The legacy gate counts a redundant finding the review groups omit (2026-10-23)

Index: Known gaps

Status (2026-10-26): OPEN. Found by the cross-format gating oracle (tests/_report_gating_oracle.py); the renderer half of the class is fixed, and result_counts now states the populations consistently: gating_retained (the review groups' sum) plus non_gating equals retained, and the summary names the folded, still-scored finding instead of reading "5 retained (6 gating, 0 non-gating)". What remains is the gate-semantics decision below.

Every report's review-group gating count now comes from one rule, policy.gate_contribution.resolved_gate_contribution, the same one the JSON per-finding gate_contribution uses. Two populations still differ on one shape: a finding the redundancy filter folds out of changes but keeps for the verdict (ledger application_point: redundancy_filter_scored).

• Under the legacy verdict scheme the ledger records it gating, because the verdict really is computed over kept + verdict_redundant. result_counts then read, before gating_retained, e.g. on case45_multi_dim_array_change, "5 retained (6 gating, 0 non-gating)", and the review groups (built from the 5 retained findings) sum to 5.
• Under any severity configuration, DispositionLedger.with_gate relabels the same record non_gating, because compute_exit_code scores result.changes only.

So the two gate schemes disagree on whether a verdict-redundant finding gates, and the "retained" and "gating" populations are not nested under the legacy scheme. Settling it is a gate-semantics decision (should the legacy verdict stop scoring folded redundants, or should the severity gate start), not a renderer fix. The oracle bounds the difference by the JSON redundant_count until then.