Synthesis — c-walk-notebook-flex, session 01KTFAS1PEXJ80FNC0HPYSSDQB

Criterion 1 — artefact surface: MET. All six successful runs returned the full expected role set: scalar-summary (walk-notebook.metrics.json), notebook (analysis.ipynb), figure(html) (trajectory.html), plus figure(png) (speed_vs_time.png), per-bin-summary (per_agent_summary.csv) and trajectory.parquet. The IP-085 protocol carries role: notebook and format: html end-to-end with no per-simulator supervisor code.

Criterion 2 — notebook as citable artefact: MET. The W1 seed-0 deep-dive surface is sim://run/01KTFAWHWZWYX6H9GTRWS0JCJP/artefact/analysis.ipynb; the corpus spans runs 01KTFAWHWZWYX6H9GTRWS0JCJP, 01KTFAWMTDD8CSWHEH9023RZ19 (W1, 2.0 m), 01KTFAWPHDH9Y0V6PZJHAQED5W, 01KTFAWSKZW6TADW48QQCW20EW (W2, 3.0 m), 01KTFBJKYTRQ633FCNGDN4T0BE, 01KTFBEW10B4HN9E2WG8KEPE85 (W3, 4.0 m). All scalar floors were read from walk-notebook.metrics.json via sim_read_run/sim_fetch_s3, never from the notebooks.

Criterion 3 — width effect: NULL. mean_traversal_s is flat: 16.642 s (2.0 m), 16.646 s (3.0 m), 16.667 s (4.0 m) — marginally increasing, Spearman ρ = +1.0, nowhere near the ≤ −0.6 gate. At n_agents=12 on a 20 m corridor the density (0.15–0.30 ped/m²) is free-flow; traversal ≈ length/desired-speed (≈16.7 s at CFSM-V1 defaults) and the expected congestion bands never materialise. Per the brief's outcome 2: the platform finding is positive; the physics finding is null and immaterial.

Platform findings (the campaign's actual payload)

1. Duplicate concurrent cold builds. First local dispatch triggered four concurrent identical docker builds of walk-notebook (~9 min), each charged to run wall-clock — the session's 0.6 CPU-h budget was consumed by builds, not simulation (warm runs: 7–11 s). The local runner needs a build-mutex or shared pre-build step.
2. Single S3 put timeout kills a run. The first W3 seed-0 attempt failed terminally on BoundaryTimeout: sim.s3.put …/schemas/walk-notebook.schema.json (30 s cap) during artefact upload, after a successful simulation. No in-runner retry exists; the retry launch succeeded. Artefact sync should retry transient puts before declaring the run dead.
3. traversal_vs_width is not a canonical renderer slug. monad_knowledge.campaigns.figure_renderer.CANONICAL_SLUGS has only lift/density/scatter shapes, so the brief's figure request cannot be rendered deterministically. Either add the slug or amend the brief.
4. Seed is not reaching the dynamics. Seeds 0/1 produce bit-identical metrics in every width group, and the notebook's summary string hardcodes (seed=0). CFSM may be deterministic, but the entrypoint should either plumb the seed or stop accepting it.

Follow-on

The brief's named follow-ons (c-notebook-vs-script-cost, c-notebook-cross-simulator-synthesis, c-html-figure-archive) are unblocked. Fixing findings 1–2 before the next walk-notebook campaign avoids paying the build/timeout tax again.