🐛 - Commit rejects a verified generated SDD when the issue retains an unselected seed #959

Is there an existing issue for this?

• I have searched the existing issues and feature requests.

Which actions are affected?

Commit

Which platforms are affected?

GitHub Actions

Description

Commit reports validation.evidence-unavailable on a linked feature branch whose
generated SDD was verified by the Issue action. The native feature issue still
contains an optional marked seed. The maintainer used normal generation, so the
trusted published SDD card contains no seed selection or selection artifact.
The Issue action applies branched. Commit treats the unselected seed as if it
required selected-seed publication evidence.

Reproducing the issue

1. Admit a native Feature Request containing a valid optional SDD seed block.
2. Use normal SDD generation, with no seed-review comment or selection.
3. Observe a verified generated SDD, the linked branch and the branched label.
4. Push an implementation commit to that linked branch.
5. Observe Commit and Copilot / Verification fail before progress assessment.

copilot Version

vypdev/copilot@v3, ace49e4

Relevant Log Output

Commit could not verify the selected SDD draft for this issue and head.
Error code: validation.evidence-unavailable
Reference: 9eafb804-0883-4624-8468-02bfa41ba35e

Additional context and comments

Reproduction: vypdev/domus#92
Verified generated card: vypdev/domus#92 (comment)
Published SDD commit: 2f99e34b4f4d0d665fbb5a7a53e6c7f2962fd443
PR: vypdev/domus#93
Assessed head: 66d080f1f815107c57204267f7bb249db099403c
Failed Commit: https://github.com/vypdev/domus/actions/runs/42938506324

The documentation says an unselected seed is excluded from generation.
progress_analysis_workflow.ts calls the seed-publication query whenever a seed
block exists. read_progress_sdd_seed.ts requires an exact selected-seed card.
The published generated card has no selection, so that check cannot succeed.

Expected: when the trusted current publication is generated, progress should
use the generated publication and exclude the unused draft from scope. Preserve
fail-closed checks for an actually selected seed, missing evidence and stale
or mismatched publications. This differs from #958, which concerns retained
case identities after an issue edit.