main was red, and the root handoff was wrong about four things out of five

[syncytium2]

main was failing both date gates and nobody had touched anything — the date rolled over. #405 and #404 edited the viewer without moving its version stamp, so the page claimed 2026-08-29 while git recorded both commits on 2026-08-30.

Bumped in docs/site/viewer.template.html, not in the built page: since #404 the viewer is assembled, and editing raster_viewer.html directly is exactly what test_assembled_viewer catches. It caught me.

The handoff had rotted in every figure

It is the first thing the briefing reads aloud at every session start.

it said	actually
in flight: #304, #292, #270 (red), #53, #50	#304 and #270 merged two days earlier
main green at ab1dbfd	ced0da4 — 61 commits back
suite 1,391	1,571
board down to 4 ACTIVE	git board 11, machine-local 30 (only 7 with a live worktree)

None of that fired a gate. test_handoff_is_honest.py only goes red when the last named PR closes, so the file could be wrong about four things out of five and still read as authoritative. Same shape as the scoreboard gate that could not fail and the suppression helper that went blind — and it earns the general form: a file whose whole job is to say what is true cannot be checked only on its own retirement condition.

Decision 2 is answered; decision 3 is added

#298 is settled. Asking for the orphaned branches to be merged answered it — ADR-0003 has been on main since #406, all nine citations resolve, and the todo is closed rather than left waiting-on-tony. What that todo got right is worth keeping: it refused to fill 0003 with anything else while the question was open, because renumbering would have broken nine citations to close one gap. Holding a number is a real option, and this is the precedent.

bench-background-is-not-flat is now the one thing nobody can advance, so it is written into the handoff as a decision. It merges cleanly and fails four tests: three test_background_curve.py tests its author left red on purpose ("re-baselining them would delete a finding"), and test_the_server_reproduces_the_published_bakeoff at fold 3, 29 != 28, because docs/learned/bakeoff.json was computed on the flat field.

Every tube-variant number in the 2026-08-29 handoff was measured in that environment. It does not exist on main, which still describes its own bench as "Not wired into the bench".

Verification

Suite 1,571 passed, 49 skipped, 0 failed. Sapper clear.

🤖 Generated with Claude Code