METHODOLOGY
Dossier watches a curated set of Solana wallets and alerts you when several of them independently move on the same token inside a tight window. This page documents exactly how that works — the entire detection logic, disclosed.
The premise is narrow on purpose. A single wallet buying a token tells you almost nothing — it could be noise, a test, a mistake. But when multiple independent wallets, with no reason to coordinate, land on the same token within hours of each other, that agreement is information. Dossier exists to detect that agreement and put it in front of you fast. It does not predict, recommend, or trade.
Measured, and it did not hold as stated. On clean data in August 2026, 28.5% of alerts described a position the wallet had already closed by the time the message was sent, with a median hold of 32 seconds among wallets we can observe selling. Agreement between wallets is still detectable; the claim that detecting it puts you in front of anything is what failed. The paragraph above is left as written because it was the claim. See correction #021.
A curated cohort — and a deliberate boundary.
The current cohort (codename Beta v2) is 38 Solana wallets selected for a track record of early, independent entries. A high-conviction subset of those is designated tier-A — the Handler feed weights its detection toward them.
Within the system, every wallet is labeled, and subscribers see those labels in their alerts — identifying which wallets converged is the product you pay for. What we publish openly is different:
We publish the method, not the roster. The detection logic on this page is complete and exact. The wallet list itself — the addresses, and which are tier-A — is the proprietary input that makes Dossier worth subscribing to, so it is not posted publicly. This is consistent with our standing commitment to disclose the methodology (logged in the corrections record), which is about honesty in how the system works, not open-sourcing what it watches. The public API and this site never expose wallet identities; counts and rules only.
Four stages, on-chain only.
No private keys. No custody. No access to anyone's funds. Dossier only ever reads the public chain.
Signals and Handler — the exact rules.
Two detectors run in parallel against the same data with different thresholds. Signals is the broad feed; Handler is the stricter, higher-conviction one.
buy the same token
24h lookback
same token, same direction
The difference is conviction. Signals catches early, broad agreement on the buy side. Handler is deliberately rarer: it requires more wallets, a much tighter window, agreement on direction (so it surfaces coordinated exits too, not just entries), and the weight of the tier-A subset behind it.
What this can't do.
- It can be wrong. Convergence is a signal, not a guarantee. False positives happen — coincidence exists, and so do losing trades by good wallets.
- Quiet markets mean fewer alerts. No activity from the cohort means nothing to report. Silence is not a fault.
- It depends on infrastructure we don't own. Detection relies on a third-party data feed; when that feed degrades, ingestion can stall. We make that visible rather than hide it — see the live status page.
- We don't publish performance numbers we can't back. No fabricated win-rates, no invented uptime. When we have verifiable history, it goes in the public forecast record and corrections log.
- The roster evolves. Wallets get added or retired as they go dormant or lose signal. The methodology stays fixed; the inputs are maintained.