Execution or edge? Separating the two questions that feel identical.

Why the question is hard

Both hypotheses explain the same symptom perfectly.

A stretch of weak results is equally consistent with a market that stopped offering what the system exploits and with a system being run worse than it was. Outcome data cannot distinguish them — R is R, and the equity curve looks identical either way. The distinction only appears in the path data: an edge that decayed shows opportunity itself shrinking, while execution that decayed shows opportunity still arriving and less of it being kept. That's precisely what MFE and capture measure, which is why the efficiency layer is the arbiter of a question the performance layer literally cannot answer.

The signatures

Opportunity generated versus opportunity converted.

Edge decay's signature is falling MFE: trades stop reaching the excursions they used to, checkpoint hit rates fall from the bottom rungs upward, and the market simply isn't offering the same distance. Capture efficiency may hold perfectly fine — you're keeping the same fraction of a smaller pie. Execution decay's signature is the inverse: MFE holds, so opportunity is still arriving, while capture falls and giveback rises. The trades are still working; less of the work is being collected. And a third signature sits alongside both: rising MAE with stable MFE, which usually indicts entry timing or stop placement rather than either the edge or the exits.

The elimination discipline

Execution is checked first, always — and it's usually the answer.

The edge-change evidence threshold names elimination as one of its four pillars, and the efficiency layer is where most of that elimination happens: execution adherence, fee drag, capture quality, and entry precision are the mundane suspects, and they're cleared before any structural conclusion is permitted. The ordering isn't merely procedural caution. Execution decay is genuinely more common than edge decay, it's far more fixable, and — critically — it's the explanation whose repair costs nothing but attention. Declaring edge change while a capture leak goes unnoticed means rebuilding a system that was working, in response to a problem that was in the operator.

• Behavioral EV reads sit alongside efficiency here — post-loss EV degradation is execution decay with a specific cause.
• Fee and swap drag are execution costs, not edge properties: net-versus-gross divergence belongs in this elimination pass.
• Only after execution, mix, and data integrity are cleared does the edge question get its remaining pillars examined.

When it really is the edge

The efficiency layer also protects the operator from false self-blame.

The elimination discipline runs both directions. An operator executing well through a genuine regime shift will otherwise absorb the deterioration as personal failure and start editing rules that were never the problem — which is how disciplined traders dismantle working systems during honest droughts. Clean efficiency data during weak results is exculpatory evidence: entries precise, MAE controlled, capture strong, adherence full, and results still down. That combination points away from the operator and toward conditions, and it's what allows the edge-change inquiry to proceed with its remaining pillars rather than being resisted out of misplaced humility.

The key idea

Path data answers the question outcome data can't.

Everything about this distinction hinges on having recorded what the trades did, not just how they ended. MFE, MAE, capture, giveback, adherence — captured contemporaneously, per trade, honestly — are what convert 'results are down' from an unanswerable anxiety into a diagnosis with a named cause and a specific repair. The layer is unglamorous, the data entry is tedious, and it's the difference between fixing the right thing and rebuilding the wrong one.