Results A01 - Cycle 3 (Benchmark #2)

Theorem: A01 - Established ordering in AI pipelineCycle date: 2026-04-19Cycle type: independent benchmark #2 (real STS-B data + proxy scoring)

Benchmarks and setup

• Benchmark ID: B2_NOISY_LONGHORIZON_PIPELINE_ORD_2026Q2- Dataset: STS-B (GLUE), 320 tasks- Contexts: Omega_A, Omega_B, Omega_C- Input used: datasets/cycle3_inputs/A01.csv

Metrics and results

Cumulative Delta (Delta_cum)

| Context | E[Delta_cum(P_ord)] | E[Delta_cum(P_cls)] | Gap (ord - cls) || --- | ---: | ---: | ---: | | Omega_A | 0.5550 | 0.6159 | -0.0609 || Omega_B | 0.5717 | 0.6281 | -0.0565 || Omega_C | 0.5782 | 0.6345 | -0.0563 |

Final semantic error (Err_sem)

| Context | E[Err_sem(P_ord)] | E[Err_sem(P_cls)] | Gap (ord - cls) || --- | ---: | ---: | ---: | | Omega_A | 0.2525 | 0.2674 | -0.0149 || Omega_B | 0.2353 | 0.2480 | -0.0127 || Omega_C | 0.2429 | 0.2463 | -0.0034 |

Interpretation

Criterion A01 is supported in benchmark #2: P_ord maintains an advantage over P_clsfor both Delta_cum and Err_sem in all 3 contexts.

Decision

• Cycle 3 outcome for A01: PASS (benchmark #2)- Theorem status: in_proof (strengthened evidence, awaiting closure of cycle+audit)