Gate Analysis

Sequential, in the order the engine actually checks each gate. A gate with a high fail rate relative to how many sweeps reach it is doing real work; one that rarely gets reached — because an earlier gate already rejected almost everything — hasn't had a chance to prove itself yet.

Shadow win rate among sweeps that reached each gate ("before") vs. the subset that also passed it ("after"). Positive lift means the gate filters toward winners; near-zero means it isn't adding much on top of the gates before it.

Is the reversal-bias no-trade filter — the newest and most aggressive gate — specifically earning its keep?

"Blocked" = trade_gate_status = filtered_continuation_bias (real trade never opened, shadow trade still ran). "Allowed" = trade_opened or awaiting_displacement (real trade did or would have opened).

Same sweeps, three gate combinations, using only data already captured — no replay needed. Comparing B and C against A shows which side of the pipeline is actually earning its keep.

Blocked count and blocked win rate per day. A shrinking blocked count over time with a consistently low blocked win rate is the filter doing its job quietly.

Same blocked-vs-allowed comparison as "The key question" above, sliced by symbol, pool type, and session — switch tabs to see whether the filter's effect holds everywhere or is concentrated in a few contexts.

Last 100, most recent first. Spot-check these directly against your own read of the chart if a verdict looks off.