We publish the kill decisions.
An independent research operation. Every hypothesis that dies gets a dated page, the evidence, and the methodology that survives it. No live parameters, no signals, no advice.
Kill ledger
- b27b ground-truth correction — re-audit of the canonical 'paper +$83/hr, live −$119' statisticThe canonical −$119 blowup was itself a measurement artifact: settlement scored against a 36-second-late Binance anchor, and a cash delta misread as PnL. Corrected bar-1 PnL: +$86.22 (−$118.75 on an unresolvable tie).
- eebd-replica — two-sided quoting-and-merge strategy replicating a pseudonymous wallet on BTC 5-minute bar marketsSim v3 said $1.69M ($56K with haircut). Live said −$253.08 across 22 bars — wallet $514.14 → $261.06. Six failure modes, five new: ghost fills, reference-price bias, concurrency races, venue-constraint spam.
- b27b — maker-buy ladder replicating a pseudonymous wallet on BTC Up/Down 5-minute marketsPaper sim reported +$83.20 over 2.5h. Live: funded $633.37, killed after 8 minutes, wallet at $514.14. Root cause: the paper fill model was a strict subset of live execution and filtered out the most toxic fills.
Free: The Fill-Model Audit — 5 checks before you trust any backtest.The procedure that has killed more of our strategies than any market event. One afternoon, no signup.
Notebook
- Copy-trading the most profitable Polymarket sports wallets doesn't work. We tested it...Copy-trading the most profitable Polymarket sports wallets doesn't work. We tested it four independent ways, out-of-sample, look-ahead-free — and lost money every time.
- Conditional re-quoting on prediction-market crypto pairs is worse than static quoting...Conditional re-quoting on prediction-market crypto pairs is worse than static quoting at every tested price.
- We ran the late-favourite sweep on 8 Polymarket crypto books for 7 days with live...We ran the late-favourite sweep on 8 Polymarket crypto books for 7 days with live money. It lost on every book that traded meaningfully. The strategy is refuted.
- A viral X post claimed BTC Up/Down 5m contracts on Polymarket were mispriced by 3–5...A viral X post claimed BTC Up/Down 5m contracts on Polymarket were mispriced by 3–5 points; we tested it against 41,317 resolved markets and refuted every edge claim.
- The Flow Gate That Fired 34% of the Time — and Why We Turned It OffThe flow gate signal is real. The economics are not — it needs 106x our current maker book to clear $5/day.
Get the kill decisions by email
New postmortems and ledger entries, as they're published. No spam, no tracking pixels of ours — it's Substack's form.
More: the kill ledger · the audit offering · about