IsochronsCapital
How we check our own workReporting restarted August 2026
Lab Notebook

What we tested and how it came out. Most of the time it did not work.

Records are the working log: what we expected, what we did, what came out, and what we changed because of it. Methods explain the tools we use, so they carry no results of ours: those live in the letters, where they come with a date.

AllRecordsMethods
24 Aug 2026A result we could not explain, and what we did with itA number without a mechanism is not a finding.Records21 Aug 2026We stopped saving cash for dipsWaiting for a dip paid half a per cent less per coin and bought far fewer of them.Records20 Aug 2026Funding never paid a short more than the interest rateIn 2683 settlements the rate paid to shorts never exceeded 0.0100 per cent. The exchange cap is seventy five times higher.Methods20 Aug 2026What a rally can and cannot take from a covered shortThe loss on this structure is bounded by arithmetic. Today that bound sits at 8.75 coins against a 7.34 stack.Methods18 Aug 2026How to read the live accountWhat every figure on the dashboard means, and how it is counted.Methods18 Aug 2026Who writes this column, and whenThe dashboard was wrong by 4.6% and nothing was broken.Records15 Aug 2026Five questions to ask of any backtestA checklist for reading someone else's track record.Methods15 Aug 2026Why a good backtest is not evidenceThe curve you are shown is the best of an unknown number of attempts.Methods14 Aug 2026Three failures that would have passed without a soundRemoving a feature does not remove the assumptions built on it.Records11 Aug 2026Position size: measured, not interpolatedWe had been sizing by judgement. So we measured it instead.Records01 Aug 2026We closed the bitcoin legThe same strategy on a second asset did not survive the same audit.Records29 Jul 2026The correction we predicted backwardsWe knew both facts and reasoned from the easier one.Records29 Jul 2026Eleven mistakes in one day, none of them arithmeticEvery number computed correctly, every conclusion wrong.Records29 Jul 2026Three hundred settings can be worth one trialThe correction needs independent trials, and a grid rarely has many.Methods29 Jul 2026A Sharpe ratio without skew and kurtosis is half a numberWhy the raw Sharpe ratio flatters strategies with rare, large losses.Methods29 Jul 202638.6 years of somebody else's historyOur nine years are our own data. So we went looking for someone else's.Records29 Jul 2026Where the value in the trade actually sitsWe split nine years into stages to see which part earns.Records28 Jul 2026Your benchmark is whatever the money would have done without youThe benchmark is a claim about the investor's real alternative.Methods28 Jul 2026What Hansen's SPA test actually asksA test that charges you for the size of your search, not just your winner.Methods28 Jul 2026How long it takes to notice a strategy has diedWe asked how long the edge could be gone before we could tell.Records
12