[RobJacobson]

Summary

per-attempt plans, results and frozen-candidate tooling
  → one historical acceptance checkpoint
  → acceptance deferred until release preparation

Remove obsolete acceptance records, raw campaign material and fixed-candidate launch/probe/replay tools. Keep contract research, ordinary regression coverage and the small fixtures used by maintained checks. Update documentation and agent guidance so ordinary changes use checks for affected behavior and do not start another acceptance campaign.

This cleanup is isolated from the pre-existing implementation work. It is stacked on PR #101 so the diff contains only the cleanup; no application/library implementation, dependency manifest, version or real lockfile changes are included.

Evidence

• Documentation structure/local-link checks, historical Git target checks and staged whitespace checks passed.
• Mocked native-driver ownership/failure/cleanup checks, observation-writer filesystem regressions and retained-contract checks passed. Targeted formatting passed.
• Preserved the maintained checkpoint verifier's identity fields and relocated the writer regression input into a portable fixture.
• The primary checkout's branch, HEAD, staged/unstaged diff and untracked file contents match the snapshot taken before isolation.
• No acceptance tests, builds, native applications or device setup were run.

Merge Danger

Door: Two-way

Historical details remain in Git history and the previous PRs. The compact checkpoint records historical f05 outcomes and makes no current release-readiness claim.

Blast Radius: Acceptance

Old fixed-candidate commands are retired. Future acceptance requires a fresh, explicitly approved preparation for the source and dependencies being shipped. The cleanup does not certify the current implementation.