Skip to main content
All AI tricks
Guided workflowProductivityBeginner

Promise Ledger: Track AI Predictions That Keep Slipping

Dates and targets in AI keep moving. Build a simple ledger of claims, who made them and what you can check, and mark each one met, moved, dropped or unchecked.

25 min practicePlatform independentUpdated · Last checked Nov 2, 2026
Jump to the lesson ↓
A ledger table with five rows and four colored status tags: met, moved, dropped and unchecked.
Designed illustration; not a screenshot or a recorded model result.

The useful result

A five-row promise ledger you can update monthly, with an honest status on each row.

  • Build a five-column ledger of claims, speakers, dates, checkable evidence and status.
  • Use four statuses (Met, Moved, Dropped, Unchecked) without letting AI decide truth for you.
  • Run a short monthly review that separates open date changes from quiet ones.

AI news is full of dates and targets: when a law starts, when a robot ships, when a service reaches a number. Many of them move. A ledger is a small table that turns that noise into something you can check.

Here are real examples from 2026, each read on its source page on 2 November 2026. The EU moved the date for standalone high-risk AI rules from 2 August 2026 to 2 December 2027, and for AI inside regulated products from 2 August 2027 to 2 August 2028 (Gibson Dunn, May 2026; Addleshaw Goddard reported formal adoption on 1 July 2026). Elon Musk said in early 2025 that Tesla would build roughly 10,000 Optimus robots that year; Electrek (2 July 2026) says Tesla missed that badly, and Musk now says production will be extremely slow at first. Waymo said it reached 500,000 paid rides a week (26 March 2026, up from 250,000 in April 2025) and aims for 1 million by the end of 2026, which cannot be checked until then. The authors of the AI 2027 scenario said in November 2025 that things are going somewhat slower and that their answer is now around 2030. At DevDay on 29 September 2026, Sam Altman said OpenAI has an AI research intern, and an OpenAI researcher said models complete more than a third of day-long research tasks; The Next Web notes no independent check was given.

The skill is not predicting. It is keeping score.

Before you start

  • Five claims or targets you have read about AI this year, with the link and date for each.
  • A spreadsheet or document, and any free AI chat assistant.

Review evidence

Documentation-based review · 2026-11-02

Sources opened and checked on 2 November 2026. Figures are dated and describe what the sources reported, not rules. Prompts and examples are illustrative; no model output or outcome is claimed.

1

Make the ledger from five claims

Pick five claims or targets about AI that you have actually read. Good ones have a speaker and a date. For each one, write the exact claim, who said it, the date or deadline, and the link where you read it. Do not rely on memory or on a chat tool for the facts. Then ask an AI chat only to format what you typed into a table, and to flag any row with a missing source or date. The five columns are: Claim, Who said it, Date or deadline, Evidence you can check, Status. Leave Status empty for now.

Try this prompt
Format the five claims below as a table with these columns: Claim | Who said it | Date or deadline | Evidence I can check | Status. Copy my words exactly. Leave Status empty. If a row has no source link or no date, write "MISSING" in that cell and do not fill it in. Do not add, correct or comment on any claim. Do not search the internet.

MY CLAIMS:
[PASTE YOUR 5 CLAIMS WITH SOURCE AND DATE]

Example / expected result

Illustrative example; not a recorded model result.

| Claim | Who said it | Date or deadline | Evidence I can check | Status | | EU standalone high-risk AI rules start 2 Aug 2026 | EU AI Act original timetable | 2 Aug 2026 | (law firm summary, link) | | | Roughly 10,000 Optimus robots in 2025 | Elon Musk | end of 2025 | (Electrek, link) | | | 1 million Waymo rides per week | Waymo co-CEO | end of 2026 | (InsideEVs, link) | |

Check your result

Every cell matches what you typed, and each row has a source link you opened yourself. The Status column is still empty.

2

Set statuses with fixed rules

Use four statuses. Met: independent evidence shows it happened. Moved: the date or target changed, with the new one stated. Dropped: it was abandoned or clearly missed with no new date. Unchecked: you cannot check it yet, or the only evidence is the speaker's own word. Apply them like this: the EU high-risk dates are Moved (to 2 December 2027, with the product-embedded date moved to 2 August 2028). The Optimus 2025 target is Dropped: Electrek says Tesla missed it badly, and Musk gave no new number, only that production will be extremely slow at first. Waymo's 1 million target is Unchecked until the end of 2026. The AI 2027 authors' timeline is Moved, to around 2030. OpenAI's research intern is Unchecked: it is self-reported with no independent check cited. Write a one-line reason in each row. Use AI only to check that your reasons follow the rules.

Try this prompt
Here is my ledger and my chosen statuses. The rules: Met = independent evidence shows it happened. Moved = the date or target changed and a new one is stated. Dropped = abandoned or clearly missed with no new date. Unchecked = cannot be checked yet, or the only evidence is the speaker's own claim. For each row, say whether my status follows the rules, using only the evidence I wrote in the row. Do not search. Do not change my status. If a reason is missing, ask me for it. Never mark anything Met on the speaker's word alone.

LEDGER:
[PASTE LEDGER WITH YOUR STATUS AND ONE-LINE REASON]

Example / expected result

Illustrative example; not a recorded model result.

Row 5, OpenAI research intern: you marked Unchecked. Your evidence says the claim comes from the keynote and the article cites no independent check. This follows the rules. Row 3, Waymo: you marked Unchecked. The deadline is end of 2026, so this follows the rules.

Check your result

No row is marked Met on a speaker's word alone, and every status has a one-line reason you wrote.

3

Review once a month and note how dates moved

Once a month, open the ledger and update only rows where you found new evidence, with the link and date. Add a short note for every Moved row: was the change announced openly, like a law with a formal new date, or did a date quietly disappear? A moved date can be sensible. The EU delay came with new, stated dates. Over time, you will see which sources keep score on themselves. Add one new row per month and remove none. If a row stays Unchecked for six months, say so in the note.

Try this prompt
Below is last month's ledger and a list of headlines I read this month, each with a link and date. For each ledger row, say whether any headline relates to it and quote the words that do. Do not decide the status. Do not search. If no headline relates, write "no new evidence". Then ask me which rows I want to update.

LEDGER:
[PASTE LEDGER]

HEADLINES:
[PASTE HEADLINES WITH LINK AND DATE]

Example / expected result

Illustrative example; not a recorded model result.

Row 3, Waymo 1 million rides per week: headline 2 says "Waymo reports a new weekly rides figure" with the date. Do you want to update the evidence? Row 1, EU high-risk dates: no new evidence.

Check your result

You changed a status only where you added a new link and date, and you added a note on how each moved date was announced.

When it goes wrong

A ledger is not a prediction. A moved date is not always a failure, and a met target is not always good news. Do not let a chat tool decide what is true: it can invent sources and dates, so use it only to format and to check your reasons. Do not use one speaker's own report as proof. Keep the ledger about claims, not about people.

Make it your own

Add one claim you saw this week, with its source and date, and give it a status using the four rules. If you cannot, mark it Unchecked and say why.

Useful next resources

Sources and review

Last checked Nov 2, 2026. Features and account limits can change. Verify important results against your original material.

Did this help you complete your task?

Tell us what needs improving. No account or private output required.

Was this useful?
Report incorrect information or a broken link

Genius Forges Editorial

Practical exercises based on checked sources. Examples are illustrations, not recorded model results.

Keep building your AI habits

Keep learning useful AI tricks

Get practical lessons and resources when we publish them.

Turn this idea into a daily skill

Practice with a small input and check the result before sharing.

We use optional cookies for analytics and referral tracking. Learn more about our privacy policy