Two handoffs leave the root, and the orchestration handoff says what it got wrong

[syncytium2]

The root is meant to answer "is anything in flight?", and two of its seven handoffs had nothing in flight. This is the cleanup from the review of #733.

Moved to docs/handoffs/, under the dates they were written. Each gets a header saying where its open items went, and its links are rewritten for the new directory. The bodies are otherwise untouched, as docs/handoffs/README.md asks.

• HANDOFF-slow-bench.md → 2026-09-21-slow-bench.md. Its own current-state section says nothing is running or held, and the one PR it waited on (Why a 0.1 s jitter is measurable on 0.1 s frames, as a figure #727) has merged.
• HANDOFF-the-readouts-landed-the-code-behind-them-did-not.md → 2026-09-20-…. Every item it carried has since been ruled, fixed or tracked elsewhere; the picker question was ruled today in gauge and chorus stay in the model picker, and chorus.py's citation resolves #734.

One new todo, for the only open item that had nowhere to go: the every-knob search has two defects.

• It extends a count it doesn't know is a count by halving it. The slow search's own log has sce.min_rois at 0.375.
• It reads a flat plateau as a climb toward the low edge. SPIKE-synch's C_min was extended four times across values that all scored the same.

Pointers updated. Everything that named HANDOFF-slow-bench.md now names the new path. That includes docstrings in bench_slow.py, three tests, two tools and --help text; no behaviour changes.

The orchestration handoff stays at the root, with corrections marked (corrected):

• It answered another session's dataset question for it ("yes, confirmed 2026-09-21"). The dataset rule forbids relaying that answer: Tony confirms the default in each session.
• It read "the meeting happened" as approval of the 0.19 participation change. Nothing in the tree records what the meeting decided.
• It didn't say that the two constant changes and the one-bench rewrite edit the same files. The correction recommends an order, and says 0.19 need not wait on the jitter ruling.
• Two housekeeping items described things that are gone: a machine-local board inside a container that no longer exists, and a branch already deleted.

The goal-2 handoff now says #687 merged, which leaves WSMIP064's crowded-allowance sweep as the only thing standing before it can be deleted.

Also done while here, outside this diff: #686 and #688 merged (both docs-only and green); #689 was updated to current main and set to merge on green. #682 stays open, because its description says the sweep's findings land on that branch.

Checks run here: sapper --all, check_quotes and check_pipelines pass, as do test_index_resolves, test_handoff_is_honest, test_sapper and the three slow-bench test files (435 passed, 2 skipped).

🤖 Generated with Claude Code

https://claude.ai/code/session_01AMvR2nf7HBnze4N87o1TB5

---

Generated by Claude Code