Leadership Under Uncertainty
Application for analysts

Tradecraft review

Paste a note, memo or briefing and the review reads it against the analytic standards of Intelligence Community Directive 203. It scores the draft, names the actions that would improve it most and marks each paragraph that falls short with the standard it misses.

A review takes about half a minute. The note is sent to a model provider for the reading. The project keeps the score and a trace identifier, holds the model’s reading for up to an hour, and does not store the note. The two samples are synthetic notes from the project’s test set.

Tradecraft reviewChecking
0 of 10,000 characters
Load a sample:
What the review checks

Five of the nine standards can be read from a single note

Sources
Where the information comes from, and how reliable each source is.
Confidence
How likely each judgment is, in calibrated words rather than false precision.
Facts and judgments
Which statements are reported and which are the analyst’s assessment.
Alternatives
Whether another explanation of the evidence is considered.
Assumptions and gaps
What the assessment rests on, and what is not known that could matter.

The other standards concern a body of work over time, its relevance to its readers, its accuracy once events have run and its presentation, and a single note does not carry enough to judge them. The review reads the text and not the world: it does not check that a cited report exists or says what the note says it says. The score combines a model’s reading of the note’s structure, its judgment on seven criteria and a set of fixed checks, and a model’s judgment can be wrong. The same engine serves the project’s application and its tools for AI agents.