Pass2 — PASS

Separate sequential actual-codec rerun at pinned local HEAD [8e4dcde14c](https://git.cafe/pingdotgg/t3code/commit/8e4dcde14c02061f11d4de36237857e4f31d86e7), merge-base [d14ace8233](https://git.cafe/pingdotgg/t3code/commit/d14ace8233ad3cf7856b6b834824f888240d67a1). Original refresh and historical pass2-final remain separate.

Command: `node /tmp/ev-24265/pass2-head8e4-refresh/benchmark.mjs /home/ubuntu/wt/pr-24265 /tmp/ev-24265/pass2-head8e4-refresh/pass2`

All equality, V4-to-HEAD, V4 resave fallback and older-reader rejection assertions passed again. Same 10k fixture byte counts. Independent Python recomputation matched min/median/mean/max for all samples in both runs. All five primary public downloads returned HTTP 200 and matched local bytes. No test rerun in pass2: 51 tests had already passed before both sequential benchmarks.

| Shape | V4 bytes | V5 bytes | V4 encode ms | V5 encode ms | V4 decode ms | V5 decode ms | HEAD reads V4 ms |
| --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
| shared | 44142269 | 22575684 | 408.660 | 173.639 | 318.826 | 185.425 | 346.516 |
| equal-but-distinct | 44142269 | 44142354 | 388.602 | 372.065 | 337.943 | 345.718 | 323.618 |

Host measurements only, no general fallback speedup claim. Android/iOS, native SQLite, web/desktop native flows and transport end-to-end flows remain unverified. No emulator/adb. Remote branch remains expected unpublished predecessor; this verifies local HEAD only. Checkout remained clean and HEAD unchanged.