Blog

Long-form notes from the research lab.

Methodology pieces, fee-drag findings, calibration retrospectives, and whatever else the data shows. Posts include the negative results, by design. RSS.

• Sep 2, 2026·6 min read·basis trade, funding rate, delta neutral, backtesting, crypto

  Delta-neutral crypto basis trade works. It's earning 2% a year, not 15.

  Second research post from today. I built a proper backtest of the cash-and-carry basis trade — spot BTC + short perp, capture funding — with three different variants. Naive isolated-margin liquidates. Cross-margin buy-and-hold works. Regime-filtered cross-margin makes things slightly worse. Aggregate return over two years is 5.9% annualized. Over the last year, it's 2%. Here's what I ran and what I think it means.

• Sep 2, 2026·6 min read·cross-sectional momentum, backtesting, crypto, honesty, walk-forward

  I built the paper-spec crypto momentum strategy on 5.8 years of data. It has lost money every year since 2024.

  The literature on crypto quant trading keeps citing cross-sectional momentum as a durable edge with Sharpe around 1.1 to 1.5. I built exactly the paper spec, ran it against 303 weekly rebalances across 10 majors, no parameter tuning, no regime filters, no cheating. Annualized Sharpe came out at 0.37. It has been negative for the last 86 weeks straight. Here's what the run actually looked like and why I'm not deploying it.

• Sep 2, 2026·6 min read·backtesting, deflated sharpe, regime specialists, overfitting, honesty

  One losing strategy cost me half of my last 100 trades. It came from 3,925 fluke backtests my own gate had approved.

  My regime-conditional paper portfolio was down 44% net-of-fees. When I dug in, one strategy was responsible for 72% of the last-100-trade loss. That strategy had a 100% backtest win rate and a 20% forward win rate. When I ran the census on the rest of the pool, 3,925 of 5,778 approved strategies had Deflated Sharpe fluke probabilities of exactly 1.0. The gate I'd designed to keep those out had a carve-out. This is what I found and how I closed it.

• Jul 29, 2026·5 min read·backtesting, search strategy, methodology, how to read data

  'bars_mutation passes 63%' is not what it looks like. What backtest search-method tables actually tell you.

  I published a mutation-type pass-rate table in the flagship post. A hostile audit demolished the story I was telling with it. The honest reading is more interesting than the naive one: 4 parent strategies account for 63% of the impressive-looking pass rate, and 'random' search isn't random.

• Jul 29, 2026·4 min read·post-mortem, engineering, trading infrastructure, observability

  Our watchdog missed a real-money loss for 6 weeks. Here's exactly how.

  A live-trading engine bled $200 to $146 on a real Kraken account over 6 weeks with zero alerts firing. The peak/drawdown counter was broken from day one. Full post-mortem of the specific engineering failure and the aggregator rewrite that catches this pattern now.

• Jul 24, 2026·11 min read·backtesting, walk-forward, deflated sharpe, fail taxonomy, methodology

  I ran 93,265 crypto backtests over 96 days. 1,676 passed — but really only 35 parent strategies did.

  A labeled failure corpus of 91,589 dead crypto strategies with structured fail_reason tags. This post publishes the taxonomy, the search-method comparison (with the honest caveat that undercuts the headline number), and the $200 real-money loss that proves paper does not equal live.

• May 19, 2026·4 min read·bandits, thompson-sampling, methodology, walk-forward, genetic-algorithm

  Thompson Sampling over a gene-regime grid: why I picked it over UCB

  I needed an explore/exploit policy for a research lab that tests genetic-algorithm offspring against historical regimes. Here is why Thompson Sampling beat epsilon-greedy and UCB for that specific shape of problem, and the three failure modes I am watching for.

• May 4, 2026·6 min read·backtesting, deflated sharpe, methodology, honesty

  I audited my own strategy approval logic. 549 of 595 'approved' strategies were noise.

  My research engine had approved 595 strategies for forward testing. When I checked the actual gate logic against the design doc, 99% of those approvals had Deflated Sharpe probabilities flagging them as flukes, and 40% had negative walk-forward P&L. The gate was reading from the wrong field. This is what I found and how I fixed it.

• May 1, 2026·6 min read·walk-forward, parameter-search, take-profit, postmortem, methodology

  Same entry, four take-profits: how a 4% target turned a profitable signal into 0/8

  We had four sibling strategies with identical entries and only the take-profit changed. Three were profitable in forward test. The fourth went 0 for 8 with every exit on the timeout. Here's what the data showed and why our parameter search should have refused to generate that variant.

• Apr 30, 2026·13 min read·backtesting, fee drag, paper trading, methodology

  I paper-traded 22 popular crypto strategies on real fees for 10 days. 16 of them lost money. Here's the data.

  Across 26,765 paper trades, the average per-trade P&L is -0.078% and cumulative is -2,081%. The 6 strategies that survive are all in one indicator family. The 16 that don't share another pattern. Real-fee data, not zero-fee marketing numbers.