RIGOR & VERIFICATION
Receipts & Methodology
If we make a claim, you should be able to see where it came from, how we evaluated it, and how confident we are in the result.
OUR PROCESS
How We Build a Receipt
01 — SOURCE
Start With the Evidence
We begin with primary documents and authoritative data whenever available — statutes, court records, government data, official votes, filings, reports and other original sources.
02 — VERIFY
Cross-Check the Claim
Important facts are checked against additional reliable sources. Disagreement, missing information and conflicting records are documented rather than quietly averaged away.
03 — CONTEXT
Ask What It Actually Means
Numbers and events do not exist in isolation. We examine timing, history, institutional rules, incentives and other context before drawing conclusions.
04 — RECEIPT
Show the Work
Published findings should trace back to their evidence. When new information changes the answer, we update the conclusion rather than defend the old one.
OUR PLEDGE
What We Know. What We Don’t. What Changed.
A verified fact is different from an estimate, an inference or an unfinished research item. We label those differences instead of presenting every number with the same level of certainty.
When something is incomplete, we say so. When credible evidence conflicts, we show the conflict. And when better evidence proves us wrong, the record changes — with the old reasoning preserved when practical so readers can see why.
SOURCE STANDARDS
Not All Evidence Is Equal.
We prefer primary records whenever they are available: statutes, regulations, court opinions, official datasets, government reports, filings, votes, transcripts and original documents.
Secondary reporting can provide context, identify leads and explain events, but important factual claims should trace back to the strongest available evidence whenever practical.
When the source is weaker, incomplete or disputed, we say so.
PRIMARY
Original records, official data, laws, filings, court opinions and direct documentation.
CORROBORATED
Reliable secondary reporting or analysis confirmed by additional independent evidence.
PROVISIONAL
Useful but incomplete evidence, unresolved conflicts, estimates or research still in progress.
NO TIME TRAVEL
History Only Gets to Know What History Knew.
Backtesting becomes meaningless if we allow a model to use information that was not available at the time.
When we test whether a warning could have been detected in 2008, 2020 or 2022, we try to use the information that would actually have been visible then.
Later revisions may be useful for understanding what happened. They are not allowed to sneak backward in time and improve the model’s earlier performance.
REAL-TIME DATA
What was actually available when the decision would have been made?
REVISED DATA
What do later corrections tell us — and do they materially change the conclusion?
NO HINDSIGHT LABELS
Useful but incomplete evidence, unresolved conflicts, estimates or research still in progress.
UNCERTAINTY IS DATA
Missing Evidence Does Not Become a Zero.
When history is incomplete, we do not invent the missing number and we do not quietly treat it as evidence that nothing happened.
Missing information lowers confidence or coverage. Conflicting information remains visible until it can be resolved.
A model that looks cleaner because uncertainty was hidden is not a better model.
KNOWN
Supported by sufficiently strong evidence.
ESTIMATED
Derived or modeled from available evidence.
UNKNOWN
Not enough reliable evidence yet.
CORRECTIONS
Wrong Is Allowed. Hiding It Is Not.
Research changes when better evidence arrives.
If a factual error is found, we correct it. If a model assumption changes, we document the change. If new evidence materially alters a conclusion, the conclusion changes too.
Where practical, we preserve enough of the earlier reasoning to show readers what changed and why.
Our promise
Show the source. Show the reasoning. Show the uncertainty. Fix the mistakes.
The receipts stay attached.