How Notes are made
Notes are the analysis I publish here. This page says where they come from, what gets in, and what each piece has to pass before it goes up.
- Written by
- Me, with AI tools
- Case study
- Clarity Today
- Cadence
- At most one a day
- Published
- None yet
- Daily Signal
- Retired 14 October 2026
How a Note is made
Five stages, in order. Nothing is generated when you load a page.
- 01
Sources
Candidates come from Clarity Today stories and Clarity’s edition Deep Dives. They are the source pack, not the piece. Each Note is written fresh from the primary documents behind them: the papers, filings, docs and pricing pages themselves.
- 02
Selection
A candidate has to pass all six tests below. Most days none does, and then nothing publishes.
- 03
Draft
I write with AI tools. Three drafts take different routes to the same question, and an editor ranks them blind. A red team then argues the best case against the winner, and the piece has to answer or concede every point. The style is Clarity’s, in the first person: plain verbs, short sentences, verbatim quotes.
- 04
Checks
Seven checks run on every draft. A piece that fails one does not publish.
- 05
Review
I read the first ten Notes before they go up, and I can take any Note down.
What gets in. What doesn’t.
Analytical and consequential, or nothing. A week with no Note is fine; a weak Note is not.
Passes all six
- Decision-relevant: a founder, CEO or CTO deciding on AI would change a plan, a budget, a hire or a design because of it.
- Mechanism, not event: it explains why, or how much, with numbers or a causal chain.
- Durable: it is still true and useful in 90 days.
- Non-obvious: the headline alone does not give away the conclusion.
- Well sourced: at least two independent publishers back it, and one primary document counts.
- On a pillar: AI product strategy and build vs buy, AI unit economics, engineering org design when agents write code, or the reliability and risk of agentic systems.
Stays out
- Amazon and AWS, including any story where Amazon is a party. I work there.
- Securities calls or trade ideas.
- Partisan politics.
- Medical advice.
- Breaking news under 24 hours old.
- Claims from a single source.
Seven checks before a Note publishes
- Fact guard
- Every number, name and date in the piece exists in a primary document, not only in a summary of one. Quotations are verbatim, in quotation marks and attributed.
- Citations
- Every factual sentence cites a primary source.
- No copied passages
- No run of 12 or more words from a source appears without quotation marks.
- Writing tells
- A detector checks the measurable habits of machine prose, and a critic reads for the rest. The critic must find none that block.
- Judge
- Two judges score the piece independently. Both must give it at least 8 out of 10 on analytical value and on consequence.
- Length
- 900 to 1,800 words, under a plain headline.
- Excluded topics
- None of the topics above appears, and no employer figures do either.
Who stands behind it
I write Notes with AI tools. Every piece is checked against its sources before it publishes, and I stand behind each one. When I say “I think”, that is my judgement; Notes never describe my day job.
Wrong is fixable. Silent is not.
If something here is wrong, I want to know: write to hi@promitb.dev. Corrections are made in the piece and logged below.
-
Retired the AI-written Daily Signal: its daily briefings, its topic hubs and its pillar summaries now answer 410 Gone. The underlying data is kept, not deleted.
-
Fixed a pipeline defect that rendered raw model JSON in place of prose on several briefing pages and share cards; added boundary validation so the failure is now loud, not silent.