How You’re Scored
This is the whole rubric. Every rung is written down before you write a word, and the grader is given these exact sentences — there is no second, secret version of this page.
A rung you have not reached is information, not a rebuke. The useful question while reading is which sentence describes what you did last time.
Read one rung at a time
View all rubricThe comic ideathe game
There is no single comic idea. The lines are talk — banter, sarcasm, or attitude with nothing underneath — or several unrelated odd things compete and none is developed. Nothing here could be pointed at and called the joke.
Half-hour comedy
The comic ideathe game
- 1There is no single comic idea. The lines are talk — banter, sarcasm, or attitude with nothing underneath — or several unrelated odd things compete and none is developed. Nothing here could be pointed at and called the joke.
- 2There is an identifiable comic idea, but the scene cancels it: a second strange element competes for the same attention, so neither is the thing the scene is about. One cut would rescue it.
- 3One unusual thing is identifiable and the rest of the scene is played straight around it. But it is the obvious reading of the premise: the idea the premise handed the writer rather than one found inside it.
- 4The idea is one the writer found rather than one the premise handed over, but it is asserted instead of demonstrated — a character names the strange thing rather than the scene showing it happen.
- 5One unusual thing, precisely chosen, with everything around it played completely real. The idea is not the premise restated — it is something true inside the premise that was not visible until this writer pointed at it.
Escalationheightening
- 1The second beat restates the first. A pattern repeats where a game should have grown, or the scene stops moving altogether and the situation at the end is the situation at the start.
- 2The scene moves, but sideways. The second beat introduces a new element rather than raising the first — a different thing happening, not the same thing getting worse.
- 3The scene moves up. The second beat is bigger than the first and follows from it, but the step is the expected one and the reader can feel it coming.
- 4The move is bigger and could not have been predicted, but it steps outside the logic the first beat established — surprising without being the consequence of what came before.
- 5The move raises the idea and stays inside its logic — it is the consequence of the first beat rather than a louder version of it. It reads as surprising in the moment and, in retrospect, as the only place the scene could have gone.
The last linethe button
- 1The scene ends on setup, trails past its own best line, or closes on the weakest thing on the page. There is nothing here to cut away from.
- 2The scene ends on a joke rather than on setup, but the laugh is diluted on the way out: the punch word is followed by explanation, a tag, or a reaction line that says what just happened.
- 3The last line is the funniest line and the scene stops there. The laugh is real, though the punch word sits buried mid-sentence or the ending is the one the setup already promised.
- 4The last line is the strongest, the laugh word is last, and it pays off a plant — but the plant is the one the setup announced rather than one the scene buried.
- 5The last line is the strongest, the laugh word is the last word, and the scene cuts on it. It pays off something the scene planted rather than arriving from outside, and it does not explain itself on the way out.
Character voice
- 1Cover the character names and the lines are interchangeable. The dialogue is generic — no detail, reference, or rhythm that belongs to one person rather than another.
- 2One character is drawn and the other is furniture. The lines could be reassigned in one direction only: cover the names and half the scene is still attributable, half could be anyone.
- 3The speakers are tellable apart and at least one line carries a detail that is genuinely this character's. The distinctions are consistent but broad — a register or an attitude rather than a person.
- 4Every line sits in its own mouth, but the distinctness is in what the characters talk about rather than how they speak. Swap the topics between them and the voices would swap too.
- 5Every line sits in one mouth and could not be moved to another. The comedy comes from who is speaking as much as from what is said: the same words in the other character's mouth would not be funny.
Compressioneconomy
- 1The scene explains itself. Characters state what they feel or narrate what the reader can already see, the opening spends lines arriving at the situation, and at least one idea is said twice.
- 2It gets in late and does not over-explain, but one line is doing a job another line already did — the same beat is paid for twice, and deleting either would lose nothing.
- 3Tight enough. It gets in reasonably late and a few lines could go without loss, but something is stated where it could have been implied.
- 4Nothing is removable and nothing is over-explained, but each line does a single job. The scene is efficient rather than compressed: no line is advancing the situation and landing a laugh at the same time.
- 5Nothing is removable. It enters after the situation has already started, every line does at least two jobs — advancing, revealing, or landing — and nothing is explained that the reader can infer. Length is never mistaken for substance: a short scene that lands beats a longer one that enumerates.
How this rubric was chosen
Why these five
Selected from ten criteria that appeared across three or more INDEPENDENT craft sources, then filtered by what is actually assessable in a 3-5 line scene. game/heightening/button are the three stages of one comic object (idea, movement, payoff), which maps onto the incongruity-detection / resolution / payoff model in humor research — stage-based dimensions are orthogonal by construction, whereas five adjectives for 'good' collapse into a single score.
How 2 and 4 are written
1/3/5 are three different questions, not three magnitudes: 1 = was it attempted at all; 3 = attempted, but is it the obvious version; 5 = non-obvious, and does it hold throughout. 2 and 4 are the boundary levels — the previous question answered yes, the next answered no, WITH THE REASON NAMED. A 2 is 'attempted but cancelled' and names exactly one defect, so it is diagnosable where a 1 has nothing to fix. A 4 is 'non-obvious but seamed' and names exactly one shortfall. No level is distinguished from its neighbour by a degree word.
What a three-to-five line scene can test
A 3-5 line scene supports ONE heighten plus a button. Improv teaching calls for three game moves minimum and five to seven ideally IN A TWO-MINUTE SKETCH; scoring 'did they heighten three times' at this length would punish correct writing. Deliberately excluded as unassessable here: rule of three, act structure, act-outs, runners, callbacks, A/B story weaving, character arc, scene turn, pacing variation, world-building, escalation ceiling.
What is deliberately not scored
'Does the cold open set up the A-story' is excluded because it is wrong, not merely unmeasurable: sitcom cold opens are conventionally disconnected from the episode. Marketability, theme, and emotional engagement are excluded because this grades craft, not commercial viability.
Why compression is scored at all
Style bias in LLM judges measures 0.76-0.92, exceeding position bias and making it the dominant failure mode: judges systematically prefer longer output. Comedy rewards compression, so an uncorrected judge is anti-correlated with the skill being taught. economy exists partly as the explicit counterweight, and its level-5 anchor names length-neutrality directly.