From the exchange to a labelled number
Four stages, and the order matters. Nothing is read before the data behind it has passed a quality gate, and nothing is published before it has a record.
Data comes in once, on a schedule
The tradable universe is refreshed each morning from the exchange’s instrument list. History is filled for everything anyone follows, topped up daily, and checked for retroactive adjustments — a corporate action rewrites past bars, and a store that never re-reads them slowly drifts from reality.
A quality gate runs before any reading
A series with too few sessions, a stale last bar, or a gap where a session should be is refused, not read. Refusals are reported with the reason. A read built on bad bars looks exactly like a read built on good ones, which is what makes this worth doing first.
Indicators and levels are computed, then narrated
Moving averages, ATR, RSI, VWAP, supertrend, swing structure, support and resistance zones — all arithmetic, all reproducible. Where a language model writes the explanation, it is handed those numbers as authoritative input and its response format contains no numeric field at all, so it cannot introduce or alter a price.
Every rule is replayed before it is trusted
Walk-forward, net of Indian costs, against a random-entry baseline with identical brackets. The outcome is one of four labels, and only the strongest permits the product to treat a rule as an edge. Published reads are then scored against what actually happened next.
Where your data lives
Market data
Fetched into our own Postgres. Candles, option chains and the instrument list. Shared across accounts — a price is a public fact.
Your account
Email, a hashed password, your simulated book and your trade journal. They are yours; nobody else sees them.
Your broker connection
If you connect one, the access token is encrypted at rest with authenticated encryption. It expires daily by the broker’s own rule and is renewed only by you completing their login.
One thing we cannot do, and why
Historical data for expired option contracts is not available from our data provider and is not on their roadmap. That means an option strategy cannot be back-tested on real past chains — only on chains we have captured live since starting. Where an option result is modelled from a pricing formula instead, it is labelled as modelled, not measured. We would rather say this than let a simulated figure read as a historical one.