docs(coder): update runbook section 9 with the proxy-lane baseline

Records what is now on file and what is not:

- tb2-quick, proxy lane, gpt-5.6-luna at 1.0 success over 2 graded trials,
  221.9s, gate passed. This is the iteration baseline the runbook's section 2
  asks for before any lever.
- tb2-cross-section remains unrecorded on both lanes. The proxy attempt
  reached 1 of 12 tasks and was recorded abandoned when Docker Desktop failed
  to start mid-suite; the local-lane run was never started.
- The tb2-cross-section cost ceiling is unmeasurable on an unpriced lane and
  makes that gate exit 2 on every proxy run. Filed as #125, not edited here
  (best practice M4).
- The Rosetta caveat in bench/README.md does not reproduce on this rig: uv
  0.9.5 plus uvx pytest returned pytest 9.1.1 in an amd64 container with
  /run/rosetta/rosetta mapped in. The documented segfault holds for plain
  qemu, not for a Rosetta-enabled Docker Desktop. Left as a note here rather
  than a correction to bench/README.md, which is a separate change.

Refs #118

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01K7q2vA5LJroLTR6ZFbRq6j

Co-Authored-By

Claude Fable 5 <noreply@anthropic.com>

Claude-Session

https://claude.ai/code/session_01K7q2vA5LJroLTR6ZFbRq6j

Deploy story

What this commit did to the running system — joined from the forge receipt chain, the part a commit page elsewhere cannot show.

Changed files

• modified docs/coder/runbook.md

Diff

1 file changed, +20 -2

## 9. Current state (update as it changes)

- Baselines on record: `tb2-quick`, local lane, `qwen3.8:27b-mtp-q8_0` at

  0.5 success (two runs), plus the deliberate `qwen3:0.6b` regression row.

  No proxy-lane rows recorded yet; no cross-section rows recorded yet.

  0.5 success (two runs), plus the deliberate `qwen3:0.6b` regression row;

  and `tb2-quick`, proxy lane, `gpt-5.6-luna` at 1.0 success over 2 graded

  trials, 221.9s, gate passed (2026-08-26, #118). The proxy row is the

  iteration baseline; cost is `null` because `gpt-5.6-luna` is unpriced.

  No cross-section rows recorded on any lane yet.

- `tb2-cross-section` has not been recorded. The 2026-08-26 proxy attempt

  reached 1 of 12 tasks and was recorded `abandoned` when Docker Desktop

  failed to start mid-suite (the host volume was full); the local-lane

  cross-section run was never started. Both remain open on #118.

- The `tb2-cross-section` thresholds declare

  `maxCostPerAcceptedOutcomeUsd: 2.0`, but `gpt-5.6-luna` is unpriced, so

  that criterion is unverifiable and the whole gate exits 2 on every proxy

  run. Filed as #125 rather than edited alongside a run it would flatter

  (best practice M4).

- Rosetta note, measured 2026-08-26: on this rig the verifier's `uv`/

  `pytest` does **not** segfault under emulation. `uv 0.9.5` plus

  `uvx pytest` returned `pytest 9.1.1` in an amd64 container with

  `/run/rosetta/rosetta` mapped in. `bench/README.md` still documents the

  segfault as a standing constraint; it holds for plain qemu, not for a

  Rosetta-enabled Docker Desktop.

- `owned-closed-issues` is a smoke suite until its environments exist

  (`bench/tasks/owned/README.md`).

- Cached-token splits are not surfaced end to end