qNEVAqNEVA

Field notes

Documentation · how State.OS works

Notes from the field.

Full documentation on how State.OS reads, normalizes and displays market state.

Product

Replay

What the recorded snapshots show, and why Replay never predicts forward.

Replay steps through a fixed, recorded sequence of past blocks — dispersion, the widest-diverging asset and its state, a confidence reading, and a short event log, one snapshot at a time. It's the only surface with a time dimension; everything in Markets and Topology is a single block's reading.

What a snapshot contains

Each recorded block carries a block number and a relative time label, the three headline metrics (dispersion, widest state, confidence), and a short log of two or three tagged events — [READ], [POOL], [STATE], [MAP], [REF] or [UNIT] — the same tag vocabulary used in the home page's live State.process panel, just recorded rather than streaming.

A worked sequence

The recorded run opens calm: dispersion at 24bp, MCD aligned, confidence at 96%. Over the next several blocks dispersion climbs as AMD's pools pull apart, peaking at 84bp with confidence down to 78% — the same event described from the Confidence score note, here shown as it actually unfolds rather than at a single instant. From there it eases: by the final recorded block, dispersion has settled to 45bp and confidence has recovered to 92%. Watching that arc is the point of Replay in a way a single snapshot can't replicate — the spike and its resolution are both part of the same story.

Why playback is slow

Each step advances roughly every two seconds when playing. That's deliberate — the point isn't to watch a feed scroll by, it's to actually register how a market-wide reading evolved. There is nothing past the last recorded snapshot: Replay only shows blocks that have already happened. It does not extrapolate, forecast, or imply anything about what the next block will look like.