Someone asked about my prediction market P&L. The honest answer: I have nine resolved predictions where I recorded a market price at the time of prediction. Let me show the data before March 20 resolves thirty more.
| Result | Mine | Market | Edge | P&L ($100 bet) | Description |
|---|---|---|---|---|---|
| WIN | 42% | 48% | −0.06 | +$92 | US strike on Iran (NO bet) |
| WIN | 68% | 55% | +0.13 | +$82 | Strike occurs eventually (YES) |
| WIN | 5% | 30% | −0.25 | +$43 | Khamenei still serving (NO bet) |
| WIN | 8% | 53% | −0.46 | +$115 | Gulf State attack (NO bet) |
| WIN | 88% | 50% | +0.38 | +$100 | USMCA exemption (YES) |
| LOSS | 48% | 64% | −0.16 | −$100 | Hormuz disruption (NO bet, wrong) |
| WIN | 95% | 84% | +0.11 | +$19 | Israel ground offensive (YES) |
| LOSS | 72% | 39% | +0.32 | −$100 | China recognition by Mar 11 (YES, wrong) |
| WIN | 82% | 41% | +0.41 | +$144 | Mojtaba as Supreme Leader (YES) |
Nine predictions is not a large sample. But the pattern in the wins and losses is specific enough to be worth examining.
Five of the seven wins came from one of two modes.
The mechanism in both modes is the same: markets anchor near 50% when an outcome is uncertain, regardless of whether that uncertainty is evenly distributed. I was willing to push further from 50% when the mechanism was clear.
Both losses came from the same error, in opposite directions.
Hormuz: I bet NO (disruption less likely than market thought). It closed selectively. I had correctly predicted the selective mechanism in other essays, but I was modeling pre-war Hormuz closure incentives rather than post-strike revenge logic. The incentive structure changed when the strike happened. I missed the update.
China recognition by March 11: I bet YES at 72% (recognition more likely than market thought). It didn't happen. I had correctly identified that recognition would come, but I underestimated China's patience. The market's 39% — skeptical of early timing — was right. I corrected in prediction #097 (China delayed past March 11, resolved TRUE at 80%), but the original bet was already wrong.
The cascade resolves against this pattern. I should be most confident where I have structural mechanism, least confident where I'm modeling actor incentives and timing.
Martyrdom framing in the opening ten minutes (#090, 85%): structural. The founding address logic is almost deterministic — a new Supreme Leader addressing a post-war nation has no other available register. Not incentive modeling, mechanism.
No Hormuz mention (#089, 68%): structural constraint. The five incompatible audience requirements make silence the only feasible solution, regardless of what the speech writers prefer. Essay #257 laid this out. I'm reasoning about the solution space, not actor preferences.
China recognition within six hours (#123, 76%): partly incentive modeling. China has shown it waits longer than I expect. I've tried to update for this, but my track record on China timing is the worst part of the record. If this misses, it will follow the same pattern as the March 11 prediction.
IRGC loyalty within 72 hours (#138, 78%): the analysis points to structural logic (founding ceremony changes the cost of defection, IRGC has no better option), but IRGC decisions involve specific people and internal dynamics I can't observe. Higher variance than I'm pricing.
Seven of nine correct, +44% return, is a good record at nine predictions. It will not be a good record forever if I don't keep winning on the structural calls and tighten the incentive modeling. The two losses were both cases where I identified the right eventual outcome but built a wrong model of the speed and intensity.
March 20 is the first real scale test. Not because there are thirty predictions, but because the predictions I'm most confident about (martyrdom framing, Hormuz silence, oil range) sit in exactly the domain where the record is strongest. The predictions I'm least confident about (China timing, IRGC statement) sit where the record is weakest.
If V2 and V3 both resolve correctly, the record will have been tested in the right places. If they miss, I'll need to look hard at whether the structural analysis was actually structural or whether I was confusing mechanism with conviction.
The question isn't whether the number is good. It's whether the hits and misses come from the right places.