This prerequisite exercise uses real issue #10 and merged PR #17. It is an attribution exercise, not a timed trial and not evidence for the go/no-go thresholds.
- Base:
5a2685cc709ca8f9f384e35e4d733012c7290d69 - Head:
442147d1bc9a33edeca1a605c9de3ada13f397f1 - Ledger setup: the squash commit was intentionally recorded as foreign so every resulting segment required an explicit manual assignment.
- Plan: P1 rename-reassignment regression, P2 literal-path regression, P3 evidence record, and P4 durable review rules.
| Plan item | Declared file | Assigned segments | Final attribution | Final scope |
|---|---|---|---|---|
| P1 | test/review.test.ts |
4 text | Attributed | In scope |
| P2 | test/plant.test.ts |
4 text | Attributed | In scope |
| P3 | docs/experiments/review-summary-edge-cases.md |
1 file + 1 text | Attributed | In scope |
| P4 | AGENTS.md |
2 text | Attributed | In scope |
Codeboost exposed 12 segments before assignment, all in the Unplanned row. Each segment mapped to exactly one item through that item's declared file contract. After the assignments, the per-item segment counts were 4, 4, 2, and 2; all four rows reported ✓ Attributed and ✓ In scope. No segment was unclear or multiply owned, so this exercise produced no ambiguity fixture. The earlier rename and literal-path concerns already have focused fixtures in test/review.test.ts and test/plant.test.ts.
The assignment was rerun from a detached worktree at the exact head using a fresh SQLite review store. The stored choices, rather than a hand-written diff classification alone, produced the final attribution and scope results above. manual-assignment.json records every stable choice ID with its kind, operation, path, original row, assigned item, and final scope so the 12 assignments can be checked individually.