Factory docs, home
Page navigation

Use this index to find the contract you need. Build review owns acceptance and review evidence. Working instructions own the contribution protocol.

1. Purpose and current truth

Accepted local infrastructure only, on one macOS host. No production, pilot, release or customer is established. All 27 customer-value epics and 15 adoption epics remain open. The build review records each accepted scope; the implementation plan describes intended outcomes.

2. Domain language as used in code

One line per term. Status says how far the term exists in src/: implemented, narrow (real but limited), pinned only (an identity string, no behaviour) or designed, not implemented.

TermIn code todayStatus
Module / Workflow / FactoryThree adoption layers: building block, complete job, full lifecycle. 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.reviewed local candidate
ProductproductId registry entry in Portfolio (ProductRecord, one verified checkout, operationScope: "metadata-inspection", deliveryAuthority: "absent"); named in Repository identities and manifests and in Guidance aggregates. The Product gate (gate:product:<productId>) binds a Product once to a Factory-owned store outside every registered checkout and admits its delivery work; only tests call it.narrow
ObjectiveOnly the objective text of an imported Portfolio focus fact. No Planning.narrow
SignalNot present.designed, not implemented
CaptureCapture (sf-capture/1) in Discovery sources (lifecycle increment 2, accepted locally): one item of at most 4 096 bytes of normalised text from a commissioned Source, with its SHA-256, SimHash and novelty (new, changed, repeat, unchanged); larger text is refused, never cut short, and only its hashes are journalled, the text staying in a cache that is readable for 35 days and pruned at each collection (§8). A Capture handed back is trusted only as its occurrence's record holds it. Tested only on a loopback corpus; nothing composes it.narrow
CitationCitation (sf-citation/1) in Discovery sources: a verbatim quote of at most 15 words (counted strictly) that must occur in a recorded Capture's text and carry no address, handle or phone number; it re-verifies from those bytes and the occurrence's record, so a quote moved to another URL, time or occurrence is rejected, and is unverifiable (unknown, not failed) once they expire at 35 days. Not registered with the Evidence store; nothing makes one yet but tests.narrow
ClusterNot present. Proposed lifecycle §4: a Product's Signals under one problem key (signal-intake), scored for synthesis.designed, not implemented
OccurrenceLifecycleSchedule in lifecycle-schedule.ts: one scheduled discovery pass with a stable ID (discovery/<product>/basic/<local date>, …/full/<ISO week>, discovery/portfolio/<rhythm>/<ISO week>), admitted, deferred, reconciled, resumed and settled once, each transition its own journal command; missed ones coalesce into the next. Only tests call it.narrow
OpportunityNot present. Proposed lifecycle §4: a falsifiable need drafted from Clusters (opportunity-synthesis) that the opportunity gate admits, parks or rejects; admission never selects it into a Slice.designed, not implemented
OutcomeNot present. Dashboard measuredValue is always unknown ["no-outcome-assessment"]. (Outcome in factory-support.ts is the CLI result word, not the domain Outcome.)designed, not implemented
CapabilityNot present (disposition versus health). Dashboard capabilities: {inspect, guidance} are server feature flags, a different sense.designed, not implemented
SliceFrozenInputs.slice {id, revision} and IntegrationPayload.slice in Repository; sliceId on Portfolio observations. The Product gate freezes a SliceEntry (scope, operations, frozen-input digests, allowance, deadline) under a grant, and the delivery records keep the frozen SliceRecord (sf-slice/1: Need, scope, non-goals, acceptance, recipe, environment, brief with an optional guidance pin, assessment plan); only tests create them. No planning.narrow
Product PackNot present.designed, not implemented
ReciperecipeRevision pin. Source guarded fixture verification uses sf-local-fixture-verify/2; the private compiled-resource draft adds /3 with distinct emitted-resource pins (/1 is read and settled, never dispatched). Repository scope retains source recipe sf-repository-verify/1; the private compiled collector draft adds /2 over actual emitted resources. Both pin the collector closure and executor bounds; authored delivery composition through local integration is a candidate awaiting complete acceptance.narrow
RunExecution aggregate run:<runId>: queued, running, completed, exhausted, cancelled; pins and Budget fixed at submit. completed means the recipe ran to its end, never a Verdict.implemented
Attempt<runId>/<epoch>, epochs 1–20, running, completed, failed or interrupted; one owner; ends only by the owner's report or a trusted cleanup confirmation for its epoch.implemented
CandidateTwo narrow forms: the fixture Candidate digest (sf-local-fixture/2 in source mode; /3 in the compiled draft, from fixture name and execution manifest digest) and Repository's SealedCandidate (sf-candidate/1: manifest plus exact bytes). The accepted Producers module can return an untrusted Proposal for seal, but nothing composes it and both live transports are unavailable, so no producer proposes anything yet.narrow
EvidenceEvidenceStore: content-addressed sf-evidence/1 records with eight EvidencePins and a ProcessObservation; deliberately no passed flag.implemented
Verdictjudge() → Verified, Failed or Inconclusive with scope: "local-fixture". judgeRepository() gives the same three results with scope: "local-repository" over store-written repository records, only in tests; the Repository collector (accepted, not composed) writes such records from real sandboxed runs, also only in tests. Portfolio verification facts carry a Verdict word plus a separate validity.narrow
ActionThe integration occurrence key action {id, occurrence} in Repository; the Product gate admits integrate-local dispatches under integrate-local:<productId>:<sliceRef> and records their outcomes; only tests call it. No Action gateway.narrow
Action grantNot present. Proposed lifecycle §4: the owner's single-use permission for one outward Action (owner-decisions), bound to an artefact digest and channel.designed, not implemented
ReceiptIntegrationReceipt {id, target}: the receipt commit after a confirmed target update; the delivery records' create-once ReceiptRecord keeps the sender's own IntegrationObservation of a dispatched move. Two other senses exist: the journal's command receipt (a stored row) and Portfolio's UsageReceipt (provider usage).narrow
AssessmentderiveAssessment in the delivery records derives supported, inconclusive or rejected from a frozen Slice record, a repository Verdict and the records it judged, Verdict-first; nothing stores a conclusion and only tests call it. No Supported, Rejected or Inconclusive record exists.narrow
AuthorityThe Product gate's grants (paths, operations, limits), revocations, stop intent and one fence-proved owner; only tests exercise them, and Portfolio still reports deliveryAuthority: "absent". confirmedBy and collector names are attribution, not authentication.narrow
PolicyNot present (only HTTP Permissions-Policy headers, unrelated).designed, not implemented
BudgetBudget {maxAttempts, attemptTimeoutMs} in Execution; never extended; no cost field. The Product gate's per-Slice allowance (GrantLimits: production runs, check runs, attempts per check, integrations, elapsed time) is counted separately and never decreases. The Lifecycle schedule's discovery-window ledger counts estimated tokens per ISO week against a ceiling set when the week opens and never raised; an unknown call stays debited at its bound; only tests call it. A shared Slice allowance across providers is planned.narrow
EscalationNot present. Nearest thing: an attention item with ownerRequired, derived from a source-reported blocker that names the owner and an exhaustion; the Factory records none itself.designed, not implemented
Command, EventCommandJournal.execute(commandId, payload, expectedVersion, aggregateId, decide): one atomic transaction records state, receipt, 1–100 Events and outbox rows. Nothing delivers the outbox.implemented
epochAn Attempt's position, 1–20, and the ownership version scope owners check; mirrored by MAX_FENCE_EPOCH and the ledger's MAX_EPOCH.implemented
pinsRunPins {candidateDigest, recipeRevision, scenarioRevision}; EvidencePins add runId, attemptId, epoch, environment, collector; RepositoryPins (Product, store, base, commit, Slice, snapshots, frozen records, Attempt) map to repository-scope EvidencePins through evidencePinsOf; GuidancePin {productId, revision, digest}. Any input change needs a fresh Verdict.implemented
ScenarioscenarioRevision = "local-fixture-scenario/" + digest(environment, timeoutMs, maxOutputBytes); the environment pin is node-<version> <platform>-<arch>. Repository scope uses sf-repository-scenario/1/<hex> over the Product, store, base, Candidate, Slice, snapshots and acceptance.narrow
Fence, WorkspaceFenceIdentity: a one-row SQLite file held by a kernel-released read lock, open → fenced one way. WorkspaceRef: one fresh 0700 directory bound to that fence. fenced proves protected work cannot resume, not process exit.implemented
CollectorAn attribution label on Evidence and Portfolio facts (sf-guarded-fixture-worker/1, local-git, claude-code-result); never authenticated. The repository collector (sf-repository-scenario-collector/1, pinned with the digest of its source closure) is accepted as step 1, increment 3, and nothing composes it.implemented
Observation, FactPortfolio Observation {collector, recordId, revision, productId, sliceId, observedAt, capturedAt, fact}; FactKind: focus, source-coverage, activity, verification, release, measurement, blocker, repository, usage-invocation, usage-session, retraction.implemented
Countexact {value}, partial {atLeast, reasons} or unknown {reasons}; only an exact Count ever shows 0.implemented
GuidanceProductGuidance: versioned values with immutable pins and compatible v1/v2 history; v3 records an exact owner decision, rationale and supporting evidence under its current fence. Knowledge advice and the selected-plan composition consume the retained pin; Guidance grants no Delivery authority.implemented
ServiceA third-party system a Product uses through a vendor account. Service access detects one from catalogue signatures (ServiceEntry in a Service map); nothing acts on a Service. Increment 1.narrow
Service kindA reviewed, versioned, frozen catalogue entry (vercel/1): signatures, credential slots, a 1Password item template, purposes and routes, desired settings, runbook templates, digested; thirteen exist (cloudflare/1, increment C1, integrated after independent review, not accepted; datadog/1, with sentry/2 and posthog/2 superseding their first revisions, packet SA5-OPS, independent review owed). Increment 1.narrow
Credential referenceWhere a credential lives and what it is for, never its value: CredentialReference (location, kind, instance, slot, class, presence, usage, consumers) with a reference id hashed from the Product and the location. Increment 1.narrow
Service mapsf-service-map/1: one Product's Services, Credential references, CI jobs, findings and Counts at one commit, derived inside the journal decision from names-only scans and listings; only test fixtures produce one. Increment 1.narrow
Product vaultA dedicated 1Password vault per Product, owner-created and bound by id, the source of truth for its secrets. Its store binding (onepassword-vault) and location type exist. Increment 4 (integrated after independent review; not accepted) lists item presence names-only through a fake-tested op item list; no real op has run, nothing composes the reader, and no field or value is read.narrow
BlockerPortfolio blocker fact (open or resolved, prerequisite, affected, remediesTried, nextAction, optional details) → an attention action. Resolution is the source's report, never verified.implemented
Work itemWork items sf-work-item/1: one unit of a Product's work with scope keys, at most one holder (attribution only, never authority) and a lane (branch, worktree, host). A claim that duplicates open or recently merged work is refused by fixed rules unless a declared comparison, a follow-up or an open takeover allows it. The register says who holds what; the Product gate still decides who may.narrow

The Temporal Planning contract separates Date claims, Capacity forecasts, Timing recommendations, Plan revisions, Commitments and Command proposals. Current Temporal, Capacity and Material modules evaluate caller inputs; durable E36 workflow APIs remain planned. These later scopes are outside the fixed 23-family audit. Outbox delivery and acknowledgement remain absent; stage-trigger proposals retain evidence, capacity and authority checks.

The future E37/E38 operational contract specifies Operational Signal Intake and Repair, outside the fixed audit. A bounded O37.1 intake candidate passed independent source/package review after its recorded-admission correction and was accepted locally on 1 October 2026; live adapters, grouping, metric detectors and Repair remain unimplemented. It retains source/metric semantics and durable case/episode identity; production resolution needs independent release/cohort evidence, and source writeback needs its own authorised Action and readback. Service access remains metadata, Discovery remains public-text collection, and local Assurance does not become production proof.

The provider event boundary projects committed scoped facts for subscribed assistants through an optional MCP notification adapter. Webhook receipt, durable domain acceptance and verified effects remain distinct. It adds no Intelligence transport or D4 ownership. No live subscription or application delivery is qualified; current local acceptance is recorded in build review.

The Guidance receiver native-fulfilment correction observes nested fulfilled data within the existing budget. Refusal, historical witnesses and unknown commit semantics remain unchanged. Build review owns current source, package and CI evidence.

The Attempt fence private-cleanup correction retries only the invocation's private name after catchable publication errors. Published targets, original errors and crash/cleanup uncertainty retain their meanings. Current gates belong to build review.

The 1 October local adoption acceptances have immutable docs/*-acceptance-receipt.json records (build review). Later source changes require fresh evidence. Live qualification, release and customer value remain unestablished; every epic stays open.

The parallel acceptance corrections add result-only Collector diagnostics and refuse fresh future-dated Owner Rule uses. Affected native, installed Collector and documentation checks passed with distinct Astra xhigh physical acceptance; exact-main full CI remains separate. Evidence bytes, Verdicts, historical replay, Ask Ledger time policy and live authority remain unchanged.

3. Module map

Each reference owns its API, invariants, composition limits and focused tests. Guides explain why that responsibility exists. For current acceptance, use the build review and joint qualification record: 70 public entries and 36 families. Historical f70 Implementation8/8 remains pinned to f70; historical 7cf focused proof is Core22/22 and Dashboard26/26. 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 bounded stage and concert evidence is independently reviewed in the original scope record. The joint qualification record and build review own current combined evidence and exact-main CI gates; historical receipts retain their exact qualification counts.

ModuleResponsibilityAgent referenceHuman guide
CalendarShared Gregorian arithmetic and fixed zone conversion rules; caller adapters retain domain boundscalendarguide
Knowledge (K1 local candidate)Permitted durable sources, independently reviewed lessons, loss-aware consolidation, current retrieval/use/history and Guidance-bound deterministic advice. Independent source, installed evidence and frozen synthetic corpus review retain their original scope. 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.knowledgeguide
Work items (WI1 candidate)One register per Product of who holds which work in which lane; refuses a claim that duplicates open or recently merged work by fixed scope-overlap rules unless a declared comparison, a follow-up or an open takeover allows it; reads lanes from Git read-only and reports merged, superseded and stale lanes with evidence and a proposed action; records attribution only and deletes nothingwork-itemsguide
Operational Signals (O37.1 candidate)Caller-authenticated redacted intake; one bounded ledger commits Signals and cursor coverage together; corrected source/package passed separate independent review; all 1,926 combined local checks passed; accepted locally 1 October 2026; CI remains openoperational-signalsguide
Marketing Workflow (W33 source drafts; private package candidate)Pure fact-linked rendering and caller-Journal drafting with original Budget, cancellation and unknown non-retry; reviewed installed replay accounting and observational assessment links. No live model or publication. 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.marketing-workflowguide
Provider Guidance tools (EV4a candidate)Inspect/apply one original Guidance offer through current receiver authority; corrected source/package independently accepted, Root's 2,121-check combined full run passed; accepted locally 1 October 2026; CI and final publication review pendingprovider-guidance-toolsguide
Guidance acceptance (T36.5a candidate)One original offered Guidance command reconciled against actual Journal/authority; native-fulfilment correction independently accepted in source/private package. Root's historical 1 October 2,121-check combined full run passed; accepted locally 1 October 2026; CI and final publication review pending; earlier full results historicalguidance-acceptanceguide
Proposal ownership (T36.4a candidate)Immutable typed proposals and permanent first-offer ownership; original integration passed 2,020 full/212 package-doc checks. Current returned-data correction independently accepted; corrected stack 6f4ff02 passed 2,045 full checks; combined package/public independently accepted; shared notification mode has independently accepted corrected source/package, with Root's 2,121-check combined full run and independent same-package Journal/EV2/notification-ownership physical proof passed; accepted locally 1 October 2026; no senderproposal-ownershipguide
Research semantics (separate local candidate)Caller-commissioned synthesis over genuine retained quotes, shared original allowance and explicit shadow policy; no feature admission or live provider. 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.research-semanticsguide
Research jobs (W32a source candidate)One originally commissioned native collection with durable markers and genuine attributed reader; no model, work admission or live source. 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.research-jobsguide
Milestone work brief (M03a source candidate)Pure selected-generation brief; no ticket, dispatch, authority or resource ETA. Independent and package gates openmilestone-work-briefguide
Milestone Research (M02 source candidate)Pure native handoff and first-plan preparation; real caller report/review ownership required, no automatic work or ETA. Source/package gates openmilestone-researchguide
Milestone objectives (M01 candidate)Component source/private-package independently accepted; corrected owner binding and plan source reviewed, with built-browser checks. Current combined package/full integration open; Research, tickets, dispatch and resource ETA unavailablemilestone-objectivesguide
Command journalAtomic command → state, receipt, Events, outbox; replay by command ID; read-only inspectioncommand-journalguide
Evidence / AssuranceContent-addressed observations; pure three-valued Verdict foldevidence-assuranceguide
ExecutionRun and Attempt ownership, Budget, durable cancellation; starts nothingexecutionguide
Local fixture workerOne built-in fixture as a real bounded process, observed as Evidencelocal-fixture-workerguide
Attempt fence / workspacePer-Attempt ownership gate and private workspaceattempt-fenceguide
Guarded verification / recoveryVerify a fixture under exclusive ownership; settle orphaned Attempts within Budget; the verify, inspect, cancel and recover CLIguarded-verificationguide
Confined executorOne bounded Node program against read-only snapshots in a macOS sandbox; observes, never judgesconfined-executorguide
Repository sealing / export and integrationSeal bounded text edits as an immutable Candidate; export exact bytes; move B → M once per prepared Action through explicit guarded or unguarded methodsrepositoryguide
PortfolioRegistry of local Products bound to one checkout each; read-only views of sourced facts and usageportfolioguide
DashboardNeutral contract, pure projection, loopback HTTP with two guarded commands, React presentationdashboardguide
Project guidanceUp to eight versioned values per Product with immutable pinsproject-guidanceguide
Required actions and blockersPure projection of current sourced conditions into actions, notices and blocker detailattentionguide
Repository AssurancePure judgement of repository-scope Evidence for one sealed Candidate (Candidate-as-program checks, pinned stdout digests, authenticated manifest and base) into a local-repository Verdict; never relabels fixture proofrepository-assuranceguide
Repository collectorRun each frozen process check of one sealed Candidate once under the Confined executor (the base or Candidate tree as the program, the named fixture input, nothing else) and store exactly one repository-scope Evidence record per run; refuse, before anything runs, pins the sealed Candidate was not sealed under, and before each launch when the gate, the hold, the environment, placement of protected paths (also checked at setup; by name and by device and inode), the entry, size or deadline says so; a write-once snapshot store for fixture inputsrepository-collectorguide
Product gate and delivery recordsOne journal aggregate per Product: store binding, grants and revocations, stop intent, fence-proved owner, frozen Slices with allowances, admissions and outcomes, with integration admitted only on a Verdict re-derived inside the decision; beside it, create-once delivery records and a pure Assessment derivationproduct-gateguide
Delivery (D4/D5 candidate)Binding, frozen inputs, proof rederivation, guarded original integration and read-only recovery/inspection/assessment on disposable local Products; trace, export and live commissioning remain separate. 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.deliveryguide
Factory Workflow (concert candidate)Stateless prepared-prefix native composition over thirteen stage kinds; original module owners retain authority, budgets and receipts. 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.factory-workflowguide
Verification Workflow (W31 candidate)Exact native external seal and original proof recovery under Delivery authority, including controller-death controls. 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-workflowguide
Implementation Workflow (W30 candidate)One caller-bound pre-seal job and original Execution Attempt with confined worker Evidence and bounded proposal/Candidate linkage; no-resend recovery preserves Product authority. 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.implementation-workflowguide
Checkout Git (R7 candidate)Bounded committed-blob bootstrap reader with pinned commit and explicit Git hardeningcheckout-gitguide
Producers (Claude / OpenAI)Turn one validated request into at most one provider exchange and return an untrusted Proposal for Repository sealing, or an explicit non-success; never applies, integrates or verdictsproducerguide
Lifecycle scheduleOccurrence identity, the discovery-window ledger and deferral for the proposed discovery design: one journal aggregate over an injected clock; admits each provider call before contact and runs nonelifecycle-scheduleguide
Discovery sources (lifecycle increment 2)Private adoption draft: required caller cache in the shared core, compatible Factory facade. Fetch a Product's commissioned public Sources (allow-listed https URLs, robots.txt first, fixed headers, no cookies, credentials or redirects) into Captures with hashes and novelty, a coverage status per Source, and 15-word Citations that re-verify from a 35-day cache; limits refuse, never truncatediscovery-sourcesguide
Service accessNames-only Service discovery, catalogue and deny-first configuration scanning; exclusive store bindings, declarations, route pins and readiness; fake-tested vault item presence; no credential values or live credential useservice-accessguide
IntelligenceOne registry of where a model takes part: each module's decision (none, optional, required) and why; for each task an owner-approved ordered list of up to four entries (provider, route, model, effort); routes and dated model facts as checked data. A pure selector takes the first entry no recorded fact refuses, records why it passed over the others, and refuses rather than substitutes. Imports only canonical JSON; nothing imports itintelligenceguide
Owner policyRecords owner asks, answers, minutes, stated rules and channel coverage; derives weekly counts, never applies a rule or grants authority. Beside the ledger, a separate sf-owner-rules/1 aggregate keeps Factory-drafted rules the owner confirms by exact-wording digest, answers whether an action needs the owner (ruled, ask or unknown), and records uses, "should have asked me" demotions and lapses; rules record owner intent and grant no execution authorityowner-policyguide
Branch healthReads GitHub branch checks and workflow history, retains failure episodes and mirrors blockers into Portfolio; no live poll is commissionedbranch-healthguide
Capacity planningConditional native-unit scenarios with caller-attributed evidence and fixed residual allowances; no admission or dispatchcapacity-planningguide
Provider subscriptionsBounded private lifecycle and caller authority/contact fence; original integration passed 1,976 full/198 package-doc checks. Current correction independently accepted; corrected stack 6f4ff02 passed 2,045 full checks; combined package/public independently accepted; accepted locally 1 October 2026provider-subscriptionsguide
Provider Events RPCCorrected bounded host-adapter source/package independently accepted; Root's historical 1 October 2,121-check combined full run passed; accepted locally 1 October 2026; final publication review, CI and live qualification remain openprovider-events-rpcguide
Provider eventsPure pinned Branch health poll projection, explicit caller permission and bounded filters; lifecycle belongs to Provider subscriptions; the separately commissioned Provider event delivery source candidate has scoped native source048 installed evidence; 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 proofprovider-eventsguide
Owner digestForms cited exception batches and bounded weekly digests; fresh green facts require complete passing evidence, and new red facts require an observed failure or open red episode. Historical /2 records replay unchanged; no delivery channel is activeowner-digestguide
Prevention register (FB-H7r draft)Three frozen guardrail templates turn caller-supplied reads into one presence fact per Product and template, never a default pass; increment 2 records results and owner exemptions in a caller-supplied journal and checks them on replay; integrated 2 October 2026 after independent review; not accepted, with no rollout, package entry or live readprevention-registerguide
Provider event delivery (EV3 candidate)Corrected bounded sender source independently reviewed; separate installed behavioural family passed scoped source048 evidence 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; separately scoped source-review and live connector qualification remain distinctprovider-event-deliveryguide
Provider diagnostics (candidate)Passive attributed facts preserve native outcomes; independent source review passed; historical 7cf bounded installed/full execution and independent local evidence acceptance passed; separately scoped final source-review gates remain distinctprovider-diagnosticsguide

The Owner digest red-evidence correction changes fresh-fact admission only; stored /2 records and delivery authority remain unchanged. Current gates belong to build review.

The Project guidance command contract reserves new save identities while retaining recorded legacy retries. It refuses surplus history unless a later valid same-Product aggregate accounts for it, and retains the captured revision under a real concurrent commit. Historical correction record; current gates belong to build review.

4. Lifecycle: what exists and what is planned

Need → Slice → Candidate → Verdict → Receipt → Assessment

The fixture CLI composes bounded execution and verification for four synthetic fixtures. Repository delivery parts exist separately; they do not yet deliver a customer change end to end. See the lifecycle explanation, proposed discovery and post-development design, and acceptance evidence.

5. Cross-cutting invariants every agent preserves

Each line names where the rule already lives, so a change can be checked against it.

  1. Unknown is not zero, and not failed. Portfolio Counts are exact only under declared complete coverage; the Dashboard renders unknown as —; Assurance folds a timeout, signal, collector error or missing reference to Inconclusive; the executor's unresolved and Repository's unknown are neither failed nor done.
  2. A historical Verdict is not current validity. Guarded verification shows historicalVerdict as issued and recomputes currentValidity from the Evidence and the source digests now; any source drift invalidates without rewriting the Run.
  3. SQLite and Git (or any provider) are never one transaction. Journal commands wrap no external effect; Repository's claim, target move and receipt are separate single-ref updates with no crash atomicity claimed; the journal's replay avoids re-running the decision only, so a caller's other effects can repeat.
  4. An unknown effect never permits a blind retry. invokeOnce throws only before submission and never downgrades unknown to "not applied"; after a lost acknowledgement, repeat the same command ID with identical inputs (journal, Execution, guidance, Product gate) and let the replay say whether it committed. The Product gate blocks every later integration of a Product while one is null or unknown, and its REPLAYED answer is history, never a permit.
  5. The implementer is never the verifier, and output has no authority. The worker and the executor observe and never judge; a producer's passed flag, printed control text or exit claim never changes admission or a Verdict; hand-built records never qualify. Model assignments follow the current working agreement; independent review is always separate from implementation.
  6. Products stay isolated. One repository binds to one Product; a Candidate manifest includes its repository identity; guidance is never copied between Products; the executor denies reads of another Product's files; usage is attributed once and never moves; the Product gate keys every record, ID and pin by Product, so one Product cannot integrate with another's Candidate, proof or store; Service access (increment 1) binds each repository and vault to one Product and hashes the Product into every reference id.
  7. Everything is bounded; input limits refuse rather than truncate. JOURNAL_LIMITS, EXECUTION_LIMITS, worker and executor bounds, REPOSITORY_LIMITS, PORTFOLIO_LIMITS, GUIDANCE_LIMITS, DASHBOARD_LIMITS, REPOSITORY_ASSURANCE_LIMITS, GATE_LIMITS, RECORD_LIMITS, SCHEDULE_LIMITS and (Service access, increment 1) ACCESS_LIMITS bound what modules accept, and oversized input is refused, not cut short; so does DISCOVERY_LIMITS (lifecycle increment 2), for fetched bodies, items and Captures. Presentation is capped instead: the worker and the executor keep a bounded prefix of process output, report the overflow and request termination (local-worker.ts:601-612, confined-executor.ts:1022-1031); diagnostic and error messages are shortened to fixed lengths (for example DASHBOARD_LIMITS.maxMessageLength); an attention item lists at most MAX_WINDOWS (5) source windows. A Budget is never extended; recovery loops have round counts; there are no leases, heartbeats or takeover by time.
  8. No human gate for self-solvable work. Attention infers nothing: empty guidance, unresolved verification, dirty files and zero counts make no item; ownerRequired needs a source that names the owner and an exhaustion; a missing fact is never an owner approval request.
  9. Never weaken tests, acceptance or pins. Drift blocks dispatch by design; never re-pin silently; never relax a corrupt check; a stored-shape change is a protocol revision, not an edit in place; a reused command ID with other content is a conflict, not an update.
  10. Validate before writing, fail closed, and never adopt, recreate or repair silently. Journal open checks required tables and columns before WAL/FULL configuration. Missing or replaced fences and repository roots are refused; the executor quarantines changed namespaces. openExisting and openReadOnly refuse missing journals, while open may create one. Journal identity is not fenced: a valid replacement is accepted. Cleanup is non-recursive and uncertainty is quarantined and reported.
  11. Local proof is not release, and neither is customer value. Fixture recovery is not customer-release recovery; a merge, an upload, a screenshot or a green suite establishes nothing about availability or benefit; all 42 epics stay open until their own evidence closes them.
  12. Plumb is editorial. The character belongs in docs, onboarding and calm empty states; in the dashboard the mark sits beside the wordmark and Plumb appears only on an empty Portfolio that a successful current read confirms is calm (calmEmptyPortfolio in apps/dashboard/src/brand.ts); failures, security, destructive actions, blockers, privacy, spend, uncertain effects, Verdicts and Assessments use plain words, no character, no colour-only meaning (brand guide).

6. Trust scope summary

Trust is scoped to each module's tested boundary. Read the module reference before composing it. Fixture recovery does not establish release recovery. The confined executor is not a general writable coding sandbox; hard RAM and disk limits and observer-loss cleanup are not established. Fresh Factory-owned repositories do not establish safe self-update of an existing product.

Build review owns the accepted scopes and unresolved gates. Repository, Confined executor and Producers specify their additional limits.

7. How agents work here

Use the observability catalogue for shared event names, identifiers, diagnostics and protocol vocabulary.

The structural graph checks use the compiler to check runtime and type-only dependencies, pure leaf closures, capability import sites, both pinned closures and restricted importers, including nested files. Boundary tests retain the existing safety restrictions; E29 adds neutral Attention/digest cores and checks that neither loads Portfolio, Dashboard or Branch health providers. Factory adapters retain those integrations. Some pure decisions load I/O modules for constants, validators or error classes; those value imports are explicit exceptions, not pure dependency chains. Owned journal prefixes are checked separately from legacy caller-chosen IDs, whose collisions remain possible. These are static coupling checks, not a sandbox or a live delivery claim.

Follow AGENTS.md for authority, model assignments, independent review, worktree isolation and delivery. Use local development for commands, receipt checks for coverage, and documentation ownership before editing prose. Preserve old receipts as historical evidence; add new receipts for new review.

8. Design language

Read this before building or changing any UI or docs page. The design system is Factory · Plumb (version 1790613544-ec35), iterated in Claude Design; its link is in the Notion status, not in this repository. docs/brand/tokens.json is the canonical copy of its tokens, byte for byte as exported; design language says where each token goes and how to re-sync.

  • Tokens only. Use the token names and values (--ink, --blue, --space-4, --radius-md, --rail-width, --focus and the rest). Add no colour, family, type size, spacing, radius, shadow or layout size that is not a token, and never edit tokens.json by hand: change the design system, then copy its export over. The design system has no token for a reading measure or a minimum width yet; the docs site names its few such measures once (SITE_MEASURES in the builder), and its only other lengths are 1 to 3 px lines and offsets.
  • Light only. One theme. Add no dark scheme and invent no dark palette until the design system has one.
  • Colour. Flat surfaces: no gradients, glow, texture or raised shadows. blue is for links and the one primary action per view. ok, warn and problem mean status only, on their status grounds, always beside a word and never in artwork; ok is for icons and borders, not text. rail, rule and select are decorative, never the only boundary or state.
  • Type. The sans and mono system stacks, no font files; weights 400, 500, 650 and 750 only; the named type styles (page-title once per view, body, supporting, code and the rest). The wordmark is live text, never an image.
  • Space and shape. Spacing space-1 to space-12 (4, 8, 12, 16, 24, 32, 40, 48 px) with nothing in between; radii radius-sm, radius-md and radius-lg (4, 6, 8 px), and nothing is a pill. Controls are control-height; targets are at least touch-min. At 1280 px and below the rail is 224 px and the inset 32 px; at 960 px and below the rail becomes a top bar and the inset 20 px.
  • Focus. Every control shows the focus ring, the only shadow, on keyboard focus; links and linked rows show a 3 px blue outline.
  • Status. Always a word, never colour or an icon alone. Unknown is an em dash with its reason, never zero.
  • Plumb. Docs, onboarding and calm empty states only, at illustration size, below the words, never replacing them, never clickable, with empty alt text when the words carry the meaning. Never in tables, metrics, navigation (the mark excepted), decisions, blockers, failures, spend or unknowns. The mark is never below mark-min, and the favicon stays favicon.svg. Module plates are shown whole, never cropped; nicknames are captions only.
  • Voice. Warm, brief and curious, in British spelling and the Factory's domain words; plain words for limits, status and anything that matters; no emoji, exclamation marks or hype. The say-and-not table is in the brand guide.

How parity is tested. test/design-tokens.test.ts fails when tokens.json stops being one light theme with system fonts, and reads both stylesheets and the docs' drawings with allow-lists, so a value it does not know fails. The allow-lists cover colour, type, radius, spacing, shadow and outline; custom properties and var(); resets; at-rules; the responsive values at 1280 and 960 px; and the drawings' elements and fills. The design language lists the rules; scripts/render_docs.py draws the diagrams from tokens.json. Its mutation cases show each check failing on altered copies. test/docs-site.test.ts checks the site's status words, notices, Plumb placement and whole plates on every page. After any UI or docs change, run node --test test/design-tokens.test.ts test/docs-site.test.ts and rebuild the site.

Where it is iterated. Visual changes start in Claude Design (link in the Notion status) and reach this repository only through a re-sync of tokens.json, followed by the stylesheets that use it.

Continuation evidence

Current continuation evidence is maintained in the build review and the handover. Historical receipts remain immutable.

Local bet commands

The command guide exposes one Branch health poll, Owner policy writes and read-only Owner policy views (which answer "no-ledger", never zero, and never word a rule as leave to act), and Owner digest batches, digests, notes and the record that a formed digest was sent, through existing local contracts. Poll retries retain their instant and watch target; a changed repository or branch at that instant conflicts. The wrapper creates no journal or Product and sends no message. Live use remains uncommissioned.

Delivery core refactors

Repository's wire grammar is shared by Assurance, the gate and delivery records; input readers and effect checks belong to the gate/records module. These helpers add no model or delivery runner. The collector's changed closure requires new pins for new work, while historical records remain intact. Review and verification are recorded in the R3/R4 acceptance record.

Adoption scope index

The adoption inventory preserves the fixed E35 audit and separately indexes later Module and Workflow scopes. Each module reference owns its contract; build review owns current evidence and gates. Old receipts remain immutable.

Delivery composition in progress

Delivery joins binding through verification and a guarded, single-occurrence owned local integration on disposable Products. R7 supplies its bounded committed-blob reader. Current integration re-derives proof before admission and submission, retains prepared/sender records, reconciles only the original dispatch read-only, and separates historical receipt from present target/proof validity. Inspection and Assessment are read-only. Stage CLI/package and combined native execution passed for the bounded local profile with independent evidence acceptance. Complete D4/D5 Workflow acceptance, trace, export and live commissioning remain separate. Existing qualification pins and historical receipts remain unchanged.

Shared value implementation

The shared libraries consolidate equivalent freezing, Milestone date/identity and bounded Promise-observation profiles. Existing Module contracts, errors, stored identities and authority remain intact; audit and remaining debt. New source bytes stale current collector recipes; historical pins and receipts are not rewritten.

Neutral adoption contracts, 3 October 2026

Schedule /2, digest presentation /1 (sf-owner-digest-presentation/1), Product binding/gate /2 and caller-pinned Claude policy /2 are additive closed profiles over the existing implementations. Dashboard /4 has explicit dual-client compatibility. Factory retains its legacy consumers; no stored records or historical receipts migrate. Current verification gates distinguish source review from installed, full-suite and CI acceptance. All 42 epics remain open.

Source: docs/agents/factory-model.md