11 Aug 2026Position size: measured, not interpolated62% of the stack short when we are fully in. The cost is stated: drawdown in coin rises from 28.6% to 34.8%. We capped ourselves at 79%, where the measurement stops being precise enough to act on.Confirmed
01 Aug 2026We closed the bitcoin legThe same audit ETH passed. On bitcoin, across 405 versions, the result could not be told apart from luck: 83% of the time chance alone would have done as well, against 2.8% for ETH. The result that looked best turned out to be a side effect of measuring in dollars. One account, one asset, from here.Rejected
29 Jul 2026Where the value in the trade actually sitsThe bare shape of our trade is worth 1.17×. The circuit breaker multiplies it by 5.29 and the regime filter by 1.48. Nearly all the value lives in one component, which means nearly all the risk does too.Confirmed
29 Jul 2026Eleven mistakes in one day, none of them arithmeticEvery number computed correctly, every conclusion wrong. What we changed in the process afterwards, and the checks we turned into code so they no longer depend on us noticing.Process
29 Jul 202638.6 years of somebody else's historyCboe BXMD, covered calls on US equities back to 1988. Beta 0.822 against our 0.809, downside capture 0.771 against 0.765. Agreement on data we could not possibly have touched.Confirmed
28 Jul 2026We paid for our own searchTry enough variations and one will look brilliant by luck alone. So we tested all 315 our parameters can produce, using a method that charges us for every one we tried. It asks a blunt question: if none of them had any real edge, how often would the best still look this good? Answer, 2.8% of the time. Our live configuration ranks 59th of 315, not first.Confirmed
28 Jul 2026We tried to break the strategy with a price stopThresholds swept from 5% to 60%. The best one is the level at which the stop never fires, which is to say no stop at all.Refuted
28 Jul 2026How long it takes to notice a strategy has diedSix to twenty years. Uncomfortable, and everything we do about position size follows from it.Confirmed
20 Jul 2026We stopped reading candles at entryWe used to hunt exhaustion patterns and take pride in it. When we measured it, the entry had stopped mattering. Entries now run on a fixed schedule.Refuted
19 Jul 2026We removed averaging downOver nine years at identical exposure, single entry returns 11.45× against 1.96× for the pyramid. A pyramid adds size precisely when the position is wrong. Removed as code, not disabled by a flag.Refuted
02 Jul 2026The position that sat still for 284 daysOur worst historical case of getting stuck. Whether that was risk or forgone return, and why the answer changes what you do about it.Confirmed