eval(deltawire): superseded v5 experiment provenance

[bigboateng]

DeltaWire v5 — superseded experiment provenance

PR #111 preserves the terminal v5 experiment and must not be merged into main.

Terminal result

• Exactly one paid D1-only range-large-v5 probe ran.
• Trial ID: 574138ea-c743-4a05-b34c-cd578d24f49f.
• Harbor agent setup ended with AgentSetupTimeoutError after the 360-second setup boundary.
• The DeltaWire environment receipt passed.
• Harbor artifact collection and the artifact-manifest receipt passed.
• No structured agent trajectory, DeltaWire plan, output, semantic result, or token usage existed.
• No probe-release.v3 or pair was created.
• No rerun, canary, or 72-run experiment occurred.

Classification

agent_bootstrap_infrastructure_failure

This was not a Gemini model failure and not a DeltaWire semantic failure. The built-in Harbor Gemini CLI bootstrap timed out before agent execution or a model call began.

Evidence

• Evidence commit: 8c7f40eb96db71b3a5ed641277026adc3e4e2ecd
• Trial evidence: labs/20-deltawire/eval/results/preflight-v5/treatment-probe-range-large-v5/
• Evidence-head CI:
  • https://github.com/operatorstack/intelligence-flow/actions/runs/29933756442
  • https://github.com/operatorstack/intelligence-flow/actions/runs/29933761169
  • https://github.com/operatorstack/intelligence-flow/actions/runs/29933761869

A separate clean integration PR will project reusable infrastructure without raw Harbor evidence. PR #111 will be closed after that clean integration merges.

READY_FOR_72=false.

[bigboateng]

v5 paid probe — terminal evidence (2026-07-22)

The single authorized D1-only range-large-v5 probe was launched exactly once from frozen commit 8ef1937e97800efa2b061009a8e5218503ffb195 with manifest SHA-256 6f99b8b29181b4bf3021397e4c3d1b7827f1850f3d1cc25ddb638f021ccdbf57.

Outcome: failed closed during agent setup. Harbor recorded AgentSetupTimeoutError after 376.568892 seconds, zero successful trials, and no retries. Deterministic environment and artifact transport passed, including binary/config/schema/public-contract hashes. No structured trajectory, DeltaWire plan, output, or token counts were produced, so treatment-use, plan-contract, semantic, and end-to-end receipts are red.

Evidence commit: 8c7f40eb
Approval receipt SHA-256: 86b5a778af39d4b41046ac85d55967ee3890580ac4dcab2c1555247c836d1b1b

The 21 published evidence files match the frozen VM byte-for-byte. Exact credential-byte scan: 0 hits. Generic secret scan: 0 hits. Full local v5 validation: pass.

Per the one-shot policy: no probe rerun, no probe-release.v3, no pre-pair report, and no pair. READY_FOR_72=false. Evidence-head CI is pending and will be reported separately from manifest-head CI run 29930309800.

[bigboateng]

Evidence-head CI

All checks on evidence head 8c7f40eb96db71b3a5ed641277026adc3e4e2ecd passed:

• v5 branch workflow: https://github.com/operatorstack/intelligence-flow/actions/runs/29933756442
• v5 pull-request workflow: https://github.com/operatorstack/intelligence-flow/actions/runs/29933761169
• main validation workflow: https://github.com/operatorstack/intelligence-flow/actions/runs/29933761869

These are evidence-head checks. The distinct frozen manifest-head CI remains https://github.com/operatorstack/intelligence-flow/actions/runs/29930309800.

PR #111 remains open and ready for review. It was not merged. The v5 probe is terminally red and the pair remains unauthorized/unlaunched.

[bigboateng]

Superseded by clean integration PR #113. The raw v5 evidence branch and evidence commit 8c7f40e are intentionally retained; no data was deleted. The v5 Git-object evidence lock merged through #113 verifies all 25 frozen v5 evidence files.