Critique Rubric

Why a rubric #

"Looks cool" isn't feedback, and neither is "the hands are weird." A shared rubric does three things:

1. Consistency. Judges, peers and creators score against the same descriptions.
2. Teaching. Every score names a craft skill and what the next level looks like.
3. Process over prestige. The rubric rewards intent, craft and honest documentation, not only polish or compute budget.

Each challenge page lists the categories that challenge is scored on. This rubric is the long form: what each score means at each level, and how to write feedback around it. Use it for votes, peer feedback, longer critiques and self-review.

1. The scorecard (100 points) #

Open (non-weekly) challenges: when a challenge has no single craft objective, reassign A's 35 points as Concept & intent 15 + Visual language 20, using the subscales in §3.

Scoring a challenge with its own categories? Use the matching descriptors below: Story → B, Technical Execution → C, Tool Use / Creativity → the Concept & intent and AI craft subscales in §3.

2. Level descriptors #

Each category is scored at one of four levels. Judges pick the level first, then a number inside its range.

A. Craft objective (35) #

Example (a lighting challenge): a 4 changes the genre with lighting alone while keeping everything else consistent. A 2 changes the light but also changes the character, set and camera, so the lesson is lost.

B. Storytelling (25) #

Check: the hook (first 3 seconds), clarity of goal and stakes, the turn, the payoff, emotional tone, and whether it's "show, don't tell."

C. Technical execution (20) #

Use the AI Artifact Field Guide for shared names (drift, flicker, morphing, mouth mush, physics errors). Score by impact, not count: one glaring face morph in the climax costs more than three tiny background flickers.

D. Breakdown & BTS (15) #

Honesty rule: an honest breakdown of a modest piece outscores a vague breakdown of a polished one. Admitting cherry-picking ratios, failed approaches and manual fixes is rewarded, not penalized.

E. Carry-forward (5) #

3. Optional subscales (open challenges and deeper critique) #

Use these for open challenges, for "Best X" micro-awards, or when a creator asks for detailed feedback. Score each 1–4 using the level descriptors.

These map naturally onto micro-awards: Best Story, Best Cinematography, Best Sound, Best World, Best Breakdown, Most Creative Tool Use, Most Improved.

4. Writing feedback #

The format: Notice, Effect, Next #

Every piece of feedback should have three parts:

1. Notice: what specifically you saw or heard, with a timestamp.
2. Effect: what it did to you as a viewer.
3. Next: one concrete thing to try.

Weak: "The middle is kind of slow." Strong: "From 0:22–0:34 we hold on the wide for 12 seconds with no new information (notice), and I started checking the timeline (effect). Try cutting to her hands on the letter at 0:26 — it gives us the detail and halves the hold (next)."

Structure for a full critique #

1. What's working (2–3 specifics, with timestamps)
2. The one biggest opportunity, tied to a rubric category
3. Two or three smaller notes (Notice, Effect, Next)
4. A question for the creator ("Was the color shift at 0:40 intentional?")

Guidelines #

• Critique the work, not the person. "This shot feels flat," not "you don't understand lighting."
• Name the craft term from the glossaries (eyeline, motivated light, J-cut), so feedback teaches.
• Prioritize. One clear main note beats ten scattered ones.
• Ask about intent before calling something a mistake. Stylized artifacts can be deliberate.
• Respect the brief. Don't mark down a 15-second piece for having no character arc.
• Don't judge tool choice by itself. A Kling piece and a ComfyUI piece are judged on the same rubric. Credit hard-won control in category D, not by penalizing hosted tools.
• Separate taste from craft. "I personally prefer warmer palettes" is fine to say, but label it as taste.
• Keep feedback encouraging and constructive in public. Raise conduct problems with the moderators privately, not in a critique.

Feedback phrase bank #

5. Judging and voting #

1. Watch once without scoring. Note your gut reaction.
2. Read the breakdown, then watch again with the rubric.
3. Pick a level for each category first, then a number in the range.
4. Write at least one Notice, Effect, Next for the entries you score highest, so the creator knows what worked.
5. Calibrate (judging panels). Before scoring, each judge scores the same two reference entries (one strong, one developing) and the panel discusses any gaps over 10 points.
6. Conflicts of interest: skip entries from your teammates or collaborators.

6. The basics (not scored) #

Every entry should meet these before it's scored. The challenge brief and the Challenge Notice have the specifics.

• Within the runtime and format specified by the brief
• AI disclosure included (tools, and the approximate AI vs. hand-made split)
• Rights confirmed for music, voice and likeness. No real-person likeness without consent. No copyrighted characters.
• Content within platform rules
• Submission package complete (see Process Breakdown Template)

7. Self-review checklist #

Run this before you submit:

• Hook: would a stranger keep watching after 3 seconds?
• Story: can I say what happens in one sentence?
• Craft objective: could someone name the challenge's main skill from the film alone?
• Continuity: have I watched it once only for faces, wardrobe, props and screen direction?
• Artifacts: have I checked hands, teeth, eyes, text and edges at full screen?
• Sound: have I listened once with headphones and once on phone speakers?
• Grade: do shots from different sessions or models match?
• Pacing: is there any shot I'd cut if it cost me $10?
• Breakdown: could another member recreate my best shot from what I wrote?
• Disclosure and rights: done?

Score yourself with the rubric. If your self-score and the scores you get back differ by more than 15 points, that gap is the most useful feedback you'll get.