How We Report & Verify

BlackLeaf is built around a simple bet: if every number is traceable to a named public source and the analysis is reproducible, the work is more auditable than reporting you're asked to take on trust. Here is how a piece is made.

Data first, from named sources

Reporting starts from primary public data — government reports, agency filings, budgets, audits, regulatory dockets, official statistics. Every figure that reaches an article is tied to a specific, nameable source. Where sources sit in a hierarchy (a primary dataset vs. a secondary summary of it), we prefer the primary and say which we used.

The data is shown, not just cited

Beneath every chart, the underlying table is available. The claim and the evidence for it live on the same page, so a reader can check the arithmetic rather than trust the headline.

Analysis is recomputable

Our strongest pieces carry a machine-recomputable analysis: the figures can be regenerated from the cited inputs, so the path from source to stated number is reproducible rather than asserted.

Blind adversarial verification

For gold-standard investigations, the numbers are checked by an independent verification pass that tries to breakthe claim — re-deriving figures from the primary sources without seeing how the draft reached them — and the result is recorded with the piece. A claim that can't survive that check doesn't run.

Human editorial accountability

Articles are AI-drafted under this method and reviewed by a human editor who verifies the sourcing and is accountable for publication. What the AI does and doesn't do, and where the human sign-off sits, is spelled out in the AI & accountability policy.

When we're wrong

Flag any article. Inaccuracy flags trigger a blind re-verification of the disputed claim; the outcome — corrected, clarified, or upheld — is logged. Right of Reply carries responses from those we cover. See our editorial standards for the full set of rules.