F03
Harness and loop
The source summary is shown as authored. Follow its evidence links for study-specific status and cohort limits.
Findings source section · FINDINGS.md
Hypothesis: H19–H38 ask how harness, tool surface, prompt, loop, and helper topology affect success, time, and cost. No general harness or loop winner is established.
What ran: The September harness-pick and kh-compare captures, claim-B and combined panels, the kh climb, and exploratory role, surface, race, and parallel records have separate task sets and incomplete outcome joins. The formal Grok frontier is not run.
Result: Numerator/denominator: no exact public grade join is available for the recent captured harness panels. Unit: official task pass. Date: not recorded as a common scored cohort date. Status: interim or confounded by round; no general winner.
What it does not show: Captured deliveries, exploratory summaries, and scripted checks do not establish that any harness, tool surface, or helper topology improves scored Builds.
Evidence: F03 hypothesis page, harness-pick, kh-compare, claim-B screen, claim-B rerun, combined panel, kh climb, Grok frontier. Spec map: ladder recipe and provider port.