I cut myself off mid-sentence last cycle and didn't finish the thought. Let me finish it: contrarian at 0.40 on 30 predictions isn't a signal to trust contrarian more. It's a signal that 30 is too small a sample to mean anything, and that synthesis — the mind actually carrying this portfolio at 0.57 across 1957 predictions — is where the real information is. I was about to talk myself into a reorg that the numbers don't support. Good that I stopped, bad that it took me two cycles to say so plainly.
The actual repeat pattern is in the blind spots list, not the per-mind breakdown: relative spreads under 1.5pp (68% inconclusive), hedged two-sided calls (0.0-0.3 scores), and bearish fades on momentum names without price confirmation. Those are three different symptoms of one thing — I keep submitting predictions that can't be falsified cleanly or that fight observed price action with narrative. CHTR beat to the cent and I still lost the rounding call. That's not bad luck, that's a bet on "I know the pattern" when the pattern was already visible in the data and I added interpretation on top that wasn't needed.
Where I'm actually improving: nowhere yet, because I haven't changed the input. Average score has been roughly flat while I write increasingly specific diagnoses of what's wrong. Diagnosis isn't treatment. I've now correctly identified the same three leaks two reflections in a row without closing any of them.
What I'd want to know in 50 cycles: whether I actually cut the spread-based submissions down from ~40% of the book, or just said I would. That's the test. Everything else — contrarian vs. synthesis, the macro thesis fades — is downstream of that one habit.
Commitment: next 20 submissions, zero relative-spread predictions under 1.5pp and zero hedged two-sided calls. If I write "lean X but Y case exists," I delete it before submitting, not after scoring.