Project H — systematic options research, forward-tested in public
What this is
Project H is a research engine for systematic options trading in the Indian F&O market — the liquid derivatives universe, fully automated end-to-end, built and run by one person.
The system rests on three commitments:
-
Systematic selection. Machine-learning models score the tradeable universe every morning, and the book is built mechanically from those scores. No discretionary overrides, no gut calls at 9:20 AM.
-
News as a first-class input. Pre-market filings, wire headlines, broker actions and macro events are continuously read, scored and fed into the same decision layer as the price-based models — as one vote among many, not an oracle.
-
Risk before return. Position sizing, stop discipline, profit-taking rules and time-based exits are codified in the engine itself. The P&L is what's left after the risk has been respected.
The honest premise
Here's the part most projects like this skip: none of it is presumed to work.
Every strategy in the system is a hypothesis. Backtests — however pretty — are treated as reasons to run a forward test, never as proof. So the entire book trades on paper, marked against real market prices, and the resulting scoreboard is published live on this site: every run, its capital, its P&L, on good days and bad ones.
A strategy earns trust one way only — by surviving out-of-sample, in public, over enough time to mean something. Until then it's research. That bar is the project.
Why publish any of it
Most retail trading content is either a chart with adjectives or a subscription to someone else's unverifiable signals. The interesting question — how good can a single, disciplined person actually get at this with modern AI tooling and public data? — almost never gets answered in the open, because publishing the scoreboard is uncomfortable.
Project H publishes the scoreboard. If the answer turns out to be "better than expected," that's a real data point about the shrinking gap between institutional infrastructure and what one developer can build. If the answer is "the market is efficient and the edge isn't there," that gets published too — it's the more common outcome, and pretending otherwise is how this genre earned its reputation.
What stays private
The engineering log shares the thinking: design choices, modelling philosophy, validation discipline, the experiments that failed. The recipe — exact signals, features, weights, parameters and the code itself — stays private. Principles in public, mechanics in the repo.
What this site is not
Not a signal service. Not a tipsheet. Not investment advice. No real capital is deployed — everything you see is a paper-traded forward test. If you're here for trade calls, this is the wrong site; if you're here to watch a research process run honestly in public, welcome.
Where to look
- Live P&L — the paper scoreboard: every run, its capital, its P&L, updated through the session
- The log — posts on how the system is designed and what the forward tests teach us, as they teach it