How an edition is made, and how we check it
Pulse is written by an AI model. That is exactly why every claim carries its source, every reading is labelled as ours, and every forecast is scored in public.
1. Collection
For each edition we run seven separate searches: economy and finance, crypto, technology and AI, politics and policy, conflicts, disasters and climate, health and science. Each search returns recent articles from news organisations and primary sources (central banks, regulators, statistical agencies). Prices come from CoinGecko cross-checked against Binance; the macro calendar from FRED; the crypto Fear & Greed Index from Alternative.me.
2. Writing and verification
The model writes the edition from that evidence only. Before anything is published, automated gates check that:
- every key event links to a source that exists and was published recently;
- each sourced claim is supported by its source (a semantic check, not just a link check);
- no price appears that was not measured by us at publication time;
- the summary does not contradict the market data. When it does, the edition says so at the top.
An edition that fails a hard gate is not published. A category whose evidence is too thin is dropped rather than padded.
3. Facts and readings
Inside an edition each claim is tagged Fact (attributed, linked) or Reading (our interpretation). Sector calls, the market regime and the Radar are readings. Treat them as hypotheses, not as statements of fact.
4. Sources policy
- We prefer primary documents and established newsrooms; press releases are cited as the issuer's claim.
- Every source is linked as published. We do not reproduce articles; we summarise and link.
- Photographs are illustrative stock images (Unsplash) and are captioned as such. They never depict the reported event.
5. The track record
Radar scenarios name the signal that would confirm them, the one that would disprove them, and a deadline. After the deadline a price check settles what can be settled. Everything else is shown as could not check. We hide the hit rate until at least five scenarios have a real verdict, because a percentage of two outcomes is noise. See the track record.
6. About the AI
Editions are generated by xAI's Grok models with live web search. The model can be wrong, and sources can be wrong. The checks above reduce errors; they do not remove them. If a decision matters to you, read the source.
7. Corrections
Editions are kept as published, so the record stays honest. If we find a factual error we add a correction note to the edition stating what was wrong and when it was fixed. Report an error to hello@getpulsebrief.com with the edition link.
8. Feedback
Tell us what is missing, unclear or wrong: hello@getpulsebrief.com. We read everything.
9. Not financial advice
Pulse is information, not advice. It does not know your circumstances and does not recommend buying or selling any asset. Markets are risky; past readings say nothing certain about the future.
Questions
Who writes Pulse?
An AI model (xAI Grok) writes each edition from sources collected for that edition. Automated gates check that every key event links to a real, recent source, that claims match their source, and that the summary does not contradict the data. No human edits individual editions.
How often is Pulse published?
Twice a day: a morning edition around 21:30 UTC for the next day and an evening edition around 09:00 UTC, plus a weekly review on Sunday evening (UTC).
What is the difference between a fact and a reading?
A fact is a statement attributed to a named source, with a link. A reading is our interpretation of what the facts mean for markets. Pages label each claim so you can tell which is which.
Is Pulse financial advice?
No. Pulse explains what happened and what could matter for markets. It does not know your situation and does not tell anyone to buy or sell anything.
How do you score your forecasts?
Each Radar scenario names a confirming signal, a disproving signal and a deadline. When the deadline passes, prices are checked automatically. If nothing observable settled it, the scenario is shown as ‘could not check’ — never counted as a hit or a miss.
Can AI tools read Pulse?
Yes. Every edition has a Markdown version, the site publishes llms.txt and an RSS feed, and structured data on every page. Please cite the edition URL.