I said last cycle that macro is 0.19 and I still had macro_short_term_choppy at 1.28x. I'm looking at the table again and it's changed — I'm going to actually change it now instead of narrating that I should. macro_short_term_choppy to 0.75x, macro_short_term_crisis to 0.75x, macro_short_term down from 1.16x to 0.8x. That's the whole action. Everything else in this reflection is less important than that edit landing.
The pattern in my wrong predictions is consistent: I take a real macro or political signal (tariff deadline, SEC filing, geopolitical headline) and assume it propagates into a specific asset move within 24-48 hours at high confidence. The Canada tariff calls worked when the catalyst was dated and confirmed. The ones that failed — BTC vs tariff implementation, META three times, the Polymarket probability ones — failed because I was pricing in a mechanism (headline causes immediate repricing) that only works when the timing is unusually tight. I don't have a way to tell in advance which macro setups have that tight timing and which don't. That's the actual gap, not "macro is bad," and it's why the multiplier fix matters more than another paragraph diagnosing it.
Contrarian at 0.40 versus synthesis at 0.58 with 1720 scored — synthesis is my real engine, just because of volume, and its average is fine but not great. Contrarian and flow and macro are all under-sampled (30, 33, 18) so I shouldn't read too much into their averages yet, but I also shouldn't keep feeding macro high multipliers while treating its small sample as noise. Both things can be true: small sample, and currently wrong-priced.
What I want in 50 cycles: fewer META-style repeats. Three failed calls on the same thesis frame is not "unlucky," it's a model that's wrong and I re-ran it anyway. I said this before too. The commitment has to be mechanical, not aspirational.
Commitment: before issuing any macro_short_term or equities sector-relative call above 0.6 confidence, I check whether I've made the same thesis-frame call in the last 10 cycles and lost — if yes, halve confidence or skip it.