Most market AI sounds certain. Ours keeps score.

Brier Labs reads the market before the opening bell, then grades every one of its own calls against real closing prices. Confidence is easy to generate. We publish the record instead.

Prediction record Scored at 1 trading day
DELL Hold · earnings caution outweighed backlog strength Correct
TTWO Buy · premium pre-order mix signalled pricing power Correct
GOOGL Sell · regulatory and cash-flow pressure Missed
BLK Buy · ETF inflows and institutional backing Correct
Every call is stored with the headlines that produced it, then re-scored at 1, 7 and 30 trading days.
3 sources
Headlines cross-checked across independent feeds before the model sees anything, so one bad article can't carry a call.
1, 7, 30
Trading-day horizons. A signal that only works on day one is a different product from one that holds for a month.
Two baselines
Every result is set against always-buy and always-hold. An accuracy figure without a baseline means nothing.

From scattered headlines to a call you can audit

The pipeline runs unattended every weekday. Nothing is hand-picked, and nothing is quietly discarded when it turns out wrong.

Step one

Gather

Independent news feeds are pulled for every position, deduplicated, and consolidated into a single evidence set per company.

Step two

Judge

A language model weighs the full set at once and returns a structured call with a sentiment score and a written rationale.

Step three

Score

After the close, real prices come back in. Each past call is marked correct or missed, and the record updates itself.

Named after a measurement, not a metaphor

The Brier score is the standard way to measure how accurate a probabilistic forecast turned out to be. We took the name because grading the forecast is the harder half of the work, and the half most tools skip.

The rationale is stored, not just the verdict

Every call keeps the exact headlines behind it. When a call fails, the reason it failed is still on file.

Baselines run alongside

In a rising market, always-buy scores well. We report it next to our own number so the comparison is unavoidable.

Horizons are measured separately

Same-day accuracy and one-month accuracy answer different questions. We don't blend them into one flattering figure.

The record is cumulative

Nothing resets. The history grows and stays queryable, which is what makes the system able to learn from its own misses.

Built for people who carry the cost of being wrong

Market sentiment moves more than share prices. It moves input costs, freight, and the pricing decisions that follow.

Pricing teams

See the macro pressure building on your inputs before it reaches a quarterly review, with the reasoning attached.

Supply chain

Track the sectors your suppliers sit in, and get told when the tone around them turns — not a week after it did.

Finance and strategy

A running record of what the market was saying, when, and whether it turned out to be right.

Early access opens with the first published scorecard.

We're accumulating the record now. When there's enough of it to mean something, it gets published — good or bad — and early users see it first.