Notes from the field.

Every tracked asset carries exactly one of five state labels at a time. This is the canonical model — the one used in the Markets table and in Replay's snapshots. Topology's graph uses a simplified three-color version of the same idea for its node coloring; see the Topology graph note for how that mapping works and where it diverges from the full model described here.

Aligned

The asset's tracked pools agree closely with the reference — the default, unremarkable state. TSM (22bp dispersion) and GOOGL (18bp) are both currently aligned: low disagreement, nothing unusual to flag.

Diverged

At least one pool sits meaningfully away from the reference. AMD is the clearest current example: 84bp of dispersion, driven mainly by a single pool (WETH 0.30%) pulling far from where the rest of the market sits. The gap is real and visible, but diverged doesn't say which side — reference or pool — is the one that's off.

Thin

Liquidity near the quoted price is low enough that the reading should be trusted less, independent of how tight the price itself looks. QCOM is a working example: 47bp dispersion isn't the widest on the board, but its depth ($260K across just 2 pools) is among the shallowest tracked, so its state reflects that fragility rather than the price gap alone. Thin and Diverged are independent — an asset can be thin and tightly priced, or diverged with plenty of depth behind it.

Moving

The reading is in the middle of changing — recent blocks show price or dispersion shifting rather than settled. IBM currently carries this label. Moving is inherently a temporal claim, which is why it's best confirmed by watching a sequence in Replay rather than judged from one Markets snapshot alone: a single block can't fully distinguish "moving" from "just diverged."

Stale

The pool hasn't produced a fresh read in a while. The last known price is still shown, but it may no longer reflect current conditions. ORCL is the current example — its 33bp figure is unremarkable on its own, but it's paired with an aging reference, which is what actually earns the label rather than the deviation size.

A known simplification

In the real world these conditions can overlap — an asset can plausibly be thin and moving at once — but the current model assigns exactly one label per asset. Where that happens, the label reflects whichever condition the underlying data judged most significant at that read, not an exhaustive list of everything true about the asset. It's a simplification worth knowing about rather than one worth hiding.