package documentation
Implementation draft independently reviewed: staged packing guide, exact draft evidence. The package now receives an installed-facing README and checked guide links through a disposable stage; ordinary source-root packing refuses. Integrated generated output/full checks and Fable remain separate gates. The original packet below records its required proof.
Inputs: actual npm pack file listing, README, package.json, scripts/build-modules.ts, every included adoption guide and the current source/artefact identity. Output: a package-facing entry document listing only actual public exports, supported environment, caller dependencies, minimal runnable examples and immutable evidence/reference links. Choose a supported staging/pack pattern that preserves the source README and private root package; do not publish, hide broken links with redirects or add private source exports.
Create a disposable consumer from the resulting tarball, open every local Markdown link relative to its actual installed path and check immutable remote reference format; reject missing files, checkout-only commands and private workspace/Notion links. Execute the documented public example with no repository present. Assert exact packed files and preserve all earlier consumer checks. Owning files are package/build configuration, the new package document and a focused test/distribution.test.ts helper. Completion is usable installed documentation plus unchanged runtime closure, not an import-only wrapper test. This is remaining distribution quality work, not a new epic or estimate.
P01 Portfolio adoption
Implementation draft independently reviewed: installed consumer guide, exact draft evidence. The unchanged runtime is exported with its Git/journal dependency closure and historical replay/error-boundary consumer proof. Integrated checks and Fable remain separate gates. The packet below retains its required proof.
Prerequisite: A01 journal distribution and current neutral reporting/attention contract decisions; inspect the latest E29/Branch health drafts. Existing sources are src/portfolio.ts and src/portfolio-inventory.ts; Portfolio reference owns the source API. Portfolio legitimately needs Git/checkout identities: removing that validation is not required for independence.
Inputs are caller journal, existing registry/import commands, explicit checkout roots, Git configuration and ViewQuery clock/window values. Outputs are unchanged registration/import results, sourced Portfolio/Product/condition views and exact/partial/unknown Counts. Existing operations include registerProduct, recordObservations, recordUsage, productView, portfolioView and currentConditions; confirm current signatures before exporting. The proposed public ./portfolio boundary must declare its whole dependency/platform closure and any explicit Node declaration requirement.
- Capture disposable pre-change journals with registrations, revised/retracted observations and usage, including receipts/events/outbox. Establish the minimal compiler closure; retain neutral shared types and errors already extracted, rather than importing Dashboard or scheduling for convenience.
- Add only the required public exports/compiler inputs. Keep machine defaults or command entry behaviour in host wiring. Construct an external consumer using two temporary real Git checkouts, caller SQLite and two logical Products; register, import, query, close/reopen and repeat commands.
- Challenge wrong/replaced/non-primary checkout identity, missing/corrupt journal, invalid time/window, stale or contradictory coverage, cross-Product observations, duplicate/revised/retracted records and concurrent writers. Verify imports cannot convert unknown usage to zero or activity to release. Old rows/results must replay unchanged.
Extend test/portfolio.test.ts, portfolio-inventory.test.ts, portfolio-observations.test.ts and portfolio-usage.test.ts only for observed gaps. Add a proposed test/helpers/portfolio-distribution.ts and external fixture to the existing distribution harness. Update exact graph/Intelligence/observability ownership, Module reference, guide and additive draft receipt. Completion: useful external registration/import/read proof, strict public types and declared Git dependency with no Factory/CLI startup or live onboarding.
P02 Dashboard adoption
Prerequisite: P01 or a separately proved neutral backend contract, current Guidance/Attention/digest cores and app build dependencies already available. Existing sources: dashboard-contract.ts, dashboard-view.ts, attention-dashboard.ts, dashboard-server.ts, apps/dashboard/src/ and apps/dashboard/public/. Read the Dashboard reference. The current startDashboardServer consumes journalPath, optional staticDir, port, guarded-command switches, Git and clock; it still constructs concrete backing modules. That is the barrier to examine, not an accepted generic backend API.
Define a proposed typed backend port from the server's actual reads and two guarded commands, preserving exact DTO protocol, provenance, Count semantics, error mapping and concurrency tokens. Make a caller-supplied backend usable without Portfolio registration or a source checkout. Keep the existing journal-path configuration as a compatible Factory adapter. Reads remain passive; no health recheck, inspection or guidance write occurs on page load. Supply explicit asset location/configuration with passive import and explicit listen/close lifecycle.
- Separate server composition from neutral contract/projection as far as the external consumer proves necessary. Test a substitute backend returning partial/unknown/error states and optimistic-save conflicts. Avoid broad callback ports that bypass command validation or authority.
- Package the actual built React/assets closure and required brand files using a reproducible build with current pinned dependencies. Resolve from the installed package or supplied absolute asset directory; no checkout-relative fallback. Require a missing/invalid asset tree to refuse cleanly, not silently serve source or initialise storage.
- In a disposable installed consumer, start loopback on an allocated port and serve the real UI against its supplied backend. Exercise Portfolio, Product, usage, blockers/digest and values editing at desktop/mobile widths. Preserve Host/Origin/token/content-type guards, static-path confinement, disabled-command behaviour and token secrecy; test traversal, malicious text/links, stale writes, backend failure, shutdown and no horizontal overflow.
Run the relevant test/dashboard-{view,server,boundaries,guidance-attention,editor,digest}.test.ts, formatting/brand tests and apps/dashboard build/browser checks. Add an actual-package server/browser fixture and packaged-link checks, not a fabricated HTML page. Completion: a usable installed browser application with declared backend/assets and compatible source Factory server. Coordinator serialises app/site builds and browser ownership; this packet does not publish or add remote listening.
W30 Implementation Workflow
3 October implementation candidate: agent contract, adoption guide and author record. Author focused/native checks passed; the installed fixture is authored but unrun. Independent final source/package review, serial full checks/CI and acceptance remain open. The packet below remains its qualification scope.
Prerequisites: accepted-for-use public Execution, fence/workspace, Repository/gate/records and Producer boundaries from the actual draft stack. Inspect D4/R7 ownership first: the separate delivery-d4 work at 42d2700 is not permission to overwrite or duplicate its delivery orchestration. Read Product gate, Producer, Repository and their tests. Reuse compatible reviewed work through a declared seam after coordination.
Proposed owning files: src/implementation-workflow.ts, its contract/test and package helper, only if no reviewed equivalent already owns the job. Define a versioned job containing job/command/Product/Slice identity, immutable ready Slice and source snapshot, permitted worker/commission, store/workspace/fence collaborators, caller clock, cancellation and fixed Budget. Results must distinguish Candidate plus execution Evidence from refusal, waiting, interruption and unknown remote effect. It cannot emit its own independent Verdict or grant integration/release authority.
- Build one end-to-end job using a conforming deterministic fixture worker. Admit/claim before contact, pin source/guidance/recipe/environment, produce bounded untrusted edits, validate and seal them through Repository, and return the immutable Candidate identity and genuine Evidence references. A supplied ready Slice is sufficient; no Factory planner or Verification job is prerequisite.
- Add durable stage/dispatch identity and restart/cancel recovery around worker invocation and Candidate sealing. Same payload/ID replays; changed payload conflicts; stale epoch or revocation refuses before dispatch. Keep an unknown provider effect debited and reconcile it rather than dispatching again under a fresh job ID.
- Substitute a second conforming worker in the same external consumer without editing it; challenge malformed/oversized edits, wrong Product/base/snapshot, missing workspace/Evidence, Budget/deadline exhaustion, crash before/after each durable boundary and recovery while another worker still owns resources.
Tests should use existing Product gate/Repository/Producer/Execution fixtures plus a new Workflow fault suite and tarball-only consumer. Live provider qualification remains unavailable unless separately commissioned and proved; deterministic doubles establish only their own scope. Completion: a useful externally requested Candidate through the public job, with durable failures and provenance, independent of Factory and Verification implementation.
W31 Verification Workflow
Historical W31 author snapshot. Contract and evidence reuse Delivery's existing verification owner. Explicit external admission requires the exact bound native store and frozen Slice, without fabricated production. The original proof witness allows read-only re-derivation after genuine native owner exclusion; a missing witness stays inconclusive without recollection. Native source checks passed; installed proof, independent final source review and combined checks remain pending. Current bounded local native and installed acceptance is recorded in the joint qualification; exact-main CI and complete live Workflow adoption remain separate.
Before extraction, independently review external-origin authority/provenance and state grammar within the same ProductGate, plus bounded original-dispatch/proof discovery after lost acknowledgement. D4 reads a retained Verdict only when an observed dispatch names it; digest-only retention does not discover orphaned proofs. Keep those proofs inert without a qualified original witness. No second owner, fabricated production, automatic retry or allowance reset.
The candidate preserves independently owned acceptance/recipe/resources; final acceptance must establish actual Evidence brands, the three-valued Verdict, historical meaning versus current validity, original identity/Budget and native exclusion. Qualify an actual external Candidate and collector through a private tarball consumer with two caller arrangements, then prove corruption, revocation, crashes, exhausted bounds and cross-Product isolation. Existing D4/R7, Knowledge and Fable/Opus gates remain unchanged; no model or integration authority follows.
W32 Research Workflow
Prerequisites: public Discovery, evidence/provenance, Journal and optional declared model/Intelligence ports. Read lifecycle for the existing planned triage/synthesis/gate rules; those proposed modules are not implemented merely because their names appear in the registry. Do not confuse public-text research intake with E37 operational telemetry intake.
Proposed src/research-workflow.ts accepts question/Objective, logical scope, commissioned Sources/brief, caller transport/cache/storage/clock, independent citation/evidence checker, optional qualified model collaborators, policy and bounds. Return attributed findings, coverage and proposals/admission dispositions, never a selected feature Slice or new Authority. Proposed collaborators/records must be strictly specified and reviewed before use; a missing optional model yields declared untriaged/parked output, not invented findings.
- Run an on-demand collection job outside Factory, preserving actual captures/citations, source coverage, novelty and bounded request/byte/deadline behaviour. No Product registry or schedule is required. Retain existing final-chunk accounting and source trust limits.
- Add bounded triage/novelty and synthesis using typed structured collaborator results. Reverify every citation/quote and scope, distinguish customer evidence from competitor-only research, retain policy/cache/watermark identity and document unavailable-model fallback. Source text cannot instruct tools or qualify a provider.
- Apply the declared admission rules as rules over facts; a model challenge may only lower admission. Return an inspectable proposal/disposition with alternatives and provenance. Selection into a Slice belongs to Planning. Complete cancellation/restart, unknown model-effect non-retry and non-renewable Budget proof.
- Add an optional schedule adapter calling the same public job and compare it with the on-demand consumer. Substitute a source/evaluator with no consumer edits. A no-change job can finish without a model call; interrupted work keeps its identity and honest coverage.
Use Discovery normal/adversarial tests, schedule/Intelligence boundaries and new Workflow cases for stale/expired/missing citation bytes, hostile sources, duplicates/rejections with new evidence, wrong scope, unknown effects and exhaustion. Completion: useful on-demand findings/proposals and schedule parity with exact attribution, not a claim that research has selected or delivered a valuable feature.
W33 Marketing Workflow
Historical 2 October 2026 snapshot: Step 1 implementation draft, integrated 2 October 2026; independent review of the final revision and acceptance pending: agent contract, draft evidence. Pure fact-linked validation and deterministic rendering only. Step 2 implementation draft (W2-08), integrated 2 October 2026 after independent review-and-fix rounds, not accepted: a durable job over a caller Journal and clock with a substitutable fixture drafter, a fixed Budget, cancellation and unknown-outcome non-retry (durable job, draft evidence, round 1 correction, round 2 correction). A substitute fact validator, step 3 publication, a live drafter and an installed consumer remain open. The original packet below records its required proof.
Prerequisites: neutral claim/evidence, Journal and optional qualified drafting ports. A consumer may supply validated external release/benefit facts; it need not run Factory, W30, release orchestration or a planner. Read lifecycle post-development gates, E33 and Temporal Planning for optional timing inputs.
Current source implements pure src/marketing-workflow.ts rendering/revalidation and src/marketing-job.ts durable caller-Journal drafting. The fact validator is fixed code; only the drafter is substitutable. The four public entries and bounded local installed proof for both steps are independently accepted in the joint qualification. Exact-main CI and live-drafter adoption remain separate. Adoption guide. Publication remains a separate unimplemented step; drafting cannot grant its permission.
- Implement useful deterministic rendering and strict fact/claim validation first. Drafting may propose prose, but code and independent review verify every factual claim against current pinned facts. Reject unsupported superlatives, pricing/availability/causality or benefit claims, wrong scope, personal data and expired/withdrawn sources. Preserve old pack meaning and invalidate new use when a fact is withdrawn.
- Use a substitute fact validator or drafting collaborator through the same external job. Prove duplicate command replay, concurrent revision, cancellation, missing storage, Budget exhaustion and unknown model results without calling a provider unnecessarily.
- Where commissioned, add a separate publication adapter with target/channel/grant/payload identity, current claims and authority at dispatch, durable Action/Receipt and execute/reconcile proof. Exercise revoked grant, stale pack, lost acknowledgement and a confirmed remote result. A fixture publication receipt is labelled simulated and establishes no live exposure. Optional Temporal timing only proposes windows; it cannot publish.
Completion: a fact-linked pack generated by a real external consumer and independently checked against changing source facts. If publication is offered, its fault/substitution contract must pass too; live E21 acquisition/activation evidence remains separate. Scheduling, owner consent and marketing effectiveness cannot be inferred from a good draft.
Evidence and stopping checklist
For the chosen packet, save a compact candidate handoff naming exact HEAD, changed-file and tarball hashes, focused commands/results, tested environment, historical compatibility, current authority and unresolved gates. Verify docs links from the installed artefact and update the affected Module reference/guide or proposed Workflow contract without changing historical receipts. Have another agent review the frozen candidate; the coordinator integrates, runs the final full checks and records draft/Fable status. Do not start the next packet automatically when the owner's instruction is to stop.
