The ledger
Claims, and what they rest on
Every essay here says it is tested rather than asserted. This is the test, made public. For each essay that went through the full line: the belief it argues, the ladder of claims it stands on with what each one does not reach, the condition under which it would be wrong, and every load-bearing claim with the primary source it was checked against, the verbatim words, where in the source they sit, the day they were read, and the claims that were struck before publication.
175 claims · 8 essays · last verified
-
Everyone Got Safer. That's the Problem.
Safety has two numbers: how often each system fails, and whether they fail together. The field has spent years driving the first one down while almost no dashboard reports the second.
19 claims 0 verified to primary
-
Confidently Wrong
The effort you're handing to AI was doing two hidden jobs. Skip them and you get faster, weaker, and blind to your own mistakes.
11 claims 10 verified to primary falsifier stated
-
A Green Score Is Not Evidence
A groundedness metric scored its best with the evidence removed. The way to tell whether your model is actually using its evidence is to change the evidence and watch what moves in the answer.
12 claims 10 verified to primary falsifier stated
-
The Limit Said 10. The Loop Made 500 Calls.
Your limit counts one cycle. The one that runs away is another. Here is how to tell them apart.
34 claims 21 verified to primary 12 first-party runs 3 struck falsifier stated
-
The Guardrail Your Agent Can Reach
Most guardrails end up with an escape hatch. Check whether the thing you are constraining can reach yours.
20 claims 20 verified to primary 2 struck
-
The Average Is Nobody's Result
Of 255 studies on AI-assisted colonoscopy, 21 split the result by who held the scope. They disagree.
35 claims 34 verified to primary falsifier stated
-
You Cannot Try to Fall Asleep
Sleep is only where you notice it first. Much of what matters works the same way.
20 claims 15 verified to primary 1 struck falsifier stated
-
Your Robot Coworker Is Still a Pilot
Built, shipped, installed, working: four counts, quoted as one. Almost nobody publishes the fourth.
24 claims 22 verified to primary 5 struck falsifier stated
How to read a row
- Verified
- The primary source was opened and the quoted words read off it, by a checker who had not written the essay. Nothing reaches this status from memory or a search snippet.
- Executed
- A first-party run: code, a query, or a count we performed ourselves. The claim is what the run printed; the method is named in the row.
- Reported
- Carried from a secondary source or a source we could not open in full. The row says which, and the essay's wording is bounded accordingly.
- Struck
- A claim the check removed before publication, with the reason. A ledger that only shows what survived is half a ledger.
Quote any row with attribution and a link to its ledger page; the page URL is the citation. The
ledgers are also available as plain text for machines: /llms.txt (index),
/llms-full.txt (full), and /md/claims/<slug>.md per essay.