
Intelligence briefings
Written positions on how financial evidence should be found, tested, quantified and defended. No detection logic, thresholds or analytical methods are published.
Library
Two systems can agree perfectly and still describe something that never happened.
Evidentiary discipline is standard in physical investigations and rare in financial ones. That gap is a liability.
A sample designed to measure a rate is usually incapable of reaching a concentrated exposure.
Most anomalies have a reasonable explanation. The work is testing it rather than accepting it.
If a number cannot be reproduced exactly by someone else, it is an opinion.
Entitlement questions are answered by the agreement in force on the date of service, not the agreement on file.
A finding should state what the evidence supports and nothing beyond it.
Language models are useful for extraction and drafting. They are unfit for calculation and signoff.
The strongest evidence is usually a combination that cannot be true, not a value that looks wrong.
Traceability, population, method and limits. If any answer is missing, the number is not ready.