The Brightroom diagnostic is live at /register. It is free. It runs in twenty-five minutes. It returns a predicted score band, not a single number, because a single number on thirty items is a lie.
This is the first public surface of the product we have been writing about since September. The engine that drives it is the v0 we wrote up in December, with three more months of alpha responses behind it. The selection logic is still simple. The output is honest about that.
What it does
Thirty items, selected one at a time from our calibrated pool. Each section of the GMAT Focus Edition gets proportional representation — ten Quant, ten Verbal, ten Data Insights — with the engine pulling harder items as the candidate's ability estimate climbs and easier ones as it falls. Time-on-task is logged per item; the candidate sees a per-section clock but no global timer.
At the end, four things appear on one screen: the predicted band, the three section bands, the candidate's pacing relative to the two-minute target, and the single sub-skill the engine wants them to work on first.
Why a band, not a number
Thirty items is not enough data to defend a single scalar. The interval the engine reports is roughly the eightieth-percentile confidence band around the maximum-likelihood estimate — the range we would expect the candidate's true ability to fall into if they sat the diagnostic again next week with a different thirty items.
On the full Mock Exam — which ships later this year — the same maths returns a tighter band, because more items produce more information about the candidate. We will be explicit about the width every time. A point estimate without a confidence interval is a guess wearing a tie.
What we cut
The original spec was a ninety-item diagnostic. We dropped it. Sixty-five minutes is the minimum responsible duration for that version, and most of the candidates we tested it on stopped at fifty. The signal we lost between thirty and ninety items is real, but smaller than the signal we lose when a candidate abandons mid-session and we report on a half-finished sample.
Pacing recommendations are not in this release. The engine has the data — every item time-stamped, every section clocked — but the model that turns pacing into actionable guidance is not yet calibrated against our small alpha pool. It is in the queue for the Q2 release.
A study plan is also not in this release. The diagnostic ends with a single sub-skill recommendation, not a roadmap. The roadmap surface — the thing that decides what to do tomorrow morning — needs more than thirty items of evidence to draw against. We will not ship a study plan the engine cannot defend.
Who gets it
Anyone who registers. There is no waitlist. The diagnostic is the front door of the product, and we want it to be the front door of a serious candidate's decision about how to prepare. Free, today, in twenty-five minutes.
Candidates can retake it every thirty days. The engine treats each attempt as a fresh sample for calibration; the ability estimate shifts with new evidence but does not forget the older sessions.
What comes next
A paid tier — Pro 1-month — opens next month with full practice access against the calibrated pool. The Mock Exam, with the test-day timing and the three-change budget, is the milestone after that. The Library, the Roadmap, and Section Analytics are further out.
The diagnostic is the first surface. It will not be the last.
— Brightroom Product