knowngrounds
The public index

What AI gets right, and wrong, about real pages

Every run here asked one model the same questions twice — closed book from memory, then open book with search and page fetching. How this works.

Open book
84%
Closed book
44%
Answers graded correct, across every public run. 40 points of that accuracy exists only while the model keeps fetching the page — open book minus closed book.
14
runs
13
pages
1,618
checks graded
for a free run
Every public run — open to read
Search them, open any report. Sign in to check a page of your own — the first run is free, and there is no password.
Run a check →
14 runs · every claim tested against a frozen question set · page 2 of 2
actionable retrieval-dependent sound not gradeable what do these mean?
Expand your payment options: Top alternative methods | Stripe
https://stripe.com/resources/more/alternative-payment-methods
tested openai/gpt-5.6-luna graded openai/gpt-5.6-terra set 1 30 claims 234 checks
On the pageView report →
open book53.3%closed book53.3%
14
misrepresented
of 30 claims
0%
page surfaced
for its own search
$1.39
cost
6m48s
Stripe Projects - Add OpenRouter via Stripe CLI
https://openrouter.ai/docs/guides/overview/stripe-projects
tested openai/gpt-5.6-luna graded openai/gpt-5.6-terra set 1 29 claims 228 checks
On the pageView report →
open book89.7%closed book20.7%
3
misrepresented
of 29 claims
82.8%
page surfaced
for its own search
$1.28
cost
6m23s
Treasury for platforms requirements
https://docs.stripe.com/treasury/connect/requirements
tested openai/gpt-5.6-luna graded openai/gpt-5.6-terra set 1 21 claims 180 checks
On the pageView report →
open book100%closed book38.1%
0
misrepresented
of 21 claims
100%
page surfaced
for its own search
$1.05
cost
5m03s
Store funds
https://docs.stripe.com/treasury/store-funds
tested openai/gpt-5.6-luna graded openai/gpt-5.6-terra set 1 19 claims 168 checks
On the pageView report →
open book89.5%closed book68.4%
2
misrepresented
of 19 claims
73.7%
page surfaced
for its own search
$0.68
cost
2m12s