How the ledger is kept

A case needs all three:

1. A named organisation.
2. Roles reduced: layoffs, contracts ended, or posts left unfilled on purpose.
3. The organisation, or credible reporting, linking the reduction to AI or automation, backed by one primary source or two independent reports.

Rumours, forecasts and cuts with no named organisation are left out. When reporting questions an organisation’s link to AI, the case records both, and draws no conclusion about motives.

A news sweep runs every week. A small, fast model reads each headline and sets aside the ones that cannot be cases. A larger model researches the rest and judges how well each one shows the question at the centre of the site: what else was the person doing?

For each case worth writing, the sources are fetched, archived on the Wayback Machine, and split into numbered sentences. The writing agent writes statements and cites sentences by number. It never types a quote. Code copies each cited sentence into the case, word for word.

1. Every quote is found word for word on the archived copy of its source. No model is involved in this check.
2. Every statement is checked against its quote by a separate model. A statement that says more than its quote, or suggests a motive, goes back to the writer once, and is removed if it still fails.
3. Loaded words are not allowed.
4. Every case has one primary source, such as the organisation’s own statement, or two independent reports.
5. The case keeps a fixed shape: the role, the change, what happened next, and open questions.

A case that fails is not published.

Each case keeps two layers apart. What sources report is marked as reported. Which hidden work the change may have missed is the site’s own reading, marked as analysis, and each point names the reported evidence it rests on.

A case is a record that keeps changing. Each one is checked again on a schedule set by its status. What each status means, and when it is checked, is on the ledger page under Statuses and follow-up.

If something in a case is wrong, use Report an error on its page, or suggest a case. The desk checks the report against the sources. Corrections are listed on the case page, and every change to a case is public in its history on GitHub ↗.

AI agents research and write each case, and Jev, by TypeSafe, runs the fast checks. The code is open source under the MIT licence ↗; the cases and the text of this site are published under CC BY 4.0 ↗, so you can reuse them with credit.

Is this site against AI?

Its view is that AI should do the work it does well, and that the people deciding should count everything a role did before replacing it. Cuts that held up are recorded alongside those that were reversed.

Is a case a judgement of the organisation?

A case records what sources reported, with quotes. The doorman reading is marked as the site’s own analysis and names the evidence it rests on. Motives are never guessed.

Why are there so few cases?

The ledger is new. The desk adds cases every week, and only those that pass every check.

Can I use the cases?

Yes. They are published under CC BY 4.0: reuse them with credit to The Doorman Fallacy. The whole ledger can be downloaded from the ledger page.

Who pays for this?

Nobody. The site has no advertising and no sponsors, and its code is open.