From a headline to a case file, in seven steps.
Every story on AlgoThesis goes through the same process. Here it is in plain words, with the real numbers — the feeds, the merge that decides which report is the original, what earns a full read, where every price comes from, and how your own side is graded.
We read the wires
About fifty feeds, every few minutes: the markets desks of the major newsrooms, the two big press-release wires, every 8-K filed with the SEC, and the Federal Reserve's own releases. Thousands of articles a day come in; most are the same few hundred events told many times.
One event becomes one story — and the first report is the original
When several outlets report the same event, they are merged into one story. The earliest report is the original; the case file calls it that in the receipts. Later reports are “also covered”, and a market wrap or commentary piece written about the event is a “reaction”. A judge reads both texts and the companies they name before merging — a shared ticker or topic on its own never joins two stories.
So “which one is the original” is answered by the clock, and “how covered is it” by the count of outlets on the case file.
Triage: ignore, brief, or read
Every merged story is scored for importance from 0 to 1. Below 0.35 it is set aside. Between 0.35 and 0.60 it becomes an In brief line — headline, company, source, no read, nothing graded. At 0.60 and above it gets the full treatment. On a normal weekday that is a few dozen full reads and a few hundred briefs.
The read is written — and kept as written
For a full story, a model writes, in our own words: what happened, what it implied for the companies it names on that day, the case it holds, the case it breaks, and the one line that would prove the read wrong. It never reproduces the publisher's article; the link out is the receipt. An editorial gate approves the read before it publishes, and once published it is never rewritten — if we correct something, the correction is printed on the page, dated.
Names and prices
The companies a story names are verified against the market's own symbol list. Every price on the site is the official consolidated closing print — Databento's EQUS.SUMMARY — published the evening of the session it belongs to and labelled with that session's date. Nothing is intraday, nothing is a quote, and a story whose read names no company draws no basket of proxies.
Later coverage joins the arc
A story is not finished when it publishes. Later reporting joins it only when the company and the catalyst match — automatically when the evidence is strong, by an editor when it is not — and is labelled for what it does: supports the read, challenges it, develops it, or resolves it. The case file shows the arc as a timeline: published, then each development, then what would move it next.
Your side, graded
There is no house call. If you take a side — bull or bear — it locks at the tap. Ten trading days later it is graded against the official close: from the close on or after the day you tapped to the close ten trading days on. It is private to you, and a story with no primary U.S. stock is recorded as your opinion but never graded, because there is nothing to price.
What we deliberately don’t do
- No house direction, no win-rate, no leaderboard of people.
- No advice. Nothing here says buy, sell, or hold — it says what the news said, and what the stock did.
- No intraday prices. Every number is a dated end-of-day print, and says so.
- No reproduction of publishers' text. Summaries and reads are in our own words; the original is one click out.
- No rewriting. A read is kept as it was written; corrections are printed, not hidden.
Research, not advice. AlgoThesis is a publisher, not an adviser — the mission and the disclaimer say exactly what that means.