[ADR-0007] Record the markets pilot's measured baseline and post-collapse cost in the package (§1, §7, §11) #2327

Why

The ADR-0007 markets pilot (#1729) is finalized. Its measured cost and final status are recorded in the #1729 comment #1729 (comment). The package doc docs/initiatives/agent-ingress-collapse-markets.md still has three gaps:

• §1 does not carry the baseline run counts, even though AC5 cites §1.
• §7 states the skip-billing ruling but has no measurement.
• §11's gap list predates the merge.

This is a docs-only change. Copy the figures below verbatim. Do not re-measure.

Do this

1. §1 Baseline (AC5)

After the stub/pin table, add a "Measured baseline (billed runs)" subsection with:

• Window: 2026-09-15T00:00:00Z..2026-09-22T00:00:00Z, 7 days, bounded at both ends. Each count is a per-workflow total_count from actions/workflows/<stub>.yml/runs?created=<window>.

2. §7 Skip-billing (AC10)

Keep the existing ruling text. Add a "Measured (post-collapse)" subsection.

Method:

• Post-collapse window: 2026-10-10T01:16:00Z..2026-10-14T01:16:00Z, 4 days, from the markets#513 merge.
• agent-ingress.yml had 382 runs. Each of the five deleted stubs had 0.
• A job counts as billed when runner_name is non-empty. Billed minutes are per job, each rounded up to a whole minute.
• Job sample: 1340 of the 1342 baseline runs (2 returned no jobs), and all 382 post-collapse runs.

Per-day figures:

Other figures to include:

• Workflow runs/day: 191.7 → 95.5 (−50%).
• Jobs evaluated/day, including skipped: 358.6 → 626.0 (+75%). These are all kind (a): 0 jobs concluded skipped with a runner assigned, in either window.
• PRs opened: 5 in the baseline window (0.7/day) and 7 in the post-collapse window (1.75/day). Post-collapse traffic was not lighter.
• Raw totals: baseline 552 billed jobs and 663 billed minutes; post-collapse 281 billed jobs and 317 billed minutes.

Conclusion to state: the collapse is cost-neutral to slightly cheaper for markets. The per-event if: filters skip union events before a runner is assigned. The main saving is pr-auto-review: the old stub gave a runner to every run (208/208), while the ingress job is billed for 95 of 382.

3. §11 Honest gap list

Bring each row to its final state and cite the closing item:

• markets PR: done, markets#513.
• Companion issue: filed, markets#506.
• Canonical template: published, fix(standards): sync agent-ingress template job blocks to markets (pr-auto-review, pr-review, ci-failure-analyst) .github#1316 (452ea69), outside DEPLOYABLE_WORKFLOWS.
• STUB_REGISTRY rows: applied (feat: implement issue #1729 — [Phase 3] Collapse the first repo's Class-1 event-driven stubs into agent-ingress.yml and rename the required checks #2182), with an ALIGNED run (Fleet Monitor run 38193525415).
• Pin corrections: already final, keep as is.
• Baseline doc: now resolved in §1, by this change.

Add one row:

• AC7 ci-failure-analyst live execution: not observed. GitHub Actions' own check runs deliver no check_run event (CI Failure Analyst never runs for failed GitHub Actions jobs: the check_run trigger only fires for third-party checks #2235), and no third-party check failed on markets in the window. Accepted by the maintainer when [Phase 3] Collapse the first repo's Class-1 event-driven stubs into agent-ingress.yml and rename the required checks #1729 was finalized. The guard is verified by markets' bats test and by package §3/§3a.

Do NOT

• Change §3 (it must stay byte-identical to markets main), any ADR, scripts, or tests.
• Re-measure, or alter the figures.

Acceptance

• §1, §7 and §11 carry the content above.
• bats tests/agent_ingress_markets_package.bats tests/fleet_stub_registry.bats stays green. In particular, the §3-to-fixture byte-identity test must not be affected.
• Only docs/initiatives/agent-ingress-collapse-markets.md changes.

Refs #1729, #2177, combsmith/markets#506

Generated by Claude Code