# Root merge 3: contract evidence, classification gains, and unresolved engine provenance The strongest supported direction in this group is to retain the cost-sensitive arb-call threshold, continue testing bounded Gaussian mixtures, and use the fresh engine audits as diagnostics rather than certification of the earlier engine blocks. This group does **not** establish first place or the required +2.7 over IB on 4,000 engine seeds. All 25 inputs were present. No strategy was submitted to the board by this merge. The assignment's baseline is IB **522.00**, 20th; 10th **524.66** (+2.7 over IB, rounded); first **528.40** (+6.4). The pinned [solved-game README][README] reports replica gains over HA of +8.70 with perfect information, +3.77 Gaussian/QM, +4.26 lattice/QM, and +6.98 with the true arb/retail flag. IB's earlier advantage over HA is about +0.59. Replica levels must not be converted into board scores. The README's exact top-ten gap is +2.66; the task's +2.7 is its rounded target. ## Evidence and scope Input citations use their supplied SHA-256 content IDs. Experiments are **worker reports**; only labeled pooling/counts are merge calculations. The appendix preserves settings, paired statistics and diagnostics; code and raw rows remain in the input artifacts. Notes are synthesized here. No policy or engine experiment was rerun; external claims are not independently certified. The requester warns that external citations failed rereading or did not substantiate quotations in: [br-1][b22-br-1] (1 source), [eng3-12003072][b24-eng3-12003072] (2), [eng3-12003584][b24-eng3-12003584] (1), [eng3-12004096][b25-eng3-12004096] (1), [eng3-12004352][b25-eng3-12004352] (1), [eng3-12004608][b25-eng3-12004608] (2), [eng3-12004864][b25-eng3-12004864] (3), [mix-2][b26-mix-2] (2), [fresh4-2][b31-fresh4-2] (1), [audit-1][b32-audit-1] (1), and [audit-2][b32-audit-2] (2). Failed URL/quotation pairs were not identified, so the qualification cannot be narrowed further. Source-dependent explanations/provenance remain qualified; local statistics remain worker reports, not independently certified. This also applies to the audits. ## Track A: implementable ports, but different artifacts must stay distinct **PB-2** blends first-trade log-fee distances with exponent g=1.5 only for 0.001