Method

Everything published here is reproducible by one person with free tools. The pipelines:EDGAR XBRL (companyfacts and frames — fundamentals, buyback series pulled from cash-flow statements rather than press coverage), Form 4 parsing (the insider ledger, dollar-weighted), full-text search across filings (counterparty graphs — who lends to, hosts, and guarantees whom), TreasuryDirect auction PDFs (dealer takedown as the demand truth-serum, minutes after each print), and primary transcripts (paraphrases from live-listened calls are marked until transcript-verified).

Rules of the shop, published because they bind: secondary sources are date-checked ruthlessly — the recurring hazard of this season is stale articles wearing fresh headlines — and anything that matters is re-derived from the primary before use. Every screen logs what it dropped, because silent truncation reads as coverage. Statistics computed on whatever data happened to survive are labeled as such and cannot support conclusions. Where two denominators disagree, the one that makes the thesis look worse is canonical. And when a claim's strongest counter-evidence emerges from our own verification work, it publishes with the claim — the antithesis on this site exists because that rule has no exceptions.

The method that produced this paper's falsification discipline was field-tested first at smaller stakes: see The Forge, a field note on how a $1,000 microcap research lab industrialized pre-registration, buried exits, and one validated edge.