How to find the new fact in overnight stock news
A recycled headline can move a stock without adding a fact. Here is how to separate the two before the open.

Overnight stock news arrives as a pile, and most of the pile is the same event printed again. The useful work before the open is not reading every row. It is finding the sentence you could not have written at the previous close, and refusing to treat a louder copy of an old sentence as a new one. A recycled headline can still move the price. That is a different decision from changing what the business is worth.
A nonevent that opened at 85
Gur Huberman and Tomer Regev documented this split in the Journal of Finance in February 2001, in "Contagious Speculation and a Cure for Cancer: A Nonevent that Made Stock Prices Soar." On Sunday, 3 May 1998, the New York Times ran a prominent article on cancer research and mentioned EntreMed, which held a license tied to that work. On Monday the stock went from a Friday close of 12.063 to an open of 85, and it closed near 52. It closed above 30 for the three weeks after that. Other biotechnology stocks rose along with it.
The article did not contain a new result. The same research had been published in Nature, and covered in newspapers including the Times, in November 1997, more than five months earlier. Huberman and Regev's point is uncomfortable if your process is "a headline moved, so the facts changed." Attention changed. The clinical facts had not. They also note that later days, when genuinely new information arrived, produced a milder move in EntreMed and little of that spillover.
The practical reading is narrower than "the market is irrational, ignore the tape." You keep two notes. One says whether the forecast changed. The other says whether a story is being widely repeated. EntreMed needed both. A desk that merges them will update a model on a reprint, or will ignore a reprint that is about to dominate the open.
Three stacks before you change a forecast
Take the window from the previous cash close to now. For a Monday, that includes the weekend. Sort what you have into three stacks. The stacks are about the sentence, not about the outlet's reputation.
- New fact. A number, a date, a counterparty, or a commitment that was not in the last primary document. A change in guidance, a contract value, a recall count, the size of a financing, a regulatory decision. You can write a sentence today that you could not have written yesterday.
- Restatement. That same sentence, issued again. The company release, then a desk rewrite, then a headline that leads with the price. One event. Keep the earliest primary source. Keep a later copy only when it adds a fact the first one did not have, and write down only the addition.
- Attention with no new fact. A feature, a recap of a move that already happened, or a line whose only news is that the shares are up or down in premarket trading. The price may still move. The forecast does not, until a sentence from the first stack shows up.
The test fits in the margin. If you cannot write the new sentence, the row is not in the first stack, however many alerts it generated. "Shares rise premarket" fails the test. It describes a trade. Park it in the third stack and go find the document the trade is reacting to. Sometimes that document is real. Sometimes, as with EntreMed, the document is a reprint.
A law-firm notice that says the firm is investigating a drop is almost always the third stack. It is a solicitation to shareholders, not the company's disclosure of a new operating result. Note that it is circulating. Do not paste the law firm's verb into the forecast. The company's release, or the primary filing, is the document that can move the first stack.
The fact may be in a different company's release
The name on your sheet is not always the company that spoke. A supplier reports a contract, and the customer is the business whose volumes change. A customer warns on demand, and the supplier's shipments are what you have to revise. If you only open items where your company is the author, you will miss the sentence and then meet it later as a mystery move.
Read the release far enough to see who else is bound by the number. The ticker on the headline is who the item was matched to. The paragraph may name a party that has no ticker in that headline at all. That party still gets a note if the number changes their volumes, their costs, or their risk. This is the same spillover Huberman and Regev saw across biotechnology names, except here the link is a term in a contract rather than a wave of attention. Treat them separately. A contract term is a candidate for the first stack. A sector reprint of an old story is the third.
Read by release time, then write four lines
Homepage order is a bad order for this job. A rewrite can sit above the release that it copies. Read by the time the publisher released the item, earliest first, inside the window that starts at the previous close. The first primary source wins. Later rows have to earn their place by adding a fact.
For each company you actually cover, stop at four lines.
- The new sentence, or "none."
- The earliest source, and the time it was released.
- Whether any later copy added a fact. If it did, add only that clause to the sentence.
- Whether you are changing expected cash flows, or only noting that attention is high.
Line 4 is the EntreMed distinction. A permanent-looking price move is not proof that the cash flows changed. If line 1 is a real sentence, the next question is which input it touches, expected results or the risk of those results. That is a different read from this sort, and it is the subject of when a story changes what a stock is worth and of reading a story for the fact that belongs in a forecast. This check stops once the sentence is clean. It does not try to be the valuation.
Try it for 7 days. Signup starts Pro. A card is required. Real-time REST, WebSocket, and MCP, with 1,000 REST requests a day. Paid Pro removes the cap. After that it's $15/month, or Free if you stop.
Pull the window for the whole book
The sort above is something you can do by hand for a handful of names. It fails when the book is long and the open is close, because you will read whichever tab loaded, not the earliest source. A feed that returns the publisher's timestamp, the source, the type, and a link lets you walk the window in release order and apply the same three stacks.
For a Monday morning check, start at the previous cash close in New York, written in UTC. Friday, 31 October 2026, was still on daylight time, so 4:00 p.m. in New York was 20:00 UTC. The request asks for that instant forward, oldest first:
curl -sS "https://api.marketnewsapi.com/v1/news?tickers=AAPL,NVDA,TSLA&from=2026-10-31T20:00:00Z&sort=published_asc&limit=100" \
-H "Authorization: Bearer mna-live-..."Walk the rows in the order they come back. Group them by the sentence, not by the row. Keep the earliest source. The company release is the document to open when you need the number itself. A publisher story gives you the headline, a short summary, and the link. How those two differ is covered in the piece on press-release text. Law-firm notices are a third kind of row, covered in how legal notices show up in a feed. You do not need a new rule for them. They go in the third stack unless they quote a company fact you did not have.
published_at is when the publisher released the item, which is the clock this sort needs. It is not when the row reached you. A free key is delayed 15 minutes, so the last quarter hour before the open is missing on that key. A live key is not delayed. If you run the request again, drop any id you have already stacked. When you show a row to someone else, keep the source name and a working link. The item is the publisher's, not yours.
The request does not do the sort for you. It removes the excuse that you ran out of time before you found the original. The sentence in the margin is still the product of the check. Everything else in the pile is a copy, a price caption, or a reminder that attention and information are not the same event.