Purpose governance conformance evidence

Start with Purpose, inceptions, signals and limits for definitions and a shared support example.

This is a source compatibility record for the bounded Room composition, not a capability-baseline promotion, production certification or npm release. The support scenario replaces the earlier LTV/CAC demonstration by owner choice.

Contract-to-test mapping

Tests prove bounded behaviors under their fixtures. They do not prove every possible concurrency schedule, every external adapter, judgment quality or production resilience. Actual PostgreSQL tests use isolated schemas and explicit integration enablement; memory-only tests cannot stand in for durable evidence.

Reproduce focused qualification

pnpm build
AGENTPLAT_POSTGRES_TEST=1 PGHOST=127.0.0.1 PGPORT=5432 \
  PGUSER=postgres PGDATABASE=postgres \
  pnpm run verify:purpose-governance /tmp/new-purpose-evidence
pnpm run verify:purpose-consumer

The output path must be new, with an existing parent. The focused gate builds the required package closure, checks public types, runs Rooms/API/PostgreSQL/Workflow and support-demo tests, rejects skips/TODOs, checks architecture/specification and adoption links, and records exact commands, TAP logs and source SHA-256 hashes. Database credentials are not written into the evidence record. Use a disposable database with schema-creation privileges; tests delete only their own schemas.

Packed verification retains the existing consumer check for Collective Runtime and Audit and adds the four governance packages, their dependency closure and TypeScript fixtures. It verifies local tarballs; it deliberately refuses registry mode as a source-consumer gate. Published 1.1.0 registry verification is recorded separately in the distribution record. It does not publish packages or change release flags.

Broader validation and limits

Objective 10 additionally runs the full workspace build, full type-check, root unit suite and adapter suite. Infrastructure-specific skipped/TODO tests in those broader suites must be reported separately; they are not treated as passed. The focused governance gate requires no skipped tests.

The initial broad run in the development checkout (which also contained the unpublished optional Jev integration) exposed stale test expectations: six adapter admissions rather than five; 66 catalog entries with 65 publishable packages and one explicitly unpublished assessor; and updated release-cohort rejection wording. Those development-checkout tests were aligned to its existing policy. The delivery branch starts from main and excludes the unrelated Jev integration and its test adjustments. Neither catalog nor release eligibility is widened by this delivery.

See adoption for host obligations, incremental rollout, safe instruction fallback and rollback refusal. The sequential source plan ends with this qualification; publication and deployment remain separate work.

Local verification record — 2026-09-28

Environment: Node.js 24.20.0, pnpm 11.25.0, disposable PostgreSQL 16.11 on loopback. No user databases, deployments, registry publications or release flags were changed.

The broad suite's skipped digest-pinned container test and six pre-existing Mesh work/lease TODO cases are not counted as successful evidence. Local PostgreSQL coverage does not imply that a live Temporal server or Redis service was exercised. CI now includes the focused database gate and packed consumer in addition to its existing checks. The edited workflow was not dispatched remotely in this task.

The default whole-worktree public audit still rejects unrelated, pre-existing multimedia files under output/ (unsupported binary types and size limits). Those files were preserved. An explicitly scoped diagnostic excluding only the listed multimedia files/directories passed, retaining all 27 frozen exact-file exceptions. It is not a substitute for the default release audit. A clean publication checkout must pass the normal release gate; no pnpm run check or release-readiness success is claimed here.

Isolated delivery qualification

The delivery branch is based on origin/main, separate from the development checkout containing optional Jev work. It excludes generated demonstration output, unrelated multimedia, uncommitted research materials and Jev-specific test changes. Historical counts above describe the original development checkout, not this branch's package cohort. The isolated checkout's normal public audit applies with no multimedia exclusions; delivery check results are recorded in the PR.

The delivery also updates existing overrides from fast-uri 3.1.6 to 3.1.7 and ip-address 10.4.0 to 10.5.1 to address advisories reported by the production dependency gate. No audit allowlist or release exception was added.