Reality probe: did Prefect #22698 now prove the two concerns from human review?

[hippoley]

This is a CounterProof Reality Probe for a real human CHANGES_REQUESTED review.

Source PR:
PrefectHQ/prefect#22698

Human review:
PrefectHQ/prefect#22698 (review)

@desertaxle @Sanjays2402 — the August 3 review asked for two concrete things:

1. stop reading variables.properties.volumes.default directly and resolve the normal job configuration contract first;
2. add a regression test proving a customized job_configuration.volumes preserves both its configured volume and the internal /tmp bundle mount.

The current PR head now contains tests for both. I independently replayed those two focused tests against the exact pre-change base.

Independent before / after replay

Exact commits:

• HEAD: 7604947bcaaec12a83505f842fbc6adb1b88275e
• BASE: ef4286181588e579f72854d9344f4a2f6d66a090

CounterProof run:
https://github.com/hippoley/CounterProof/actions/runs/35979249605

Result:

• HEAD: 2/2 PASS
• BASE + the exact same changed test file: 2/2 FAIL
• protocol: json-v1
• evidence mode: precise
• digest: sha256:4d16ed79a10dacbfe15f025c39dbf9c4f0d7eb724eaa7c4647182ed305b9434f

The two failures match the human review concerns

Default work-pool volume

BASE:

assert 'result-storage:/result-storage' in ['/tmp/...:/tmp/']

The old code kept only the temporary bundle mount and dropped the work-pool default.

HEAD passes.

Customized template volume + internal bundle mount

BASE:

assert any(volume.endswith(':/tmp/') for volume in call_volumes)

The custom custom:/custom volume is present, but the internal temporary /tmp mount requested in the review is missing.

HEAD passes with both.

Claim / evidence matrix

I wrote the compact matrix here:

https://github.com/hippoley/CounterProof/blob/reality/reviewer-requested-matrices/docs/reality/prefect-22698-claim-matrix.md

Current shape:

default work-pool volume propagation       WITNESSED
custom template volume + internal /tmp     WITNESSED
live Docker runtime                        UNPROVEN

This is deliberately not a “PR approved” badge. It only says the current head now has red-before / green-after behavioral coverage for the two concrete failure modes requested in the human review.

Question for @desertaxle

For a re-review like this, is this claim/evidence matrix materially useful — enough to reduce the work of checking whether your previous concerns were actually covered — or would you rather see a different artifact?

I am asking because another external reviewer independently asked CounterProof for exactly this matrix shape. I want to know whether that generalizes across reviewer workflows before making it a product feature.