Method note
Severity is a screening tool, not a verdict.
A useful alert system reduces a large stream to a smaller review queue. It does not turn unusual activity into proof. This note explains the reasoning boundary behind Polytrack severity labels and the checks a reader should make after a record is flagged.
The classifier answers a narrow question.
Polytrack severity asks how strongly a monitored trade matches configured activity conditions. It does not ask whether the trade will be profitable, whether the trader is informed, or whether any law or platform rule was broken. Those questions require evidence the classifier does not possess.
The distinction matters because a large execution is compatible with many ordinary explanations: portfolio rebalancing, hedging, liquidity provision, conviction, closing exposure, or a trader whose capital base is simply larger. An alert earns attention; it does not settle the explanation.
Several weak signals can describe more context than one cutoff.
A universal dollar threshold ignores market sensitivity, existing flow, price movement, wallet history, and the difference between an isolated trade and a short-window cluster. Polytrack can evaluate trade size alongside short-window volume, change from a market baseline, directional concentration, coordinated-wallet counts, category sensitivity, new-wallet context, accumulated volume, and prior flags.
The strongest triggered condition raises the record's severity. This makes the output useful for queue ordering while preserving the source reasons a researcher can inspect. It does not make the dimensions independent or convert the label into a statistical confidence score.
- Trade size alone is insufficient to infer motive.
- Context signals explain why a record was prioritized but do not prove causation.
- The same observed trade can deserve different attention when the market and surrounding flow differ.
The next action is review, not imitation.
After a flag, the useful path is to open the market and wallet context. Check the outcome and side, nearby activity, liquidity, position state, prior monitored records, and whether a source timestamp or profile field is missing. Compare the record with the selected time window rather than with an undefined idea of normal activity.
Copying a flagged trade would discard most of that context. Polytrack does not execute or copy orders, and severity is not calibrated as an expected-return signal. A researcher should also account for the possibility that market conditions changed between execution, ingestion, processing, delivery, and review.
A responsible public description keeps four negatives visible.
A severity label does not verify identity. It does not establish coordination. It does not establish non-public information. It does not predict the resolution. Repeating those limits is not defensive copy; it is part of the data definition.
Polytrack therefore uses words such as monitored, flagged, tracked, and heuristic on public pages. More dramatic language may attract a click, but it would make the product less accurate and less citable.
Limitations
Keep these with the conclusion.
- This note explains classifier interpretation, not the exact production thresholds.
- Polytrack severity is evaluated only for records the monitor receives and processes.
- The listed signals can be correlated and are not presented as a validated causal model.
- No label is a legal, compliance, identity, or financial determination.
Sources