The Taste Rubric

Taste is a skill with a rubric.

Ten criteria for scoring finished creative and design work, each with the way it fails and the anchors two people need to agree on a number. Written for creatives directing models and engineers designing things, because both now face the same problem.

Version 1.0 · August 2026

What is taste?

Taste is the ability to tell which of several competent options is the right one, and to say why. It is judgment about finished work, and it is learnable, because the criteria it runs on can be written down and applied until they are fast.

The reason it matters more now is arithmetic. Generation collapsed the cost of producing competent-looking output and did nothing to the cost of knowing which output is worth producing. The scarce skill moved from execution to judgment, and it moved for the engineer designing an interface at the same time as for the director cutting a sequence.

What follows is the instrument, not the argument. Score the work, read the low numbers, go back in.

How do you use the rubric?

Score a finished piece against all ten criteria, zero to three each, for a total out of thirty. Six rules keep the number honest.

  1. Score the work, not the maker or the process. How hard it was is not a criterion.
  2. Score alone before anyone talks. Two scorers who discuss first produce one score written down twice.
  3. Score the whole piece in one sitting, in the printed order.
  4. Write one sentence of evidence for every score below 2. A number with no evidence attached is a preference wearing a rubric.
  5. Treat a gap of more than one point between scorers as a definition problem, not a disagreement. Resolve it by pointing at the work, never by averaging.
  6. Rescore after the revision. A single score is weak information; the delta between two is the whole point.

The ten criteria

Each one carries a definition, the way it fails, and four anchors. The anchors are the working part: two scorers agree about anchors and argue about adjectives.

01 Intent legibility

A stranger can say what the work is for after ten seconds with it.

Fails when the work is impressive and unreadable, and the maker has to explain it first.

  1. 0 No stated or inferable purpose.
  2. 1 Purpose emerges only after explanation.
  3. 2 Purpose is clear on a careful read.
  4. 3 Purpose lands in the first seconds and shapes everything after it.

02 Restraint

Everything present earns its place, and nothing is there because the tool made it easy.

Fails when every effect, transition, gradient, or model capability appears exactly once.

  1. 0 Feature tour. The work is a list of what the tool can do.
  2. 1 Two or three effects present with no reason to be.
  3. 2 Mostly deliberate, with a decorative habit or two left in.
  4. 3 Removing any element would cost something specific.

03 Coherence

One set of decisions governs the whole surface: palette, type, grade, motion, pacing, tone.

Fails when three good ideas are at war and each was right somewhere else.

  1. 0 Reads as assembled from unrelated sources.
  2. 1 A system exists and is abandoned in places.
  3. 2 Consistent, with a section that belongs to another piece.
  4. 3 Every part could be identified as belonging to this work alone.

04 Reference literacy

The work knows what it descends from, and departs from it on purpose.

Fails when the result is accidental pastiche, or the model's house style left exactly as it arrived.

  1. 0 Default output of the tool, unexamined.
  2. 1 Imitates one obvious reference without knowing it.
  3. 2 References are chosen and named, departures are timid.
  4. 3 The lineage is legible and the departure is the point.

05 Detail under scrutiny

The work holds at full resolution and at half speed: edges, spacing, hands, contact, seams, kerning.

Fails when it passes as a thumbnail and collapses as a frame.

  1. 0 Breaks at ordinary viewing size.
  2. 1 Survives the thumbnail, fails the full view.
  3. 2 Holds at full size, with known soft spots.
  4. 3 Rewards inspection. Detail was decided, not survived.

06 Rhythm

Time and space are shaped: cut lengths, spacing, easing, density, and where emphasis falls.

Fails when everything is given equal weight, so nothing is emphasised.

  1. 0 Uniform. No pacing decisions are visible.
  2. 1 Rhythm attempted, applied unevenly.
  3. 2 Clear pacing with one dead stretch.
  4. 3 Attention is directed. You could not reorder it without loss.

07 Material honesty

The medium is used as itself rather than as an imitation of an older one.

Fails when generated work pretends to be photographed, or an interface apologises for being software.

  1. 0 An imitation, and a poor one.
  2. 1 Imitation that mostly holds, with tells.
  3. 2 Honest about its medium, not yet using it.
  4. 3 Does something only this medium can do.

08 Evidence of cutting

Things were removed. The piece is visibly shorter and tighter than what was made.

Fails when everything produced was kept, because producing it was cheap.

  1. 0 Everything made is present.
  2. 1 Obvious weak material survives.
  3. 2 Cut well, with one indulgence left.
  4. 3 Nothing left to remove without damage.

09 Defensibility

The maker can name why each major choice beats the alternative they actually tried.

Fails at the sentence "the model gave me that."

  1. 0 No alternatives were tried.
  2. 1 Alternatives tried, chosen on feel alone.
  3. 2 Major choices defended, minor ones inherited.
  4. 3 Every significant choice has a rejected alternative behind it.

10 Effect on a stranger

Someone with no context has a specific reaction, and a second stranger has the same one.

Fails when the only people who respond are people who know the maker.

  1. 0 No reaction beyond politeness.
  2. 1 Reaction is generic approval.
  3. 2 A specific reaction, not yet repeatable.
  4. 3 Two strangers describe the same effect in their own words.

What do the scores mean?

Four bands out of thirty. The band matters less than which criteria produced it, so read the low numbers before the total.

What is this rubric not?

It is a scoring instrument for finished work, and it is wrong to use it as anything else. Five limits.

  • Not a style guide. It never tells you what to make, only how the thing you made holds up.
  • Not a checklist for work in progress. Applied while making, it produces safe work that scores 20 and moves nobody.
  • Not a measure of a person. One low score is one piece, on one day.
  • Not usable on a fragment. Score a whole piece; a single shot or screen has no rhythm to read.
  • Not a substitute for a room. The rubric makes a critique specific. It does not replace people who will tell you the truth.

How do you cite it?

Freely, with the version. The rubric is published so it can be used, argued with, and forked by anyone teaching this work.

The Taste Rubric, version 1.0.
The Multimodal Society, August 2026.
https://multimodalsociety.com/taste-rubric

Version 1.0 is in force until it is replaced here. A change to any criterion increments the version and is listed on this page; the capstone of the Multimodal Credential is scored against the version in force when the cohort begins.

Learn to use it in a room.

Membership is free. Masterclasses taught by industry leaders, hackathons where you finish something the same day, and screening nights where the room is honest.