Skip to main content
Public scorecard

We grade our own predictions

Every win probability Grantverse publishes is logged the moment it is issued, graded when the real funding decision lands, and reported on this page. Wins and losses alike.

Track record · In progress

Building our track record, in public

This page never shows a number we cannot defend. Until enough real funding decisions have come back, it stays empty rather than estimated. Here is the commitment running underneath it:

  1. 01

    Logged at issue time

    Every win probability is recorded the moment we publish it. No retroactive edits.

  2. 02

    Graded on real outcomes

    When the funder decides, the prediction is scored against what actually happened. Win or lose.

  3. 03

    Published here, unedited

    Calibration results land on this page automatically, whether they flatter us or not.

Predicted probabilityActual win rate
Illustrative · Not data

An example of the calibration view this page will show. Dots near the diagonal mean our percentages match reality. Nothing real is plotted until graded outcomes exist.

The intelligence behind each prediction is built on 160,000+ analyzed funding awards. This page tracks something stricter: how our own published predictions perform against real decisions.

Accountability

Why we publish this

Funding platforms love a confident percentage. A 92% match. A “great fit” badge. Almost none of them ever go back and check whether those numbers meant anything, and a prediction nobody audits is just decoration.

We treat our numbers as commitments. When Grantverse says 60%, we expect roughly six wins in every ten of those calls. Holding ourselves to that standard in public is the only honest way to ask you to plan your raise around it.

Principles

How we keep the numbers honest

Bands, not fake precision

A figure like 47.3% implies a certainty nobody has. Where the evidence supports a range, we show you the range, and we round to what we can actually stand behind.

Calibrated against reality

Our probabilities are tuned so the number means what it says. Across many predictions, 30% calls should win about 3 in 10 times and 70% calls about 7 in 10.

Outcomes feed back

Every graded outcome becomes a correction. When reality disagrees with us, reality wins, and future predictions move toward it.

Methodology

How we measure ourselves

We use the same standard statistical yardsticks professional forecasters are judged by. Nothing exotic, nothing self-invented.

Calibration

We group predictions into probability bands and compare each band's predicted rate to its real win rate. A band that predicted 60% and won 6 of 10 lands on the diagonal. Distance from the diagonal is exactly how wrong we were.

Brier-style scoring

Each prediction is scored with a Brier-style measure, the standard test used in weather and election forecasting. It rewards being right and appropriately confident. 0 is perfect, and anything under 0.25 beats coin-flip guessing.

Overall accuracy

The share of individual predictions that landed within 20 points of the real outcome. A blunt measure on purpose: easy to read, hard to dress up.

What a 60% prediction actually means

Not a promise, and not a vibe. It means that of every ten opportunities we score at 60%, about six should end in a win. If they do not, this page will show it, and we will fix the model, not the chart.