Expose full merge train feedback and read models

[cbusillo]

Objective

Make train state visible to PR authors and operators. A full train needs feedback beyond “selected PR”: queued position, current candidate, blocked-by entry, validation state, and next action.

Finish Line

Expose full merge train feedback and read models

Current Status

State: Planned.
Next action: Define feedback payloads once queue records define status fields.
Blocked by: Queue records; authz/feedback actor rules; UI/read-model direction.
Last verified: 2026-05-13; current dry-run records queue entries but does not publish durable position feedback.

Scope

• In: PR labels/comments, service read models, operator UI/API projection, and redacted audit summaries.
• Out: replacing GitHub checks UI or exposing secrets/provider tokens.

Acceptance Criteria

• Queued PRs receive clear feedback when they enter, move, block, reflow, validate, or merge.
• Later PRs can see which earlier PR or candidate state they are waiting on.
• Operator read models show active train generations and entries.
• Feedback writes are idempotent and do not spam comments on every scheduler pass.
• Authorization for feedback actors is explicit and audited.

Relationships

• Blocked by queue records.
• Related to Authorize trusted Every Code feedback actors #514 for trusted Every Code feedback actors.

Validation

• Unit tests for feedback diffing/idempotency.
• Service/read-model tests for redacted train state.
• Live dry-run or low-risk smoke verifying labels/comments are understandable.

Decisions

• Feedback must be derived from records and fresh provider evidence, not from log parsing.

Open Questions

• Should Launchplane use one evolving status comment per PR or multiple event comments?