Latest News
Introducing the Brightroom Library.
Twenty interactive lessons live, scaling toward two hundred and twenty. Predict, manipulate, derive — the Library is opened.
Introducing the Brightroom Mock Exam: a faithful test-day rehearsal.
Full, Half, Section. 135 minutes, three sections, the 3-change budget, the break room. Faithful to test day, to the minute.
Most GMAT prep promises a 700. Almost none of it delivers.
A founder's letter on what serious preparation actually requires — and why most of the industry can't honestly offer it.
GMAT Focus 2026: what changed, what we updated.
GMAC quietly retuned Data Insights weighting. Our engine was already calibrated; here's the diff.
Section Analytics: now in private beta.
Per-section percentile, time-per-question drift, predicted band — refreshed every session.
The guarantee, written down in one place.
Improvement bands on every Pro tier. An absolute 715+ floor on every Ultra plan. Renewal as remedy. One document, one URL, one contract.
How our adaptive engine actually decides what to ask next.
Item Response Theory, mastery vectors, and why most 'adaptive' prep isn't.
What the first Ultra cohort taught us.
One hundred and fourteen of one hundred and sixteen who reported back cleared 715+. Two renewed at no charge. Three patterns separated the floor-hitters from the misses.
Year Two.
All five of last year's forecasts shipped; three slipped a quarter. What we got wrong, what the contract proved, what 2026 looks like.
A calculator that knows when to disappear.
Present on Data Insights. Absent on Quant. Same Tab keystroke; the right pane carries the section's rule. A faithfulness change, not a feature.
Section Mode is its own surface.
Adaptive item selection inside the section, sub-skill scope picker, configurable length. The conditioning surface the Mock alone never was.
Verbal: Critical Reasoning, first lessons.
Argument structure first; the six CR question types fall out as questions about the same object. Nine lessons in the second Library batch.
The table moved again. Less, this time.
0.4pp at the 705 mark, against last year's 0.8. The curve is stabilising. What we updated and what we did not.
First lessons. Number Properties.
Fourteen interactive lessons in the foundational topic — predict, reveal, synthesise across ~20 slides each. The first Library batch is live.
The scratchpad, inside the runner.
Press Tab from inside any Mock section: a side panel opens with free-form notes on the left and a five-line calculator on the right. Per-question state, no rendering latency.
Introducing Ultra Plan.
$1,599, six months, every Pro surface — plus an absolute 715+ score floor for every Ultra customer, no diagnostic threshold.
A first look at Section Analytics.
Three signals per section — predicted band, percentile, median time/Q. A fifty-candidate private alpha, instrumented to tell us which view drives behavior.
Roadmap is live. A study plan the engine writes, one day at a time.
Per-day session blocks — surface, length, focus, the one-line reason — rewritten every night against the candidate's mastery vector and target date.
Practice is live. The engine, on demand.
Ten, twenty, or forty-five minutes against the mastery vector. No scored band, no test-day frame — just adaptive sessions whenever a candidate has the seat.
Data Insights needed three dimensions, not one.
The mastery vector is a selection input as of last Thursday. Interpretation, inference, integration — three axes, learned loadings, a new routing rule.
Year One.
Twelve months. A diagnostic, three engines, three Pro tiers, the Mock Exam, a written guarantee. What shipped, what we got wrong, what 2025 looks like.
Pro now carries a written score-lift guarantee.
Diagnostic-banded lifts on every Pro tier: +110 down to +20 across the score range. Three eligibility checks. A free six-month renewal as the remedy.
Pro 6-month is live. $699, half a year, the engine.
Three tiers now — 1-month, 4-month, 6-month. Same product, three durations. Built because 28% of the June cohort needed it.
The Mock Exam is in public beta.
Three modes — Full, Half, Section — share one runner. 64 questions, 135 minutes, the three-change budget, the break room.
The percentile table moved. Here is the diff.
GMAC's August refresh compressed the top of the curve. 705 is no longer 99th. What changed, why, and what we updated.
GMAC dropped “Focus Edition.” The test is unchanged.
Sixteen months after launch, the qualifier is gone. The sections, the band, the percentile table, the items — all identical.
Pro 4-month is live. $599, 120 days, the v1 engine.
Two durations, one product. The engine, the diagnostic, the analytics — sized for an eight-week sprint or a four-month run.
The pipeline behind every item parameter.
Manual weekly re-fit → nightly rolling re-fit on thirty days of responses. Usable pool grew 19%; standard error on θ tightened by 0.04.
Pacing, finally, as a first-class signal.
Time-on-question was logged but never consumed. v1.1 routes on it. The back-third accuracy gap closed by nine points.
An engine that picks the next question, properly.
v1 replaces the rolling-window scalar with 3PL Item Response Theory. A 2,400-item pool, Fisher-information selection, honest stop conditions.
The legacy GMAT, retired.
Three weeks after Jan 31. Old scores still count for five years; new ones are Focus only. The concordance table is the only honest translation between them.
Introducing the Brightroom diagnostic: 30 questions, 25 minutes.
Free, adaptive, twenty-five minutes. Returns a predicted band, three section scores, and the single sub-skill to work on first.
An adaptive loop, before it has anything to teach.
Three weeks of private alpha. Thirty testers. A seven-line loop, a too-small item pool, and the calibration problem behind both.
GMAT Focus, day one.
The new test went live yesterday. The bookmark-and-review mechanic is the rule that will catch most candidates off guard.
A small note on what we are building, and why.
Four people in Zurich, no product yet, and a read of the new GMAT that we think is worth a company.