F01
Shape, Intent and plan
The source summary is shown as authored. Follow its evidence links for study-specific status and cohort limits.
Findings source section · FINDINGS.md
Hypothesis: H01–H15 ask whether shaping, plans, probes, and stage choices improve completion or cost under defined conditions. The public record does not establish a shaping effect or a Build guarantee.
What ran: R69 and R71 have interim stage and Intent records; R72 is a separate descriptive ladder comparison; R73 is not run; R74 is withdrawn. The L3 repair lane is DESCRIPTIVE: two rescues among six selected failed implementations, with per-test preservation unverified. R69's L0 reconciliation reconstructs 400 assigned and 396 included ITT rows; 150 of 327 internal grades exact-join the refreshed 2026-10-07 public outcome export by cell_id, so CL-WA-r69-source-reported-denominator-variants remains INTERIM. Offline Intent, test-quality, and saved-candidate audits are different units from live Builds.
Result: The L3 DESCRIPTIVE round records 2/6 repairs rescued, below its three-rescue threshold; preservation is unmeasured, environment evidence is absent, and rule chronology is unproven. CL-L3-REPAIR-RESCUE-RATE is STATUS_ONLY because the CSV summary does not include per-test pass sets or independent grade receipts. No reconciled H01–H15 shaping cohort is recorded. Status remains round-specific: R69/R71/R72 are interim, R73 is not run, and R74 is withdrawn.
What it does not show: The L3 result does not establish that repair increases completion over restarting or preserves each previously passing test. These records also do not establish that added Intent detail always helps, that a witness guarantees a successful Build, or that shaping is worth its cost.
Evidence: F01 hypothesis page, L3 repair, L3 status-only claim, R69, L0 reconciliation, R71, R72, R73, R74, Intent coverage audit, over-strict replay, candidate counterfactual. Spec map: Intent and Shaping and approval.