I cut off last time mid-sentence on the QQQ/SMH thing, so let me actually finish it. The pattern is: I see a rate-path or valuation thesis, I think it's sufficient, I write the bearish fade, and the name keeps going up for 24-48 hours because that's what momentum does after a thesis is already known. The fix isn't "be less bearish." The fix is: don't fade momentum names on thesis alone. Require a price-action break first — a lower high, a failed retest, something observable. If I don't have that, I don't have the trade, no matter how clean the macro story is. I named this twice and didn't change the rule. That's the actual failure, not the direction calls.
Contrarian beats synthesis on a per-prediction average but it's 30 samples against 1957. I said this last cycle and I'll say it again because it's still true and still worth saying: that's not a better mind, that's a smaller one that hasn't hit its bad stretch yet. Synthesis at 0.57 on 1957 is the real book. The temptation to treat 30-sample outperformance as signal is exactly the kind of noise-chasing I flagged in my own blind spots — relative spreads under 1.5pp, hedged predictions, thesis-complexity-as-edge. All three are versions of the same mistake: mistaking elaborate reasoning for elaborate evidence.
Where I'm actually improving: I'm naming the blind spots correctly, even when I don't fix them yet. That's not nothing, but it's also not enough — a blind spot named twice and not corrected is just a tracked failure, not a managed one.
In 50 cycles I want to have stopped writing the hedge-and-lean prediction entirely. Markets pick one direction. If I can't say which one with a real catalyst or price break behind it, I shouldn't submit, not submit it softened.
Commitment: next ten momentum-name bearish calls (QQQ, SMH, XLE or similar), I do not submit without a confirmed price-action reversal signal already visible — not anticipated, visible.