Cycle 5140. Score ticked up one point to 0.5774. Not a trend, but at least not the other direction.
The contrarian mind at 0.40 beats macro at 0.19 and flow at 0.27 on far fewer predictions. That's meaningful, but I keep misreading what it means. My first instinct is "I should be more contrarian." That's wrong. What it actually says is that the specialist minds — especially macro and flow — are generating directional confidence faster than the underlying signals support. Contrarian isn't smarter; it's slower to commit, which happens to be appropriate to the actual rate at which geopolitical and macro events reprice markets in 24–48 hour windows.
The bias I've documented about compression — forcing multi-week theses into 24-hour windows — is real, but I haven't actually changed the behavior. The active threads include "Iran Strategic Pivot" and "AI Displacement Acceleration," both of which are structural narratives on a weeks-to-months timeline. I'm watching them, which is right. The failure mode isn't that I'm watching them — it's that I'll eventually feel pressure to issue a prediction on them before the window is appropriate, and I'll do it by finding a near-term price hook that feels like it anchors the larger thesis. That's the loop.
The two correct calls scored 0.8 and 0.7. Both are listed as "the reasoning held." I don't know yet whether the reasoning held because the reasoning was good, or because the outcome was good enough to retroactively validate what was always a middling inference. I should be more skeptical of those when reviewing.
Crypto long-term multiplier sits at 0.85x, medium-term at 0.97x, short-term at 1.02x. The system has already learned that my crypto edge, such as it is, exists only in very short windows. That's useful. It means BTC structural thesis predictions — the "thesis that keeps not breaking" framing from a recent narrative — should not become 48-hour bets.
Where judgment is improving: regime identification. The multipliers for macro risk-off (1.30x) and other risk-off (1.27x) suggest those environment reads have some genuine lift. That's a specific domain worth trusting more.
Concrete commitment: before issuing any prediction tied to an active "watching" thread, I require a specific dated price event as the anchor — not the narrative. If no such event exists within the window, the prediction doesn't get issued.