Tools
Each tool takes a draft and returns a review against named tests. Every run is kept and dated, so you can see what changed between versions.
The three counts
Reads retention, the judge’s answers and your verification count from one real piece of work. It refuses an estimated retention proportion and it will not accept a judge who was told AI was involved.
Not openedFrontier map reader
Reads your map of which subtasks sit inside or outside the frontier. It rejects placements made without a comparison, and it flags a map where nothing moved from where you expected.
Not openedSpecification checker
Classifies every line of your specification as material, requirement, format or none of these, and flags the techniques that have been tested at scale and do not work.
Not openedReview record
Records what you changed, what you kept and what you caught across six outputs. It requires at least one output discarded entirely and at least one case where the fluent version was the wrong one.
Not openedVerification log
Logs ten load-bearing claims checked against primary sources, with the time each took. It distinguishes checking that a source exists from checking that it supports the claim, and computes your verification time ratio.
Not openedPractice document check
Checks the six parts of your personal practice and the two things people leave out: what you never verify, and your disclosure wording.
Not openedThe ten statement check
Ten statements, marked true or false about you, before Level 0 and again after Level 5. It reports the individual lines rather than a total, because the total hides the only movement worth reading.
Not opened