design-critique-and-review-facilitation

name	design-critique-and-review-facilitation
description	Use when facilitating a live or asynchronous design critique that needs a clear ask, safe feedback, a decider, and recorded outcomes. Unlike design-audit or heuristic evaluation, this runs the human review session rather than producing expert findings.
metadata	{"portable":true,"category":"00-cross-cutting-ops-qa-a11y","compatible_with":["claude-code","codex"]}

Design Critique & Review Facilitation

Acknowledgement: Shared by Peter Bamuhigire, techguypeter.com, [phone].

Use When

• You have to run the meeting, not just have an opinion — a live or async design critique / design review where several people will react to a concrete piece of work and a decision must come out the other end.
• Feedback in your reviews is unstructured, demoralising, or goes in circles — people argue preferences, the loudest voice wins, the author leaves bruised, and nothing is actually decided.
• You need a repeatable session protocol: how to frame the ask, which feedback framework to use, how to separate taste from requirement, who decides, and how the decisions get recorded.
• You are setting up an async review (PR-style comments, a Figma thread, a recorded walkthrough) and need the same rigour without everyone in a room at once.
• You want the critique to protect the maker (psychological safety) while still being honest — the two are not in tension when the session is run well.

Do Not Use When

• You want an expert evaluation of the artifact itself — a structured walkthrough against Nielsen/Tognazzini heuristics with per-finding severities → use heuristic-evaluation-and-design-critique. That skill produces the findings; this one runs the human session in which findings (and taste calls) are discussed and decided.
• You want a scored 0–100 quality audit across 10 dimensions → use design-audit. (Audit = one expert's graded scan; critique = a facilitated group session. They feed each other: an audit can be an input to a crit, but the crit is where humans decide what to do about it.)
• You need a ship go/no-go gate with spec/pixel parity and signed verdict → design-qa-and-pre-launch-review.
• There is nothing concrete to react to yet — only an idea in someone's head. Build the smallest real demo first; a critique with no artifact is the "imaginary-puppy" anti-pattern (doctrine/references/creative-selection-and-taste.md). Return when there is a demo.

Required Inputs

Input	Supplied by	Required?	Why
Review artefact and maturity	Maker	yes	Calibrates feedback depth
Decision, constraints, and open questions	Maker and sponsor	yes	Keeps critique on scope
Participants, roles, and decider	Organiser	yes	Establishes authority and safety

• A concrete artifact to react to — a working demo, screens, a prototype, a flow, or a redline. Never run a crit on a verbal description or a mock-of-a-mock (creative-selection rule 1: the working demo is the unit of progress). The closer to real, the better the feedback.
• The author's framed ask — what stage the work is at, what one or two questions the author wants answered, and what is explicitly off the table for this round (locked decisions, out-of-scope). An unframed crit invites scattershot feedback on things already decided.
• The decision model — who is the single decider for this work, and what is merely advisory. Reviews without a named decider end in mush (Kocienda: every review has one decider).
• Mode — live (synchronous) or async — and the roster (author, facilitator, reviewers; keep it small — 3–6 reviewers, not the whole org).
• Where decisions will be captured — the doc/ticket/thread that becomes the record.

Workflow

Run the session in seven moves. The full protocol (timings, scripts, framework choice) is in references/critique-protocol.md; an end-to-end worked session is in examples/critique-session-worked.md.

1. Set the frame (before any feedback). The facilitator states the goal of the session and the author states the ask: the stage, the 1–2 questions they want answered, the audience/context of the work, and what is locked / out of scope. Write the ask at the top of the capture doc. A tight frame is the single biggest lever on crit quality — it turns "what do you think?" (which yields taste noise) into "does this onboarding reduce drop-off for first-time users?" (which yields useful signal).
2. Present the artifact, not the rationale-as-defence. The author walks the concrete demo briefly, then stops defending and starts listening. Reviewers react to the artifact in front of them, not to an imagined one. (No artifact ⇒ no crit — send it back to be built.)
3. Choose and announce a feedback framework. Pick the one that fits the stage and say which it is, so everyone gives feedback in the same shape (references/critique-protocol.md §Frameworks):
   • I-like / I-wish / what-if — early/exploratory work; keeps it generative and safe.
   • Problem, not solution — name the problem you observed and leave the fix to the author ("users can't tell which field failed", not "make the border red"). Protects authorship.
   • Ladder of feedback (clarify → value → concerns → suggestions) — climbs from understanding to critique so concerns land after the work is understood, not before.
4. Separate taste from requirement — out loud, every time. Tag each piece of feedback as either a requirement (it violates an objective standard — a11y AA, a broken red-route, a stated brand rule, a measured perf budget: an algorithm, the arrow points one way) or a taste call (a heuristic with a real value but no objective arrow — a colour, a duration, a word). This is the algorithm-vs-heuristic lens from doctrine/references/creative-selection-and-taste.md. Requirements are non-negotiable and go straight to the decision log; taste calls are input to the decider, never settled by vote, volume, or seniority. ("I'd have used another font" is taste; "this label fails 4.5:1 contrast" is requirement. Saying which prevents taste being smuggled in as fact.)
5. Name the decider and decide — on the spot where possible. After the input, the single decider makes the calls. Taste decisions trace to a skilled human's conviction, not to a tally and not to an A/B metric (the "41 shades of blue" lesson — do not outsource a taste decision to a number). Decisiveness is the job; a crit that defers everything was theatre.
6. Capture decisions, not just comments. For every open question record one of: decided (what + who + why, one line), author's call (owner + by-when), needs a demo/spike (the variation to build before the next round), or parked (with the reason). Comments evaporate; the decision log is the deliverable. See references/critique-protocol.md §Capture.
7. Close the loop with "live on it" where it's a heuristic. Some calls can't be settled in one review — an animation's feel, whether the flow "adds up". Mark these to be dogfooded over time ("live on it"), not re-litigated by reopening the same argument (doctrine/references/creative-selection-and-taste.md rule 2). Schedule the next demo, not the next debate.

Async variant. The same seven moves apply, time-shifted: the frame + ask is written at the top of the thread; reviewers comment in the chosen framework's shape and tag each comment [requirement] / [taste]; the decider resolves each thread with a decision tag; the captured log is the thread's summary. Async trades immediacy for a written record and wider time zones — see references/critique-protocol.md §Live-vs-async for which to choose.

Anti-Patterns

• No artifact in the room. Critiquing a description or a mock-of-a-mock — "arguing about whose imaginary puppy is cuter." Build the smallest real demo first (doctrine/references/creative-selection-and-taste.md).
• Unframed "what do you think?" With no ask and no out-of-scope, the crit sprays opinions over settled decisions and wastes the room. Frame first, always.
• Taste smuggled in as requirement. "That's just wrong" about a colour/word/duration, stated as fact. Tag it as taste and route it to the decider; reserve "requirement" for objective standards.
• Decision by vote, volume, or seniority. Letting the loudest or most senior voice settle a taste call, or counting hands. Taste traces to the decider's conviction, not a tally (the "41 blues" error on the process side).
• Feedback aimed at the person, not the work. "You always do this" / praise-sandwich theatre. Critique the artifact; protect the maker (see Psychological-safety, references/critique-protocol.md).
• The author defends instead of listening. Turning the crit into a debate the author "wins" kills the signal. The author asks clarifying questions, captures, and decides later.
• Comments captured, decisions not. Leaving a pile of notes and no record of what was decided, by whom, and why — so the same argument reopens next week.
• Seagull feedback. A senior who swoops in, drops opinions, and leaves without engaging the context or owning a decision (Kocienda's Seagull Manager failure mode).
• Re-litigating a heuristic instead of living on it. Re-arguing a feel-call every meeting rather than dogfooding it over time and re-deciding from real use.

Outputs

Output	Consumer	Evidence / acceptance
Critique brief and agenda	Participants	Ask, stage, constraints, and protocol visible
Structured feedback register	Maker and decider	Observation, impact, question, and source captured
Decision and action log	Delivery team	Decision, rationale, owner, and due date recorded

• A run session (live or async) that produced decisions, not just comments.
• A decision log — each open question resolved as decided / author's-call / needs-a-demo / parked, with owner, one-line rationale, and (for taste calls) the decider named.
• A requirement list vs. taste-input list — the objective must-fixes separated from the conviction-led calls, so nothing objective is voted away and nothing subjective is faked as fact.
• A next-step: the next demo to build (not the next debate to have), and any "live-on-it" items to dogfood before re-deciding.

Examples

• Use the worked example in examples/ as the facilitation and decision-log model.

Quality Standards

• Critique the work and its objectives, never the maker's competence or identity.
• Distinguish evidence, requirements, hypotheses, and taste before deciding.
• Stop the session when scope, decider, psychological safety, or artefact readiness is absent.

Decision Rules

Condition	Decision	Wrong-choice failure
Work is early and divergent	Use questions and possibilities	Premature polish feedback narrows exploration
Decision is constrained by a requirement	Cite the requirement and owner	Taste is debated as preference
Feedback conflicts	Decider resolves against stated objective	Loudest voice wins
Maker needs time to process	Collect async and decide later	Defensive real-time reactions replace judgement

Capability Contract

Read and facilitation are required. Writing is limited to the authorised brief, feedback, and decision record. Do not alter the reviewed artefact, expose private feedback, or infer consensus without participant evidence.

Degraded Mode

If required evidence or tooling is unavailable, use the scoped fallback below and mark the result unverified. Without synchronous access, run an asynchronous critique with deadlines and the same structured fields. Without a named decider or objective, pause decisions and return the missing-context checklist.

• examples/critique-session-worked.md — a full worked critique session of a real artifact (a first-time-user onboarding flow demo): the agenda and frame, the author's ask, the feedback captured under the chosen framework, the taste-vs-requirement split, the single-decider decisions, and the recorded decision log with next-demo. Concrete, not lorem.

References

• references/critique-protocol.md — the session structure: agenda + timings, the three feedback frameworks (I-like/I-wish/what-if; problem-not-solution; ladder of feedback) and when to use each, the taste-vs-requirement split rule, the single-decider + decision-capture model, psychological-safety practices, and the live-vs-async decision.
• doctrine/references/creative-selection-and-taste.md — the doctrine spine of this skill: no critique without a concrete demo (rule 1); the algorithm-vs-heuristic lens that powers the taste-vs-requirement split; the "41 shades of blue" lesson (don't outsource a taste decision to a vote/metric); "live on it" for heuristic calls that can't be settled in one review.
• doctrine/design-doctrine.md — the Mission (the moat is looking human-made) and the Anti-Slop Charter; a crit also surfaces convergent/slop choices, and the rule that taste choices trace to human design authority, never to a vote or a model's pick.

Plugins (Load Alongside)

Companion Skill	When to Load
heuristic-evaluation-and-design-critique	When the session needs an expert evaluation as input — run that to produce severity-rated findings, then bring them into the crit for the team to decide on. (That EVALUATES; this FACILITATES.)
design-audit	When a scored diagnostic of the artifact should seed the discussion — its triage queue is good crit input.
design-qa-and-pre-launch-review	Downstream — once the crit's decisions are built, run the ship gate.
ux-remediation-and-redesign	Downstream — to execute and re-validate the fixes the crit decided on.

---

1. The method in one paragraph

A design critique is not a vote and not a venue for opinions — it is a facilitated session in which a small group reacts to a concrete piece of work so the author and the single decider can make better, faster decisions. Its discipline comes from four moves: a tight frame (what stage, what question, what is off the table), a named feedback framework (so everyone gives feedback in the same shape), separating taste from requirement (objective standards are non-negotiable; taste calls are the decider's, never the room's vote), and capturing decisions, not comments. Run it this way and honesty and psychological safety stop being in tension: the work is critiqued, the maker is protected, and the meeting ends with decisions and a next demo rather than a pile of notes and a bruised author. (Spine: doctrine/references/creative-selection-and-taste.md — demo-grounded critique, the algorithm/heuristic lens, the single decider, and "live on it".)

2. How this differs from the EVALUATE pair

	Critique facilitation (this skill)	heuristic-evaluation / design-audit
What it is	Running the human session	Producing an expert evaluation of the artifact
Unit	A meeting / async thread + a decision log	A findings report / a 0–100 score
Who acts	A facilitator, a small roster, one decider	One (or a few) evaluators against a rubric
Output	Decisions, requirement-vs-taste split, next demo	Severity-rated findings / gated score
Question	How do we run this review so it's productive, safe, and decisive?	What's wrong with the artifact, and how bad?

They compose: an audit or heuristic evaluation can be an input that seeds the critique, but the critique is where humans discuss and decide. Keep the boundary — do not re-derive heuristic findings here; bring them in and facilitate the decision.

---

Sources: doctrine/references/creative-selection-and-taste.md (Kocienda, Creative Selection — demo-grounded critique, the single decider + decision capture, the algorithm-vs-heuristic lens, the "41 shades of blue" taste-not-vote rule, "live on it"); the design-crit feedback frameworks (I-like/I-wish/what-if; problem-not-solution; ladder of feedback) as used in design-studio practice; doctrine/design-doctrine.md (the Anti-Slop Charter and the human-design-authority asymmetry rule). Companion to the EVALUATE pair heuristic-evaluation-and-design-critique and design-audit.