First seen is not first trade: a 500-wallet audit

A wallet can look new because the observation window is new. Vultax checked 500 addresses from a historical fill dataset against the oldest TRADE returned by Polymarket's public activity API. In the January 2025 first-seen stratum, 134 of 250 had traded earlier. In the later-first-seen stratum, 17 of 250 had traded earlier. These are two deliberately equal-sized samples, not estimates of the whole platform.

The difference matters when an alert describes a large bet as coming from a fresh wallet. An earliest recorded fill tells a reader when a dataset begins to observe that address. It does not identify account creation, the arrival of funds, a person's identity, or access to private information. The same distinction applies to a vendor's impressive-looking timeline.

The archive starts on 1 January 2025. Among the 134 January-stratum discrepancies, the median earlier-history gap was 42.5 days. Among the 17 later-stratum discrepancies it was 145.6 days. That second number describes the flagged subset only; it is not the typical age of later-seen wallets.

Download table figures:CSVJSON with sources

Download chart figures:CSVJSON with sources

An older internal calibration used the earliest activity of any type. This follow-up queries TRADE explicitly. Minting, merging and redeeming positions can have different timestamps from a trade. Reusing the earlier all-activity rate as a first-trade rate would mix definitions. The later-stratum trade result is 6.8%; the internal all-activity calibration was 8.4%. Neither should be presented as an account-birth statistic.

No sample address is identified as suspicious. The study measures a data boundary. An alert vendor could reduce ambiguity by publishing the observation start, the earliest observed trade, the oldest public trade checked, the check time, and an explicit unknown state.

The fixed sample was originally selected on 16 September 2026 by ordering wallet hashes in each stratum and taking 250 addresses with at least five fills. The source archive covers 1 January 2025 through 28 April 2026 and observes maker and taker addresses. This follow-up does not regenerate that full fill archive; it audits the frozen first-seen values against newly retrieved public histories.

For each address, the API query requests type=TRADE, sortBy=TIMESTAMP, sortDirection=ASC, start=1 and limit=1. We retain request times and response hashes, check the address and activity type, and compare Unix timestamps. All 500 histories were available, and none returned an earliest trade later than the archived first fill. A complete response is not proof that the upstream history is complete.

We report strata separately. We do not weight them using the share of all wallets in the archive: the sample requires at least five fills, and the eligible-population weights were not verified. The intervals also do not account for multiple addresses controlled by one operator.

wallet-age-audit.csv · results.json · METHODS.md · Reproduction bundle. The bundle contains the calculation, frozen analysis inputs, results and a SHA-256 manifest. Public inputs use sample IDs and timestamps, omitting wallet identities. They reproduce the arithmetic; they do not permit an independent address-by-address API re-fetch. The private research archive retains that provenance.

Reuse the figures with a link to this study and the stated population and observation window. Vultax welcomes corrections and independent methods reviews. A citation or collaboration does not require a positive conclusion or a reciprocal link.

Freeze an alert rule before the next observation window, record every alert and every eligible non-alert, and retain the same evidence for failures as for successes. Add a first-seen evidence field without treating the field as proof of wrongdoing. A partner can contribute a small historical alert export and help define the denominator. Vultax can contribute the timestamp audit and publish disagreements alongside matches.