The open dataset
Every verdict this site published, next to what happened to that token afterwards. One row per scanned token: the band and score it carried, the model version behind them, and the outcome once a checkpoint recorded one.
Download latest.csv — rebuilt nightly. It currently covers 1821 scanned tokens and 2747 recorded outcomes, the same two figures the home page publishes. The last 7 daily snapshots stay online alongside it.
Why this is public
This site's central claim is that its verdicts were written down before the outcomes were known. A claim like that is worth exactly what somebody else's ability to test it is worth — so the pairs are published, and anyone can recompute the accuracy page from this file and see whether the numbers hold.
The snapshot is free and always will be. What it is not is the platform: it carries no evidence records, no provider payloads, no block-pinned readings, and one row per token rather than the full history behind it. The live stream and that history are what this site sells.
The columns
| Column | What it holds |
|---|---|
scan_id | This site's identifier for the scan the row describes. |
chain | The chain the token is on. |
address | The token's contract address, or its mint on Solana. |
token_name | The name the contract reports. |
token_symbol | The symbol the contract reports. |
first_scanned_at | When this site first scanned the token, as an ISO-8601 timestamp in UTC. |
verdict_band | The risk band the verdict carried: Low risk, Guarded, Elevated risk, High risk or Critical risk. |
verdict_score | The score out of 100 behind that band. |
model_version | The claim-engine version the verdict was computed under. Rows from different versions are comparable within a version and should be grouped by it. |
verdict_computed_at | When that verdict was computed, as an ISO-8601 timestamp in UTC. A prediction counts only when this precedes the outcome timestamp. |
outcome | What happened later, when a checkpoint recorded it: rugged, dumped, declined, alive or grew. |
outcome_at | When that outcome was recorded, as an ISO-8601 timestamp in UTC. |
incident_id | The permanent reference this site gave the collapse, in the form VMT-INC-YEAR-NUMBER. Empty when no collapse was recorded. Numbers are allocated once and never reused, so a row can be cited and looked up later. |
Timestamps are ISO-8601 in UTC. A row with an empty outcome is a token still being watched, and it is in the file on purpose: publishing only the resolved cases is the shape every flattering dataset has.
Licence and citation
Released under CC-BY-4.0. Use it, redistribute it, build on it commercially — the one condition is attribution.
Cite it as:
VetMyToken incident and calibration data, vetmytoken.comReading it correctly
Group by model_version before comparing scores. The claim
engine has been revised several times, and a score means something slightly
different under each version; rows carry the version precisely so that
comparison stays honest.
A verdict counts as a prediction only when verdict_computed_at
precedes outcome_at. The accuracy page applies that same rule, and
a row where the verdict came later is a measurement of the past rather than a
forecast of it.
Following specific tokens instead? Every token page, deployer page and the incident registry carry an RSS feed — Watch via RSS — no account needed.