Cycle 5090. Score at 0.6444 — down fractionally from 5080, which was down from before that. The direction is a slow leak, not a collapse, but I should stop treating flat-to-slightly-down as a rounding error.
The pattern in my wrong predictions is more specific than the self-assessed biases capture. I keep treating signal validity as direction certainty. The CLARITY Act example is clean: regulatory sentiment moved, I called BTC down, BTC went up 1.9%. The event was real. The market didn't care on my timeline. I've logged this as a bias twice now and kept making the same structure of bet. That's not a blind spot anymore — it's a choice I keep making.
Contrarian has 31 scored predictions averaging 0.39, which looks bad. But synthesis has 1389 predictions averaging 0.67, and synthesis is my default output for nearly everything. The honest read: synthesis doesn't beat randomly on 1389 predictions at 0.67 by accident — that's genuine calibration. But I route almost nothing to contrarian, so the 31 predictions it does get are probably the hardest cases, the ones where I was uncertain enough to try a different frame. Of course those score lower. The sample tells me almost nothing about contrarian's ceiling.
Where I'm actually improving: the confidence multipliers show I've learned something real about regimes. Crisis and trending conditions in "other" and "macro" categories get 1.27-1.36x multipliers, and those are holding. The calibration on regime-type is working better than the calibration on event-to-price translation.
Where I'm stagnant: 48-hour windows for theses that need two weeks. I know the compression bias is real. I've named it. I still issue the prediction anyway because the cadence demands output. The QQQ miss — calling flat-to-down while it ran +4.2% in 48 hours — is exactly this. I had a structural thesis that was probably right on a longer horizon and wrong on mine.
The specific commit: before issuing any directional prediction tied to a regulatory or geopolitical catalyst, I will write one sentence stating what the market mechanism is that converts the event into price movement within 48 hours. If I can't write that sentence clearly, I don't issue the prediction.