Holdout SetEvery hypothesis registered before it was scored. Every verdict published.

Who and why

About

A solo quantitative research programme on Indian equities, run by one person alongside a full-time job. It consists of a tick-capture system that records the order book of a fixed universe every trading day, a decade of daily history reconstructed to be free of survivorship bias with every corporate action hand-verified, a replay engine, and a set of scoring harnesses that measure whether a candidate signal predicts anything.

Everything runs unattended on a single cloud machine. Nothing about this site touches that machine.

What is not on offer

Being explicit about this is not modesty. Each of these would require something I do not have, and saying so is more useful than the alternative.

Why publish at all

Because the failures are the interesting part, and almost nobody publishes them. The public record of retail quantitative research is a highly filtered sample: the strategies that worked, described after the fact, by people with a reason to describe them favourably. The base rate is invisible.

The Ledger is an attempt at the opposite — the complete denominator, including every idea that looked good in preview and died out of sample. If it is useful to one other person building something similar, that is enough.

Contact

Corrections are especially welcome — if something here is wrong, I would rather know. The working code lives at github.com/arkagimt.