The Confirmation Dividend

The finding

A thousand headlines cross a screen on a normal day, and almost none of them move a market anyone cares about. The useful question is not what happened, which every feed will tell you, but which of these a market will actually react to, known early enough to matter. Vera reads incoming news and sorts it in advance, before the price moves, into stories worth watching and stories to set aside.

The gap between the two is wide and consistent. On markets sitting still, with no move already underway to take the credit, the news Vera flagged went on to move a price 2 cents or more about 23% of the time. The news it set aside did so about 8% of the time, roughly a third as often. Across every market in the window, not just the quiet ones, the flagged rate is about 35% against 13% for set-aside news, a gap of about 21 points.

The more sure the tool, the more often the market moved.

The Confirmation Dividend · Vera Research

The test

The study covers more than 200,000 news-to-market pairings from April 22 to June 30, 2026, matched from tens of thousands of stories to more than 3,000 prediction markets. For each pairing the group was fixed in advance, from what Vera read the moment the news arrived, and never from the move that followed.

That last point is the whole game, so we checked it against the data directly. The label that defines the two groups reconstructs, to better than 99.9%, from Vera’s pre-news scores alone, and the table that stores it holds no price-outcome field at all. This is not a case of grading the exam after seeing the answers.

It tracks how sure the tool was

The flag is not a single yes or no. Vera also records how sure it is, and the reaction rate follows that confidence in the right direction. Among the stories it marked worth watching, about 33% went on to move a market 2 cents or more. Among the stories it was most confident in, about 43% did. A signal that strengthens exactly where the tool says it should is harder to dismiss as an accident than a single average is.

Not just the obvious markets

The easy objection is that the sort only re-discovers something mechanical. Prices near their edges cannot move much, some categories are twitchier than others, and a market already drifting will keep drifting whatever the news says. All true, and all tested. Holding category, how extreme the odds were, and the market’s state of motion constant at once, the gap narrows, as it should, but it does not close: flagged news still moved markets about 9 percentage points more often than set-aside news in the same conditions, several times the size of the effect we said in advance would count as real.

What it does not tell you

Here is the part easy to leave out, so we are putting it in the middle rather than burying it. The sort separates news by whether a market will react. It says nothing useful about when. Among the markets that did move, the flagged ones and the set-aside ones took about the same time to reach their peak, a little under an hour and a half either way (84 minutes against 86). The stronger flag did not buy a faster move, and we are not implying one.

The takeaway

1. On quiet markets, the honest comparison, flagged news moved a price 2 cents or more about 23% of the time against about 8% for set-aside news, roughly 2.7 times as often.
2. Across all markets the flagged rate is about 35% (34.76) against 13% for set-aside news, a gap of about 21 points.
3. The rate climbs with the tool’s own confidence: about 33% for the WATCH tier and about 43% for the highest-confidence TRADE tier.
4. Holding category, odds extremity and market motion constant, the gap narrows but survives at about 9 percentage points, several times the pre-declared floor.
5. The sort separates news by whether a market reacted, not when: flagged and set-aside movers peaked in about the same time (84 against 86 minutes).

Every figure is a reaction frequency, the share of pairings that moved a market 2 cents or more. Nothing here is a return, a direction, or an accuracy. The sort was fixed before the price moved, reconstructable to better than 99.9% from Vera’s pre-news scores alone.

Method and data

Built on the Vera dataset: more than 200,000 news-to-market pairings, April 22 to June 30, 2026, matched to more than 3,000 markets. A move is a repricing of 2 cents or more; every rate is a reaction frequency, not a return or an accuracy. Corrected figure: two independent data defects (float-subtraction residue, and a backfill that wrote nothing where it meant zero) were repaired. They push in opposite directions and partly cancel, and every claim holds.

Odds shown are each market’s own price at the time Vera flagged the story, stated as fact, not advice. Nothing here is a recommendation to buy, sell, or trade any market or asset. Vera and Crypto Briefing are not registered investment advisers.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.