The evidence threshold before declaring the edge has changed.

Why the bar is high

Edge-death is declared a hundred times for every time it happens.

Variance produces edge-death experiences on schedule: any honest positive-expectancy system delivers stretches that are statistically indistinguishable — to the person living them — from a dead edge. The benchmark quantifies this: material fractions of fully healthy governed futures spend months below median, and some spend them below P25. Declaring edge-change from inside one of those stretches doesn't just misread the moment; it destroys the sample. A rule set abandoned at trade 300 never reaches the trade count where its expectancy would have been provable at all. So the threshold is deliberately asymmetric: slow to declare, structured to diagnose, and impossible to clear from placement alone.

The four pillars

Independent instruments, one verdict.

The declaration requires convergence across four evidence sources that fail for different reasons — which is what makes their agreement meaningful. Persistence: sustained adverse placement in the benchmark bands across multiple review periods, not one bad month. Statistical failure: the EV evidence itself degrading through the analytics stack's own gates — the expectancy lower bound, the probability-EV-positive read, rolling EV deteriorating rather than oscillating — on a sample the lab's minimum-size gate accepts as meaningful. Elimination: the mundane suspects cleared one by one — execution adherence, fee drag, branch-mix drift from the benchmark's assumed weights, data-entry integrity — because an edge taxed isn't an edge changed. Structural corroboration: the diagnostic layer independently flagging abnormal state rather than reporting a healthy system having a normal bad stretch.

What happens during

The system already responded — that's the point of governance.

The threshold's patience is only affordable because protection never waited for the diagnosis. If the adverse stretch carried drawdown, the gate ladder already compressed deployment at −7%, −13%, −19% — the account was de-risked by arithmetic while the question was still open. This is the division of labor that makes the high evidence bar safe: gates protect capital in real time on drawdown alone; the edge inquiry proceeds on evidence time. The operator suspecting edge-change needs no emergency powers and gets none — no preemptive rule changes, no discretionary de-risking below what the gate already imposes, and absolutely no aggression to 'test' whether the edge is back.

After the declaration

A legal declaration opens the sandbox, not the toolbox.

Clearing the threshold doesn't authorize live surgery. It authorizes an R&D question: candidate adjustments — branch re-weighting, filter changes, management variants — are formulated as scenarios with explicit expected behavior and run through the simulation layer against the live evidence, exactly as any proposed change would be. The declaration also triggers a benchmark obligation: if the edge has genuinely changed, the simulation assumptions built on the old edge are stale, and the Operating Standard's refresh rule applies — benchmarks update after approved model changes, so the ruler is rebuilt to measure the system that actually exists now. Diagnosis, sandbox, promotion, refresh. The same pipeline as always, entered through a heavier door.

• The declaration is a research trigger, never a permission slip for live improvisation.
• A confirmed edge-change stales the benchmark — refresh follows the approved-change rule, not the emotional one.
• If any pillar later reverses, the declaration reverses with it. Evidence conclusions stay evidence-bound in both directions.

The key idea

Make the biggest conclusion the hardest to reach, and it will usually be right.

The threshold exists because 'edge changed' and 'edge experiencing variance' demand opposite responses and feel identical from inside. Four independent pillars, mundane suspects eliminated, structure corroborating, protection already engaged — a declaration that clears all of that is probably true, and a suspicion that can't is probably the benchmark's below-median quartile doing what it always said it would. The system's answer to the scariest question in trading is the same as its answer to everything: evidence, stacked, before action.