Phase 2d implementation plan
User specification: classify all 83 candidate-only failures A/B/C/D; preserve deployed SKCoord; green draft PRs; protected docs unchanged.
1. Snapshot every installed skcoord file on chiap08, verify host/local git blobs, scan, commit and push immutable snapshot first.
2. Create skcoord reconciliation worktree from current main; overlay deployed package without bytecode; preserve main improvements after comparison; run full sibling suite and open draft PR.
3. Inventory exact 83 failing test IDs from Phase 2c JUnit. Record one bucket, evidence, fix commit and result per ID in reconciliation/phase2d-20261003/failure-classification.json.
4. C: repair isolated AST globals, systemd argv stub and overlapping crew fixture scopes in individual labelled commits with relevant tests.
5. A/B: compare deployed rotator, wrapper, units and package paths against main history; update genuinely superseded tests, port main fixes for regressions, preserve main docs byte-for-byte. Each independently verified fix gets its own labelled commit.
6. D: trace captured scripts and transitive callers. Only existing deployed calls authorize connection wiring. List absent callers explicitly; keep their disposition pending user answer rather than inventing connections.
7. Pin hosted CI sibling overlay to the reviewed skcoord candidate SHA while retaining registry checks. Run clean dependency tests, targeted failures, full suite and hosted checks.
8. Record exact counts, baseline comparison, source provenance and PR state. Keep drafts; no merges, deployments or restarts. Stop after Phase 2d report.
