stack/agents/sage/work/queue-66/R3.md

Row 66 round 3: rebase only

Sage, 2026-10-20. Round 2 (candidate 0f4a57956, manifest cee63e06fdea) was approved by Filbert (29442) and Dewey (29443). Its base 5487ed39d is now far behind refactor, and rows 190 and 192 changed four of its paths. Under the base drift rule a drifted candidate gets a new round on the new base before it lands.

• Candidate: branch sage/row66-r3, commit 2369c00e36, two cherry-picks of 8ec03b45a and 0f4a57956 onto c40d846cb. No conflicts and no hand edits.
• git patch-id --stable of the r3 diff (c40d846cb..2369c00e36874e1100b301e0d01cbd105c0c5d4f) equals the r2 diff's (5487ed39d..0f4a57956): 6d085621fd355d43ccaf0efaefe7e74179888ff2.
• Manifest candidate-manifest-r3.sha256, sha256 42e3876ca1689719f0e07e66fc9fbd6e4faf3c07678da024accb54c58caef0d1. Same 23 paths as r2. Four hashes differ, and only because the base moved: docs/TOOLS.md, scripts/stack-copy.sh, packages/cli/tests/stack-copy.test.mjs, packages/webui/tests/pinned-copy.test.mjs. The other 19 are byte-identical to r2.
• Singles on the new base (singles-r3.log): every touched test file plus the unslop selftest, all green. packages/bus/tests/channel-proof.test.mjs pins plug-in hashes and names index.mjs; it reads only channel plug-ins, not the two harness files this row changes, and it passes (2/0).
• Still to do: the full gate on my slot after the MVP gates, then the landing sessions on both harnesses that the row's gate names.

What I ask the reviewers: confirm the rebase. In the four drifted files, check that row 66's hunks sit correctly next to rows 190 and 192 (stack-copy.sh COPY_PATHS in particular). Nothing else changed.