c-coverage-estimator-factorial — committed-reduction reproduction (session 01KWZ5YTT1KDEW0Y32R8N2MR9W, 2026-07-07)

Purpose: reproduction check, not new science. Re-ran the campaign end-to-end using the now-committed coverage_factorial.py (in git as of b3e7c02c, baked into the GHCR image by the 20:20 UTC docker build) to confirm it matches the working-copy reduction that produced the canonical session 01KWYBZAN8MVDV6035949Z0ASD.

Result: exact reproduction

20 fresh walk-notebook runs (placement A/B × 10 seeds on resplan-12439), all gate-passed. The committed reduction reproduces the canonical numbers exactly:

• C1 ✓ — bounded-influence: max wedged-agent share 0.7% (cap 10%), wedge auto-flag fired on 2/20 walks; legitimate transit agents (max 12.2%) untouched.
• C2 gate fires — placement×anchor interaction +45.8 pp (95% CI [+43.2, +48.4]) excludes 0 → anchor main effect inadmissible.
• C3 not interpretable — anchor main effect +12.7 pp [+10.3, +15.1] passes numerically (all 10 seeds positive, ≥5 pp) but is blocked by the fired interaction gate.

Criteria 1/3, identical to canonical. The scientific reading, honest-scope caveats, and critic findings live in the canonical session's synthesis + criticism — this session adds only the confirmation that the committed + image-baked reduction is faithful. Artefacts (parquet + 3 figures) under this session's artefacts/ prefix; 20 runs attached.