The Claim Ledger
Claims, and what they rest on
The claims behind 23 of the 88 pieces here, each with the source it was checked against and a label that says how far it was checked. Claims that failed the check stay visible, crossed out, with the reason.
- Claims
- 544
- Verified at the primary source
- 440
- Ledgers
- 23
- Removed before publication
- 14
Verified rows were checked against the primary source itself, usually with the exact words, where they sit and the date read. Last checked . Download every row: CSV · JSON, or query them from Claude or ChatGPT with the connector.
How to read a row
- Verified
- Checked against the primary source itself by a checker who did not draft the piece, usually with the exact words, where they sit and the date read.
- Executed
- A run or observation done for the piece: code, a query, a count, or a check made live in an app. The claim is what it returned. Run records are not published.
- Checked
- Checked against its source by a separate checker in the older ledger format, which recorded no quote, locator or date.
- Reported
- Not verified word for word against a primary source: carried from a secondary source, from one that could not be opened in full, or supported only in part. The essay words it accordingly.
- Struck
- Drafted, checked and removed before publication. Kept here, crossed out, with the reason.
- Excluded
- Considered and deliberately left out of the piece. Kept here with the reason.
Primary source: the original paper, filing, page or dataset itself. Locator: where in the source the words sit, such as a section, table or page. Load-bearing: a claim the argument needs; if it fell, the piece would change.
The ledgers
-
AI Can Take the Practice Without Taking the Job
The practice that builds judgement can disappear while every junior keeps their seat. This audit checks whether it has on your team.
67 claims: 63 verified, 4 executed
Last checked
-
Your AI Grader Is Only as Good as Its Answer Key
I went through 77 studies of AI graders. Without a verified answer key, they are least reliable on exactly the questions they can't answer themselves. Here is a five-step check on twenty items for yours.
68 claims: 57 verified, 11 executed
Last checked
-
Amazon Shut Out Meta's Agent. Shopify Gave It Shop Pay.
Also this week: an OpenAI agent inside an Australian government portal, another supplier handing a big customer the right to buy its stock, and six calls I've put a number on.
33 claims: 33 verified
Last checked
-
One Question for Any AI Model Headline: Can You Use It?
It may be in a mode you didn't open, on a plan you don't have, or kept for partners. Here's the check.
21 claims: 18 verified, 3 executed
Last checked
-
Use AI to Learn Without Getting Worse at It
Answer-giving AI left learners worse off once it was gone. Three setups built to keep the thinking with you, and a way to check what stuck.
70 claims: 57 verified, 7 executed, 6 reported
Last checked
-
Your Health Has a Dashboard. Your Relationships Don't.
One you schedule. The other you leave to a mood.
10 claims: 10 verified
Last checked
-
You Can Win the Wrong Game for Years
You usually cannot tell whether you are playing the wrong game. Write down what would prove it, and by when.
17 claims: 11 verified, 6 reported · plus 1 removed before publication
Last checked
-
The Planes That Didn't Come Back
Success stories and the antiques that outlasted the rest. Your evidence was filtered by what survived to reach you.
7 claims: 7 verified
Last checked
-
The Graph Looks Right. The Merge Is Where It Breaks.
A multi-agent graph can lose a verdict with no error and nothing turning red. The bug is one line in the state schema.
19 claims: 4 verified, 14 executed, 1 reported
Last checked
-
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.
17 claims: 15 checked, 2 reported · plus 2 removed before publication
Ledger dated
-
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, 1 reported
Last checked
-
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, 2 reported
Last checked
-
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, 12 executed, 1 reported · plus 3 removed before publication
Last checked
-
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 · plus 2 removed before publication
Last checked
-
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: 23 verified, 11 executed, 1 reported
Last checked
-
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, 5 reported · plus 1 removed before publication
Last checked
-
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, 2 reported · plus 5 removed before publication
Last checked
-
The Difficulty You're Escaping Was Making You
AI can lift the effort out of almost anything you find hard. Some of that effort was the thing turning you into someone.
9 claims: 9 verified
Last checked
-
Your Research Agent Cites Sources It Never Read
The same trap has killed pricing models and trading desks for decades. One move tells you if your number is next.
11 claims: 11 verified
Last checked
-
Your Multi-Agent System Is an Org Chart
Cognition said don't build them. Anthropic said do. A year on, they converge on the one question that decides it.
9 claims: 9 verified
Last checked
-
The Seven-Layer Agent Audit
Your agent is starved on one layer of seven. It is rarely the harness everyone argues about.
8 claims: 8 verified
Last checked
-
Skills Are Package Management for Your AI
There are millions of agent skill files to choose from. The hard part is deciding which deserve trust, how to version them, and how to retire them.
8 claims: 8 verified
Last checked
-
Same Model, Different Product: The Case for Harness Engineering
Hold the model fixed and the surrounding harness can still move task success, cost and reliability. Here are the layers that do it.
14 claims: 14 verified
Last checked
How checking works
What goes in
The load-bearing claims a piece rests on: the figures, quotations and factual statements its argument needs. Opinion stays out.
Who checks
For a verified row, a checker who did not draft the piece opens the source and records what it says, usually with the exact words, where they sit and the day they were read. Runs and live checks the site did itself are labelled Executed. Anything short of a full check says so in its label. What fails is struck and kept, with the reason.
What is not covered
23 of the 88 pieces have a ledger so far. The others predate it or were not run through it, and make no claim to this standard.
When something is wrong
Every row has a “Report an error” link. Changes to a published essay are logged on the corrections page, or write to [email protected].
Cite and reuse
The ledger data is licensed CC BY 4.0: reuse it, quote it or train on it, with credit to The Durability Curve and a link. Read the licence.
Every row has its own link (/claims/<piece>/#c-<id>) and a ready-made citation that names the original source first. Each ledger downloads as CSV, JSON and Markdown from its own page.
For machines: all ledgers as JSON, as CSV, as Markdown, the site index at /llms.txt, and a read-only MCP connector for AI assistants at /connect.