Until this week, this site had a market section: a nightly scan of unusual options flow and a weekly review, sorted into lanes and tiers, each posted after a delay with position sizes removed. The scan ran from 22 June to 4 September. As of today those pages are gone from the site. The scan's own repository is public and keeps all 54 scans and the weeklies, at github.com/ewanlimr25/market-analysis.
I stopped it after asking it three plain questions.
Does it help me plan a trade? No. It had not sized a position since 31 July. The positions it did size returned −1.24% against the market per trade, and 37% of them won. Each lane's average sat between −1.2% and +0.14% before costs, and the scan never modelled costs. To trust any lane, it wanted an amount of evidence that would take about fourteen months of fresh data to collect. With the hundred-odd days I had, even a genuinely good strategy would pass the usual statistical bar less than one time in ten, so anything that did pass was more likely luck than edge. The scan was right to stop sizing positions, and it had no way to start again.
Is options flow a signal at all? Not on this data. I tested 32 trade-level flow measures, each at four horizons: 128 tests, and none survived a correction for testing that many. Directional options flow is retired as a signal.
Can the data support anything? It can price option strategies, which is a different question from which stock goes up. The first idea it priced, selling the volatility around earnings, looked good before costs and lost after them. The second, a SPY put spread that trades only when the volatility curve slopes up and the VIX sits above its recent median, cleared the bar I set before testing it, on three years of history. It is now running on paper, and I will read its record once, on 1 December.
So the nightly scan is gone, and in its place is the trading desk, at ewanli.com/trading. It shows my paper trading on the same day it happens: three Alpaca paper accounts, the simulated books, every trade with the rule's numbers behind its entry and exit, and the P&L as each engine records it. Every strategy there is a pre-registered experiment with a date or a trade count fixed in advance for its one read; until then its record says little either way, and the page says so. None of it is advice. It is a record of what each rule did, kept honestly, including the ones that lose.
The old journal was the evidence for this change, and the evidence was right. A scan that publishes a confident lane every night is more fun to read than a desk that mostly says "not yet". The desk is the one I can stand behind.