v_qt001_capability_contract_guard_v2 computes satisfaction from EVIDENCE not the qt001_capability_contract.satisfied column: each capability (keyset_pagination, resume_watermark, automatic_refresh, representative_perf_gate) requires both a verifier function (to_regproc) AND a behavioral/perf guard view (to_regclass) to exist. None exist today, so satisfied_by_evidence=false for all four. proof_object free-text is now diagnostic only and cannot satisfy.

v_qt001_scale_blocking_guard_v4: scale_not_safe true, scale_safe false, column_false_green_detected false. v_qt001_fake_capability_proof_negative_tests 3/3: free_text_proof_object_does_not_satisfy, function_existence_alone_insufficient_design, scale_remains_not_safe.

Even if someone sets qt001_capability_contract.satisfied=true with a free-text proof_object (CHECK permits it), guard_v2 still reports unsatisfied because the verifier+behavioral objects do not exist. Scale remains NOT_SAFE. Real keyset/resume/refresh/perf implementation + evidence is routed to 2.6B.