Task completion
Did the app complete the same requested study task?
Method · Evidence · Corrections
How we test AI study apps.
We use the same permitted materials and task instructions, then separate sourced facts from reproducible test observations.
No product scores or “best” verdicts are published until the benchmark set, procedure, date and evidence are available.
What we test
Factual support
Can important claims be traced to the supplied material?
Citation fidelity
Do links identify the correct source page or passage?
Formula handling
Are notation, units and transformations preserved?
Quiz quality
Are questions answerable, distinct and supported?
Exports and limits
Are formats, gates and free limits recorded with evidence?
Test procedure
Freeze permitted sample materials
Run the same task instructions
Check claims against sources
Record evidence, failures and limits
Sample materials
Owner-created, public-domain, permissively licensed or explicitly permitted materials only.
Free-plan checks
Signup, card requirement, task limits, file limits, export gates and reset timing require dated evidence.
Evidence and verification
Marketing copy is not treated as independent test evidence.
Sponsorship policy
Commercial relationships never change the test result or editorial placement.
Corrections
Corrections retain the source, reason, reviewer and change date.
Report a correctionVersion history
The public revision log remains empty until a benchmark release is approved.