Where the cases come from, what counts as evidence, and how every comparison is kept honest.
Every incident in the book comes from OnChainAttack — a public, free catalogue of 650+ documented crypto incidents from 2010 to 2026, maintained by the author, with a citation for every claim. Cases were chosen for mechanism, not fame: each chapter needs one crypto danger that recurs across many incidents, and a twin of that danger that predates crypto. A spectacular one-off that teaches nothing general did not make the book.
Every load-bearing claim prefers primary sources, in this priority:
Secondary summaries are used only where no primary evidence exists. Numbers from interested parties — vendor estimates, industry lobbies — are named and sourced; when such a number only flatters the book's argument, it is declined, in the text, visibly.
The book separates what is established from what is claimed, and never strengthens a verb the record does not support. Documented and established are reserved for court judgments and regulator findings. Investigators concluded covers forfeiture actions and settlements without admission. According to the indictment and prosecutors say mark charges — because charged is not convicted, and a settlement is not an admission. Where a link is contested, the book says the company disputes this, and means it.
One rule governs every comparison: like-for-like. Same year, same categories, both sides. If a category of loss is counted on one side, it is counted on the other. Rates whose denominators measure different activities are refused even when they flatter crypto — on-chain volume and card commerce do not measure the same thing, so no rate built on them appears in the book. Where sources give ranges, the book prints ranges. Moving figures are dated in the sentence that uses them.
Every load-bearing number lives in a manifest with its source and as-of date. The public slice of that manifest is the verification index: claim by claim, next to the public artifact that checks it — on-chain transactions anyone can open in a block explorer, court dockets, regulator reports.
Mistakes are expected and corrected publicly — upstream in the research catalogue first, then in the book's errata. A found error, credited to its finder, strengthens the record. A hidden one destroys it.