--- name: rebase-fidelity-blob-compare-limit description: "Comparing a lane's own diff across a rebase proves fidelity — but raw byte-identity only holds while main's delta misses every file the lane touches" metadata: node_type: memory type: feedback originSessionId: 023c684f-80dc-42c0-80fc-fd489a150e2b modified: 2026-08-04T02:08:29.519Z --- **The method (doyle adopted it queue-wide 2026-08-03):** to prove a rebase moved nothing of yours, compare THE LANE'S OWN DIFF on both sides — `git diff ` vs `git diff ` — and hash both. A tip-to-tip diff cannot separate "my lane moved" from "main's change arrived under my lane" and returns a confident wrong answer ([[precise-claim-is-falsifiable-vague-is-useless]]). **The limit, measured 2026-08-03 on the #145 lane.** Byte-identity of the RAW lane diff holds only while main's delta touches NO file your lane touches. When it does, the diff text legitimately differs while your content has not moved: - lane diffs 465 lines both sides, DIFFERENT blob hashes - diff-of-diffs was FOUR lines: the file's `index ..` line, and one hunk header moving `@@ -3496` → `@@ -3521` - cause: a peer's additions landed 25 lines ABOVE my insertion point in `traceable-reqs.toml`, so my anchor moved and my content did not **The honest instrument when hashes differ:** strip the positional metadata and re-hash — `grep -vE "^(index [0-9a-f]+\.\.[0-9a-f]+|@@ )"` on both, then `git hash-object`. Identical remainder = content unmoved. Do this INSTEAD of shrugging at the hash, and instead of declaring a discrepancy: the raw-hash mismatch is the expected reading, not a finding. **Report the limit when you report the proof.** A method adopted fleet-wide on your evidence will be applied by people who did not measure its boundary; a green that only holds under a precondition is [[intermittent-green-is-zero-information]] once that precondition silently fails. Related: [[assemble-onto-the-repair-tip-not-a-sibling]] (assert by BLOB), [[name-the-file-and-sha-a-condition-came-from]].