
Lectures · classrooms · assignments
A lecture theatre is a conference hall that comes back every week.
The engine that holds a keynote holds a first-year lecture without noticing. What education adds is the part after the room empties: assignments with a window, courses with a certificate, and a report that says what the class believes rather than who was wrong.
- 17 question types
- Live or self-paced
- Works with no internet
- Full reports, free
Three ways to set the same thing
Live, self-paced, or a course with a certificate.
One quiz, three deliveries. Nothing has to be rebuilt to move between them, and the differences between them are deliberate rather than incidental.

Assignments score on correctness, not speed
A live game scores on think time because the room shares a clock. Attempts hours apart share nothing, and a learner can put the kettle on mid-question, so timing them would produce a number that looks like the live score and measures something else.

Retries are a switch, deadlines are not capped
Let a learner have another go, or do not. Set a window as far out as the course needs. Neither is a decision the tool makes for you.

A correction reaches everybody
Editing a quiz updates an assignment already issued: the question is read at answer time, so a fixed typo reaches everyone who has not yet answered it.

Courses, narrower on purpose
An ordered list of assignments with sequential unlocking and one certificate across all of it. Documents and videos would mean a second content system for material a link in a slide already carries.

Missed questions come back on a schedule
A question got wrong on an assignment lands in that learner's practice deck. It is asked again a day later, then two days, then four, eight, sixteen and thirty-two: one step further out each time they answer it correctly, and straight back to a day whenever they do not. Leitner boxes, not a difficulty score we would have had to invent.
5,000in one room
Measured on one node against the real server over real sockets: 5,000 of 5,000 players joined with zero failures, every one received the question inside a 34ms window, and every answer was counted with the exact expected split. The same room was killed outright with a SIGKILL and came back with all 5,000 players in under two seconds.
After the bell
A report that says what they believe.
Counting who was wrong is a number. Naming what they were wrong about is a lesson plan, and that needs the chosen answer recorded rather than inferred.

Per-option distributions
Eleven people missed question four is a number. Eleven people all picked the same wrong option is a shared misconception with a name.

Wrong is not the same as silent
A learner who answered incorrectly and one who never answered are different rows, recorded as a fact at play time. Without that, a silent learner tops the needs-help list.

AI misconception analysis, on your key
Optional, off until you connect your own provider, and it is sent aggregates (question text, accuracy and the option distribution) rather than a class list.

Full report on every plan
Including the free one, and exportable as CSV at any size. Reporting is not the thing being sold here; room capacity is.
A slow connection cannot cost a learner marks.
Scoring credits the think time measured on the learner's own device, clamped by the connection's server-measured round trip and by any outage the server itself observed. Two equally fast thinkers score the same across a 25× latency gap, over real sockets.

The room in front of you
It has to work on the day.
Unglamorous, until the moment it is the only thing that matters.
- Host from a laptop over the local network. The phones join the laptop rather than the world, which is the answer for a school with filtered wifi or a trip with none.
- Every answer tile has its own silhouette, so a learner with a colour vision deficiency and a room whose projector has flattened the palette both still work, and everybody can say 'the triangle' out loud.
- The tile palette is checked for separation under simulated protanopia, deuteranopia and tritanopia, and the check fails the build.
- axe-core runs over every signed-out page in both light and dark on every build, at zero violations. WCAG 2.2 AA is the target we test against; no conformance level is claimed.
Stated plainly
What is not built.
On the same page as everything that is, because a gap somebody finds after the demo costs more than the gap itself.
- LMS integration and rosteringNo LTI, no grade passback, no roster sync. It is real, expensive, and worth building for a named institution that needs it rather than speculatively.
- Spaced repetition from live gamesThe practice deck is fed by self-paced work only. A live player is identified by a nickname typed on the day, and keying weeks of somebody's revision to that would hand one learner another's deck.
- A public content libraryDeliberately absent. Built for a single workspace it is an empty shelf, and an empty shelf is worse than no shelf.
Free plan, real room
Fifty people, no time limit, commercial use allowed.
The free plan is not a demo: it runs a real room of fifty with no cap on session length and no restriction on using it commercially. Larger rooms are a one-time pass, priced per event.