Determinism, explained

EdgeDepth replay is 100% deterministic: replaying the same moment produces exactly the same result, tick for tick, frame for frame, on every run. Not “very similar”. The same.

How

Two ingredients, both necessary:

The capture is lossless. Every trade, every order-book delta at full depth, every liquidation print, open interest and funding - recorded as it arrived, nothing sampled, nothing thrown away. Most “replay” tools store candles or periodic snapshots and interpolate between them; interpolation is invention, and invented data varies. Ours has nothing to invent.

Replay re-runs the same engine that renders live. The recorded feed is fed back through the identical pipeline - the same book-building, the same VPIN computation, the same liquidation-field accumulation. Derived instruments aren’t stored and played back; they are recomputed from the same inputs, which is why they land on the same values every time. Same inputs, same code, same output. There is simply nothing left to vary between runs.

What non-determinism looks like elsewhere

It’s worth knowing what this replaces. Chart-scroll “replay” plays candles, so the book, the tape and every microstructure instrument are missing entirely. There’s alot that happens within each candle. Snapshot-based heatmaps bake periodic images of their state, so rewinding shows you approximately what the map looked like, at whatever cadence the snapshots ran. Sampled feeds drop ticks under load - precisely during the violent moments you’d most want to study. Each of these is a small edit to history. They add up to studying a market that never quite existed.

Why it matters

• Repeatable study. You can’t drill your read of a move if the move changes on every viewing. Deterministic replay makes practice work like film review: rewind, re-watch, test your read against the same tape, as many times as it takes.
• A shared reference. Every student of an archived event sees exactly the same market. Lesson annotations pin to exact ticks, and those ticks never move - an author can write “watch the bid at 60,791 pull at 03:47:12” and it is always there.
• Honest hindsight. A replay can’t be cherry-picked or redrawn. What printed is what plays. Any claim anyone makes about how an event unfolded - including ours - is checkable against the tape.
• Measurable instruments. Determinism is what lets us score the tools instead of marketing them. The liquidation heatmap’s capture numbers exist because we can re-run history against the model and get the same answer every time. An approximate replay could only produce approximate claims.

Determinism sounds like an engineering detail. It’s actually the foundation: the archive, the lessons, the scoreboard and the study method all stand on it.

The same principle extends beyond replay. A Research result carries the query grammar version, feature-library version, dataset revision and canonical query hash. The same document under the same receipt returns byte-identical result bytes whether it came from the browser, the public API or an AI agent through MCP. Replay makes a moment repeatable; the receipt makes a market-wide claim repeatable.