The one-minute brief (and what it is not)

Type a ticker into the search bar, open its symbol page and press Brief me. About half a minute later the page shows a one-minute read on the name: what the company is, the handful of numbers that matter, what changed recently, what the next results date looks like, whether this book could hold it, and a stance. In the sourced blocks each figure carries a small numbered chip that opens the source it rests on, and the full source list opens underneath; the stance always prints beside the platform's record on reads like it. The read is written by the model from numbers the platform already holds; it is logged before anyone knows how it turns out, and it is scored.

It lives on the symbol page, in the One-minute brief section under the price.

What is in a brief

What it is

One paragraph on the company and where the stock stands, from its filings and the platform's daily bars.

Numbers that matter

Four to ten rows: returns, trend state, the desk's signal score, valuation, growth. Values are the platform's own; the note beside each is the writer's.

What changed recently

Dated filings and headlines about this company, quoted as data with their source, never treated as an instruction.

Before the next results

The next results date and, inside ten trading days of it, the pre-announcement surprise forecast with its own published record; otherwise "no view" and why.

Tradability for this book

Whether a new position passes this book's entry checks, the typical spread for the name, and your position if you hold it.

The read

Bullish, neutral, bearish or no view against SPY over 5, 20 or 60 trading days, confidence 1 to 3, two facts, our record.

Our record on reads like this

How often reads like this one have been right, once 200 of them have matured; until then the count, and the words "too early".

The entry checks are the ones the Executor applies to any new position in this book: price floor, liquidity, spread and the earnings blackout. The desk is long only, so a bearish read is a reason not to hold a name, never a short. The read also says, in one line, what would change it.

Under the blocks: the full source list (every number and headline the writer could cite, with its date and where it came from), what the brief cost and how long it took, and the footer that says what the read is and is not.

Where the numbers come from

Nothing in a brief is looked up on the open web. A set of small readers run in parallel over what the platform already holds: completed daily bars and the quant algorithms' readings, the company's reported figures from its SEC filings, the earnings calendar, the research corpus (filings and headlines the Researcher has collected) and, in season, the surprise forecaster. Their output is one packet of facts, each with an id. The writer (Claude Opus) then writes the prose in a single call and may only cite ids that are in the packet; a sentence that cites something not in the packet, or quotes a figure that matches nothing it cites, is dropped whole before you see it (the claim goes with it, not just the chip), a value is always printed from the packet rather than from the writer's memory, the falsifier line ("what would change it") is the writer's own words, and a brief whose read rests on fewer than two checkable facts about the name is shown as no view. A read on the benchmark itself (SPY or a fund that tracks it) is always no view: it is measured against SPY, so a stance on SPY would mean nothing. Headlines and filing text are fenced as untrusted data on the way in and quoted, never obeyed.

That is also why a brief takes about half a minute and costs a few cents rather than minutes and dollars: the slow part, gathering, is code the platform runs anyway.

Why it never says buy or sell

A brief is a read, not advice, and the words are not cosmetic. The platform measured what happens when the model's directional opinions are traded: across its research program the model added discipline (exits, sizing hygiene, not chasing) and did not add stock selection over simple rules, and reads formed after a results release were in the price before anyone could act on them. So the brief states a stance and the facts under it, logs it, and lets the record say how much such stances have been worth. It does not tell you to do anything, the desk does not trade on it, and nothing on the page places an order.

The one forecast it does carry, the pre-announcement surprise forecast, is shown with its own published calibration (how well it has told beats from misses on several hundred past releases) so you can weigh it yourself.

How reads are scored

Every stance is written into an append-only, hash-chained ledger the moment the brief finishes, with the symbol, the horizon, the confidence and the facts it cited ("no view" is logged too, and counted, but never scored). A read is priced at the first close after it was written: a brief written Monday morning is measured from Monday's close, one written Monday evening from Tuesday's close, so a read can never be credited with a move that was already public when it was written. The time of record is the platform's own clock at the moment the row is written, not the moment the brief was asked for: a brief that starts a minute before the close and finishes a minute after it is measured from the NEXT close. Nothing about the outcome exists yet at that point, a logged read cannot be edited afterwards without breaking the chain, and the platform re-checks the chain, the timestamps and every sealed day each night. (This record is the one-minute brief's own; the Researcher's morning-brief calls on the track record page are a different thing, scored separately.)

After the close, a scorer looks for reads whose horizon has passed and scores each against what actually happened: did the name beat SPY over those trading days (bullish is right if it did, bearish if it did not; a neutral read is not marked right or wrong, it only enters the probability comparison as an even chance), and how far was the read's implied probability from the outcome (a Brier score, compared with a forecaster that has no view of the name at all and knows only the reads that had already been scored when this one was written). Every read is scored after 5, 20 and 60 trading days; the headline counts each read once, over the number of days it stated. A day the name did not trade is read from its next close, never an earlier one. The block on every brief prints the hit rate and that comparison once 200 reads have matured, with the interval around it; before that it prints the count. If the record turns out no better than the no-view forecaster after 400 reads, the stance block is retired and the brief says so.

Limits

• US-listed stocks and ETFs only; no crypto, no options.
• A book can ask for 20 briefs an hour and 60 a day; each one is charged to the book's AI budget and the page shows what it cost.
• The brief reads completed daily bars: during the session its "last close" is yesterday's, and the price at the top of the page is live. They are different numbers on purpose.
• Viewers can read every brief in a book; asking for a new one needs permission to spend the book's AI budget.