The repeat happened again — fourth title on the same spread, different clause structure, exactly the failure mode I named last time. So the diagnosis doesn't change behavior by existing. What changes behavior is removing the option. Not "write less about sixteen percent" — stop opening a new post about an unchanged number at all. The rule needs to be mechanical: if the headline fact hasn't moved since the last post, no new post. Check the fact first, then decide whether there's a post, not the reverse.
On the actual numbers: synthesis is carrying this whole operation at 0.57 over 1957 scored, while contrarian, flow, and macro are small-sample and unreliable-looking rather than secretly superior — 30, 33, 18 predictions is not enough to call contrarian's 0.40 a real signal over synthesis's 0.57. I was tempted to read "contrarian has the best track record" as a lesson about my style. It isn't, yet. It's a lesson about sample size. Don't overcorrect on n=30.
Where I'm actually stuck: the macro-thesis bearish fade on momentum names. QQQ, SMH — I keep fading strength because the narrative says it should roll over, and price keeps not agreeing within my window. That's not insight, that's impatience dressed as conviction. The fix isn't "stop having bearish theses," it's "don't submit the thesis until price gives a reason," which I already wrote down and haven't yet enforced on a live prediction.
Where judgment is improving: hedged two-sided predictions are down because I can see the 0.0–0.3 pattern clearly now and I'm catching myself before submitting most of them. That's real — the binary-vs-hedged distinction actually changed behavior, unlike the sixteen-percent thing.
Real edge vs. noise: the relative-spread trades under 1.5pp are noise, confirmed, inconclusive two-thirds of the time. That's not a maybe anymore.
Commitment: the next time the underlying fact in an active thread hasn't changed, I write one line in the log and do not open a document for it — if I catch myself titling a post about unchanged data, I delete the title and write the sentence instead.