test(ci): prove changed-base reruns and unaffected imports

[ooples]

ISOLATED ACCEPTANCE FIXTURE: never merge proof branches into master. Uses unchanged production CI and genuine Validation certificate from run 34965621059 / head 7bf8750. Base e1cb0ee adds executable ExponentialLoss zero-label rejection and three assertions in a concrete test class; the base was deliberately not prevalidated. Local production PlanDelta against the actual certified map and YAML manifest returns Partial: rerun Integration C-Core, D, H-L, N-O, and ModelFamily Generated Layers E (5/116); import unaffected Integration A-B. H-L and Generated Layers E were NOT in the source run, proving the new-shard edge rather than only intersection. Old fixture #2209 remains evidence of safe abstract-class fallback; no safety rule relaxed. Negative fixture #2208 completed with Validation Gate failure and no expensive proof. Required live evidence here: Partial resolver decision, actual five-shard build/test execution, new assertions executed, actual A-B artifact import, and aggregate/gate conclusions. None of those live outcomes is claimed yet.

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

🗂️ Base branches to auto review (2)

• master
• integration/.*

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Advanced

Run ID: b7fd147f-1e10-4dd0-b3a2-94aea83e98ca

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[ooples]

Live result so far: run https://github.com/ooples/AiDotNet/actions/runs/34980747565 resolver job 104420112781 passed. At 14:19:39 UTC it accepted source run 34965621059 and reported Partial, rerun 5 shards, import 1 unaffected PR shard. H-L and Generated Layers E are included despite being absent from source validation; A-B is the import. Select shards job 104420543955 passed; Build job 104421115668 is running. This proves the live changed-base routing decision, not completed tests or imports. Full execution evidence remains pending.

[ooples]

Completed live changed-base execution/import evidence for run 34980747565: Build 104421115668 passed. Exactly five test shard jobs executed: C-Core, D, H-L, N-O, Generated Layers E; no A-B test job reran. H-L and Generated Layers E (absent from source PR validation) passed. All three new ExponentialLoss assertions passed in H-L at 14:39:17 UTC. Test-result import job 104436139690, diagnostics import job 104436139821, and coverage import steps in 104436139646 all succeeded. Published ledger has six shards (five executed plus A-B imported), four passed, two failed, zero incomplete. Downloaded source ledger 057f896 from source run 34965621059 and target ledger d9a2ffc: all 2230 A-B records match exactly across every serialized field, zero differences. Both failing test identities/messages also match source failures exactly (DDP builder and InputLayer cloning). Validation Gate 104436453880 failed: this push did not enforce a baseline verdict, so raw shard failures remain blocking; expensive proof correctly not published. Thus selective changed-base scheduling, fresh executable build/tests, unaffected artifact import, and complete aggregation are VERIFIED LIVE. This is not an all-green run or proof that existing runtime defects are fixed. Sonar still running at inspection and is not counted as correctness evidence.