Explore 21 capability topics, logic and statistics foundations, and practical browser tools—from evaluating evidence to delivering projects.
Understand the world.
Start with a question.
Start with logic and statistics. Practice reasoning, make sense of numbers, and bring what you learn to your own research and projects.
Try one question
From 20% to 30%: what is the relative increase?
An AI answer says ‘a 10% increase’. Calculate it yourself before deciding how to revise it.
Show the calculation
The increase is 10 percentage points. Relative growth divides by the starting rate: (30% − 20%) ÷ 20% = 50%.
Practice percentages and percentage points →What would you conclude?
From 20% to 30%: what actually changed?
A completion rate goes from 20% to 30%. Did it rise by 10% or 50%?
Open this lessonNecessary or sufficient: is registration enough?
'You may enter only if registered.' Does being registered guarantee admission?
Open this lessonWhat does a small p-value actually say?
If p is about 0.04, does that mean the new method has a 96% chance of being better?
Open this lessonExecuted tests · 9 September 2026
Wrong percentage, yet a score of 82?
A deliberately wrong explanation of a change from 20% to 30% scored 82 in the English demo and 73 in Chinese. The rules rewarded structure and signals without checking the arithmetic. Read the inputs, outputs and calculation, then rerun them.
Open the case and checkerInspect the actual grouping of an image-export request.
See a citation marker change review priority.
Inspect what still needs removing from redacted text.
Things I have built
Projects you can open
An HTML5 product experiment around focus sprints and local reflection.
A browser experiment testing whether game structure lowers the friction of starting work.
Browser tools for repeated work and local-first checks.
Research status · 9 September 2026
Two-person review has not started
Twenty synthetic fixtures and a local review workbench are ready. There is no human agreement result. The tool cases above are executed software tests, not conclusions from two reviewers.
View calibration material · Research notes and methods