Inputs: brainstorm, acceptance, ADR-062. Owner final scope supersedes the initial three-pipeline draft. Lead integration owner approves the bounded contracts under the owner's explicit CI/Benchmarks separation direction.
Parallel workstreams: provenance audit, disjoint C# protocol/regression updates, and production isolated artifact adapters; lead alone owns workflow YAML, shared policy, feature/ADR integration, site handoff and delivery. The existing legacy CI producer must not be silently renamed in retained historical JSON.
| Task | AC | Owner/tier | Permissions/ownership | Dependencies/start | Artifacts/verification/join |
|---|---|---|---|---|---|
| TASK-WF-AUDIT | 003/004 | capable read-only worker | inspect scripts/tests/pages, no writes | accepted scope | exact identity routes and migration risk report; terminal complete/blocked/failed/cancelled |
| TASK-WF-CS | 001/002/003 | cost-efficient capable C# worker | named clean C# comparison protocol/fixture files and new unit layout tests only | contracts and lead naming decision | diff + source checks; no local tests; lead review before join |
| TASK-WF-SCRIPTS | 003 | cost-efficient capable JS worker | isolated producer scripts only, explicit named files | audit and accepted new identity | diff + syntax checks; no local tests; exact provenance remains fail-closed |
| TASK-WF-ROOT | all | lead planner/integrator | workflows, policy, docs, site adapter/constants | serialized contract ownership | review all diffs, static checks, scoped commit/push, GitHub results |
Ordered steps and done conditions:
- Read root/local policy and architecture; record owner correction first.
- Write brainstorm, detailed AC/test matrix and this task graph before workers.
- Inspect baseline full relevant GitHub CI/Website failures at existing main SHA; track failing job/step/root cause before attributing any result to this change.
- Approve ADR-062 implementation contract: final producer Benchmarks; current CI/Website names change while historical KeyLoad CI main-push identity remains exact.
- Join read-only audit; update only new isolated producer identities, retaining authentic legacy historical CI proof. Delegate bounded C#/script scopes in parallel.
- Move comparison graph and TimeSeries image checks into benchmarks.yml, with complete prerequisite build/format/rules; move standalone rules into standard CI.
- Update Pages completion trigger and matching measured-source producer copies.
- Add/update TUnit regressions derived from AC; inspect every worker diff and exact native job/step/artifact names. Workers cannot commit or push.
- Run static governance, syntax/whitespace and integrated source review. Preserve all pre-existing dirty work; stage only task-owned hunks/files.
- Commit and push the stable layout milestone on current main. Inspect actual push-triggered Tests/Benchmarks, manual CI/Release and Website exact-SHA jobs. No duplicate full dispatch.
- Download changed-job artifacts, record actual SHA/run/job URLs and gate status; fix layout/provenance failures and repeat relevant GitHub checks if required.
- Close all AC or state exact pre-existing/external blockers without claiming failed/skipped/incomplete measurement, readiness or publication success.
Baseline failures: main9b3bd8ed64b42e7e3bd37d454face7984dfc6b5f CI37111280400 and Website37112008157 fail. CI verify job111169530401 fails only the existing AcTsi006UnknownProblemNamesAndUnavailableStatusStayUnknown assertion at line44; that test file was already dirty before this task and remains outside this scoped layout change. Website qualify job111171578311 fails authenticated comparison selection before tests (no complete isolated producer), followed by missing native coverage inputs. Governance37111280336 succeeds, but its standalone workflow contradicts owner-directed layout. These remain existing qualification gaps; never weaken tests, use partial results or invent coverage to bypass them.
Validation guidance: no extra skills are applicable to this bounded infrastructure change; install none. TUnit/MTP runs only in GitHub. Every new assertion targets caller-visible pipeline composition/provenance rather than performance arithmetic. Build precedes tests; full ordinary CI and preserved native comparison gates remain. Existing incomplete product qualifications are not silently included as new repairs.
- TASK-WF-ROOT separates CI PR checks from Tests main/PR/manual project suites, names Benchmarks/Website plainly, and adds Release tag/manual package builds.
- TASK-WF-CS updates the bounded C# fixtures and minimal source-contract regressions to all five final workflow roles/names; no runtime behavior changes.
- TASK-WF-SCRIPTS updates exact isolated Benchmarks routes; adds explicit current CI metadata name separately from historical legacy KeyLoad CI run name.
- TASK-WF-AUDIT reviews final migration and package/provenance boundaries.
No public package publishing is requested: Release retains real generated NuGet packages as run artifacts. Current CI removes main-push triggers, so legacy push history can remain strictly KeyLoad CI while its current metadata is named CI. Final shared contracts are lead-owned; workers receive disjoint named file scopes.
Both bounded C#/JS packets are COMPLETE and joined after lead diff review. The independent audit's undefined EventBlock call and incorrect RF3 assertion location were corrected before staging. Ruby parsing confirms the five final names; exact original native job definitions remain equal apart from their prerequisite. All seven modified JS modules pass node --check and scoped whitespace checks pass.
The shared checkout acquired an unrelated new Diagnostics project during this task; its in-progress inventory briefly fails static governance. An exact current HEAD archive plus only this task's source changes passes the existing validator: 25 projects,4 modules, original policy prefix and unchanged skill inventory. The sandboxed development build returned exit1 with0 warnings/0 errors after 5min and is not a passing build result. Required compilation/runtime evidence comes from the exact delivered GitHub workflows. No local tests were run.
Commit 9e0532fdeed7c55a17f9c857f8d5a264215341e7 is pushed to main.
The authenticated workflow inventory contains exactly the five active names
CI, Tests, Benchmarks, Release and Website. CI run
37113420787
passes repository rules, the complete Release/format build and analyzer tests.
Release run
37113424709
passes and retains nine actual 0.1.0-dev NuGet packages. Downloaded package
versions, sizes and SHA256 values agree with its source/run-bound packages.json.
Artifact packages-0.1.0-dev has authenticated id 11271266373.
The qualification receipt retains the exact
run/job/artifact identities, verified package manifest and real image test summary.
Benchmarks run
37113337516
passes its complete build/format/rules prerequisite and the moved real pinned
TimeSeries image job 111176089078. The unchanged native resource-model suite
then fails four Timescale image-tag assertions (expected 2.30.2-pg18, actual
empty tag). The incomplete cohort cannot refresh website metrics.
Tests run
37113337508
passes complete build/format/rules and 118 analyzer cases. All four new
workflow-layout regression cases pass, as do all 183 process-recovery cases.
The ordinary unit suite passes 1856/1857; its sole failure is the same pre-task
SDK-status assertion listed in the baseline above. RF3 passes 62/63 and reports
one existing SQL admission assertion
AcAisql007InsufficientDataBytesRejectSdkAndOfficialSqlBeforeIdClaimWhileDirectControlKeepsRf3Healthy.
The moved RF3 job definition is unchanged. Downloaded genuine TUnit reports bind
these counts and focused assertions to the exact milestone source SHA.
Website run 37113337499 fails authenticated comparison selection and missing coverage inputs, matching the pre-task publication blocker. The complete isolated archive needed by the SiteTests before-session setup remains unavailable, so a new source/provenance test is compiled but cannot be counted as runtime-qualified. Publication gates remain intact. These failures do not establish successful database, benchmark or website qualification. The follow-up preserves the original PR trigger and PR cancellation in the separate Tests workflow.