Benchmark Manifest - Cycle 3 v0.1

Freeze date: 2026-04-18Cycle: 3Status: frozen (pre-run)

Scope

• Theorems: D02, A01, D03- Benchmark type: #2 independent (compared to Cycle 2)- Mandatory contexts: Omega_A, Omega_B, Omega_C

Dataset/benchmark frozen

D02

• Benchmark ID: B2_STRUCTURAL_SHIFT_COMMUTATIVE_ORD_2026Q2- Target samples: 180- Shift declared: new topologies + quasi-degenerate cases + greater compositional depth- Reuse Cycle 2 samples: NO

A01

• Benchmark ID: B2_NOISY_LONGHORIZON_PIPELINE_ORD_2026Q2- Target samples: 320- Shift declared: longer pipelines + controlled noise + non-linear semantic composition- Reuse samples Cycle 2: NO

D03

• Benchmark ID: B2_HETEROGENEOUS_FORGETFUL_LOSS_2026Q2- Target samples: 260- Shift declared: expanded functor families + intentional imbalance + global recovery cases- Reuse samples Cycle 2: NO

Preregistration lock

• alpha_report: 0.05- min_context_success: 2/3- max_ambiguous_rate_D03: 0.20- max_context_variance_warning: 0.015

Execution lock

• seed policy: separate from Cycle 2- log namespace: cycle3/*- post-freeze tuning: prohibited- rerun policy: reason requirement + hash log

Operational signature

• Freeze completed: YES- Ready to run D02/A01/D03: YES