How a finding earns its conclusion.

1. Question

   Every investigation starts with a specific question that the evidence can test, rather than a predetermined allegation or an open-ended search for something interesting.

2. Data provenance

   Identify the sources relevant to the question: public on-chain records, venue-published records, official records, and other verifiable public sources. Each published investigation names its source set.

3. Semantics

   Determine what the fields, events, endpoints and venue representations actually mean before using them — a field can be present and still mean something other than what it appears to.

4. Completeness

   Test whether the data is sufficient for the claim at hand. A completeness signal can read clean on a fetch that is mostly missing; this step exists to catch that before anything is built on top of it.

5. Reconstruction

   Reconstruct the underlying economic activity — fill by fill, transaction by transaction — and reconcile it against any displayed or claimed figure before explaining the result.

6. Alternative explanations

   Actively test competing mechanisms before accepting the interesting one. We first test whether our own data or methods caused the discrepancy. We consider an external cause only after that internal account is closed in writing.

7. Materiality

   Determine whether the pattern is economically or analytically meaningful — a real discrepancy that doesn’t move the conclusion is disclosed, not dressed up as a finding.

8. Conclusion

   State what survives and what does not. “Insufficient evidence” is a valid conclusion, published as plainly as a positive one.

9. Reproduction

   We preserve enough supporting material for a finding to be reviewed, and publish it when permitted.

10. Corrections

    Material corrections are dated, not silently replaced. In public research, a correction is appended to the record; the original entry stays visible.

This is the general standard, not a single engine.

The July 2026 wallet-copyability program applied this standard to one specific question: whether a leaderboard wallet’s profit mechanism was structurally replicable. It used a purpose-built reconstruction engine. That historical program remains public, and its engine, measured error, and limitations are documented in the Reconstruction Workpaper. This page describes the general standard illustrated by that program; it does not overwrite or restate the workpaper.

Commissioned investigations apply the same ten steps to whatever question is in scope — a market, an account cohort, a funding path, or a claim — not only a wallet. See Capabilities for the categories of commissioned work, and Independence for how a conclusion here cannot be bought.

The cartel-arrest markets: two law-enforcement announcements tested for pre-public trading — none found. The displayed $117,153 in trading volume traced to two counterparties trading against each other for a combined net cost of $171.71, not independent directional participation.