{
  "intent": "Act on two direct user verdicts: the statusline glyph told them nothing (drop it — the plain-words SessionStart line is the surfacing channel), and explain-diff is not needed at all (drop the skill entirely, do not compile it). Additionally stop the nudge counting candidates that already have a verdict.",
  "scope": [
    "python_pkg/session_autopsy/report.py (handled_ids; table marks handled rows; Reviewed section renders 'dropped'; write_report excludes handled from the count)",
    "python_pkg/session_autopsy/cli.py (mark-reviewed uses the same handled filter)",
    "python_pkg/session_autopsy/tests/test_report.py (updated + new assertions)",
    "~/.claude side (committed separately in that repo): statusline badge removed, skills/explain-diff/ and scripts/explain_diff.sh deleted, finish skill/script references cleaned, compiled.json verdict 'dropped'.",
    "Non-goal: existing docs/explanations galleries stay — they are inert committed docs."
  ],
  "changes": [
    "unreviewed_count now means 'significant AND not yet handled': candidates with any compiled.json verdict (compiled / keep-llm / dropped) leave the count, so resolving a candidate visibly shrinks the nudge (7 -> 4 on the real corpus).",
    "The ranked table's action column shows 'already handled — see Reviewed/scoreboard' for such candidates instead of re-suggesting a review.",
    "The Reviewed section renders dropped workflows as 'dropped (workflow removed)'."
  ],
  "verification": [
    {
      "command": "python3 -m pytest python_pkg/session_autopsy/tests/ -q",
      "result": "pass",
      "evidence": "81 passed; report.py and cli.py at 100% branch."
    },
    {
      "command": "PYTHONPATH=~/testsAndMisc python3 -m python_pkg.session_autopsy report; jq . ~/.claude/autopsy/state.json; statusline pipe-test; start-hook pipe-test",
      "result": "pass",
      "evidence": "unreviewed_count 7 -> 4 (err-549ef470, crdt-sync-migration, design-audit, update-config); statusline renders '[Fable] ~' with no glyph; SessionStart hook still prints its unchanged plain-words line (hook text untouched, so the mutation gate is unaffected)."
    }
  ],
  "risks": [
    "A candidate marked handled disappears from the nudge even if its compiled script later regresses; the scoreboard ('measure') is the instrument that would show that.",
    "Historical explain-diff docs remain in repos but the tooling to regenerate their gallery index (build_index.py) was deleted with the skill."
  ],
  "rollback": [
    "Revert the three python files to restore counting of handled candidates.",
    "The ~/.claude deletions are separate commits in that repo; git revert there restores the skill, template, and scripts."
  ]
}