Strong leans need a strong recent record

There is a subtle weakness in any conviction score, and it has nothing to do with the math. A score can say "+8, strongly bullish", but the number alone never tells you whether the method behind it has been right lately. A confident read from a system that has been wrong more often than right is not confidence. It is noise wearing a number.

A conviction is only as good as its record

Orion's cumulative verdict is a conviction score. It fires only when many independent readings align in the same direction, and it leans contrarian by design: accumulation into weakness, distribution into strength. The score is the headline. But a headline conviction is worth little if the method that produced it has been missing lately.

So we added a cross-check. Before Orion issues a high-conviction verdict, it now asks a second question: how has this method's own recent accuracy been? If the track record sits below a floor, the high-conviction read is held back. The strong lean does not fire.

The rule, in plain terms

A high-conviction read must earn its confidence. The conviction score measures how many independent readings align. The accuracy score measures how often those readings have actually landed. The cross-check ties the two together: a strong lean cannot come from a method that has been wrong more often than right in its recent past.

When the recent record is thin or below the floor, Orion steps back. The verdict still exists as data, but it is not promoted to a strong conviction. The engine declines to lean hard on a method that is not currently earning it.

Why this matters for trust

The easy thing is to let conviction stand alone, a big number with no memory. The honest thing is to make the number answer for its own record. A track record that shows only its wins is not a record. A conviction score that ignores its own recent misses is not a score. It is an advertisement.

The cross-check closes that gap. A "+8 conviction" can never come from a system that has been wrong lately, because the system would have held itself back before it ever fired.

A cross-check, not a mute button

This is not the engine going quiet. It is the engine being honest about when it has earned the right to lean. When accuracy is strong, the high-conviction verdict fires exactly as before. When accuracy is weak, the verdict steps down and the engine says so. The cross-check is a gate that opens and closes with the method's own performance.

The result is a conviction score you can actually read: strong when the method is earning it, restrained when it is not. That is the whole point of keeping a record in the open.