[lpcox]

Summary

• add deterministic measured pilot-100 and pilot-1000 tiers ahead of the existing full Phase 4 campaign
• keep pilot evidence immutable and explicitly promotion-ineligible, with exact sample/cell/minute ceilings and unchanged bounded resume/retry safeguards
• require successful same-baseline pilot-100 evidence before pilot-1000, and successful pilot-1000 evidence before full
• document exact initial dispatch and immutable resume commands for both pilot tiers

Pilot plans

Tier	Runs	Recoveries	Cells	Estimated runner-minutes
pilot-100	6	4	10	100
pilot-1000	96	4	52	1,000
full	960	40	76	10,000

The 100-minute plan covers both workloads on matched sequential/cold/default lanes, a matched concurrent build pair, and both recovery kinds on both runtimes. The 1,000-minute plan runs two matched samples across every workload/mode/cache/resource/runtime cell plus all recovery cells.

Validation

• npm run type-check -- --pretty false
• npm run build
• npx jest scripts/ci/nvx-benchmark-campaign.test.ts --runInBand --silent
• npx jest scripts/ci/nvx-matched-runtime-adapter.test.ts scripts/ci/nvx-promotion-evaluator.test.ts --runInBand --silent
• npx markdownlint --config .markdownlint.json docs/nvx-promotion-evaluation.md
• deterministic CLI plan/matrix generation for pilot-100, pilot-1000, and full

No pilot or full campaign was dispatched during validation.

[github-actions]

Documentation Preview

Documentation has been built for this PR.

Download preview artifact

To view locally:

1. Download the docs-preview-pr-9931 artifact from the workflow run
2. Unzip and open index.html in your browser

Built from commit c3c8d56

[Copilot]

🟡 Changes recommended

Progression validation can accept incomplete or mismatched evidence and select ambiguous rerun artifacts.

3 open findings

• Pilot success accepts missing evidence records · New
• Reruns select arbitrary aggregate artifacts · New
• Same-baseline gate omits runner image · New

What changed in this PR

Adds staged, deterministic Phase 4 benchmark pilots before the promotion-eligible full campaign.

Changes:

• Adds pilot-100 and pilot-1000 plans and progression validation.
• Extends workflow dispatch, resume, aggregation, and documentation.
• Adds pilot planning and progression tests.

File	Description
scripts/​ci/​nvx-benchmark-campaign.ts	Implements pilot tiers and progression validation.
scripts/​ci/​nvx-benchmark-campaign.test.ts	Tests pilot plans and progression.
docs/​nvx-promotion-evaluation.md	Documents tiers and dispatch/resume commands.
.github/​workflows/​nvx-phase-4-benchmark.yml	Executes and gates measured pilot tiers.

🧠 Review effort: Balanced

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[github-actions]

✅ Coverage Check Passed

Overall Coverage

Metric	Base	PR	Delta
Lines	93.17%	93.18%	📈 +0.01%
Statements	91.72%	91.73%	📈 +0.01%
Functions	90.38%	90.38%	➡️ +0.00%
Branches	85.48%	85.49%	➡️ +0.01%

📁 Per-file Coverage Changes (1 files)

File	Lines (Before → After)	Statements (Before → After)
src/log-directory-setup.ts	96.8% → 100.0% (+3.18%)	96.9% → 100.0% (+3.13%)

---

Coverage comparison generated by scripts/ci/compare-coverage.ts

[lpcox]

@copilot address review feedback

[Copilot]

@copilot address review feedback

Addressed in commit 57e65ed: pilot record IDs must match observed IDs, progression now checks runnerImage, and predecessor downloads are restricted to attempt 1.

[github-actions]

✅ Copilot review passed with no inline comments.

@lpcox Add the ready-for-aw label to this PR to trigger agentic CI smoke tests.