claude-code-nse-trading

A swing/positional trading system for Indian equities (NSE), built as a Claude Code workspace — deterministic Python for anything measurable, versioned prompts for anything requiring judgment, and a SQLite layer that scores both afterwards.

It is a measurement system first and a signal generator second.

It is an experiment in a specific idea: that the useful edge for a retail trader is not a better indicator, but a disciplined loop — detect where informed money already moved, genuinely try to kill the idea, size it against a fixed rupee risk, and score the outcome afterwards so the system can be proven wrong.

Status: archived. This ran on real money from July 2026. It is published as a design and engineering artifact, not as trading advice or a product. See What I'd do differently.

Why this might be worth your time

Most public trading repos show a backtest that works. This one includes the backtest that didn't:

**strategy/learnings/07-price-data-alone-found-no-edge.md** — two years of systematic testing could not make a single price-derived entry signal beat buying on a random day. The full teardown is in research/phase1_price_signals/CONCLUSION.md.

That negative result is the most valuable thing in the repo, and it is what pushed the design away from indicators and toward information flow — delivery percentages, bulk/block deals, insider filings, corporate announcements, and relative strength against a stock's actual sector peers.