Research Log
Each post traces to a dated Decision Record written at the time, not reconstructed afterwards. Where a post disagrees with an earlier one, both stay up — the record is append-only.
- How raw NSE Bhavcopy beat premium vendors at survivorship-free data
Why most Indian equity datasets quietly inflate every backtest run on them, how to rebuild a decade of history that doesn't, and the eight-month hole the acceptance tests found in my own archive.
- We built a gate, then failed it twice in one day — on purpose
How pre-registration works in practice, why a threshold chosen after the result isn't a threshold, and what it costs to run a research programme that can tell you no.
- Eleven hypotheses, two rounds, zero trades
A year's worth of quantitative research discipline compressed into six weeks, and what it cost to find out that nothing worked.