Nielsen Heuristic Evaluation Reference
Purpose: Structured expert-review methodology applying Nielsen's 10 usability heuristics (1994, still the field's de-facto baseline) plus domain-specific extensions. Surfaces usability issues without user recruitment, at roughly 1/10 the cost of moderated testing, while preserving traceability from finding to heuristic violated.
Scope Boundary
- echo
heuristic: persona-agnostic heuristic review against Nielsen's 10 + extended sets. Produces severity-ranked issue list with heuristic-citation audit trail. - echo
walkthrough(default): persona-based cognitive walkthrough — simulates specific user minds through a flow. - palette (elsewhere): remediation. Once heuristics surface violations, Palette owns the redesign.
- canon
wcag(elsewhere): WCAG compliance assessment. Heuristic evaluation is broader than accessibility — overlap exists around heuristic 4 (consistency) and heuristic 9 (recognize/recover errors). - field (elsewhere): empirical validation with real users. Heuristic evaluation predicts issues; research confirms them.
Workflow
PREP → define scope (screens/flows), recruit 3-5 evaluators, agree heuristic set
→ brief evaluators: independent passes first, group reconciliation second
PASS 1 → each evaluator walks the interface alone, logs findings per heuristic
→ issues independent — do not discuss until logged
PASS 2 → evaluators walk the interface again with task scenarios in mind
→ adds context-dependent violations missed in open exploration
RECONCILE → merge findings, dedupe, assign severity (0-4), tag heuristic(s) violated
→ resolve scoring disagreements via worst-case consensus
REPORT → severity-ranked issues with heuristic citation, evidence, and fix category
→ hand off fix candidates to Palette; critical issues to Judge for gatingThe 10 Heuristics (Nielsen 1994, Refined 2020)
| # | Heuristic | Core question | Common violations |
|---|---|---|---|
| 1 | Visibility of system status | Does the UI tell users what's happening? | Silent loading, hidden progress, unclear save state |
| 2 | Match real world | Does language match user mental models? | System jargon, reversed metaphors, cryptic icons |
| 3 | User control and freedom | Can users undo / escape? | No cancel on long operations, no confirm-undo pair, forced paths |
| 4 | Consistency and standards | Does it follow platform and product conventions? | Mixed button styles, inconsistent gestures, off-pattern modals |
| 5 | Error prevention | Does it prevent errors before they occur? | Destructive actions without confirm, free-text where constrained input possible |
| 6 | Recognition over recall | Are options visible rather than remembered? | Hidden menus, context-free icons, no recent/suggested values |
| 7 | Flexibility and efficiency | Does it serve novice and expert? | No shortcuts, no power-user batch ops, one-path-only design |
| 8 | Aesthetic and minimalist design | Is noise minimized? | Information overload, competing CTAs, decorative over functional |
| 9 | Recognize, diagnose, recover from errors | Are errors plain-language, diagnostic, recoverable? | "Error 500", no next step, no retry |
| 10 | Help and documentation | Is task-scoped help available? | Only general docs, no contextual tips, stale screenshots |
Severity Scale (Nielsen)
| Score | Meaning | Action |
|---|---|---|
| 0 | Not a usability problem | Dismiss |
| 1 | Cosmetic only | Fix if time permits |
| 2 | Minor — low-frequency or easy workaround | Low priority |
| 3 | Major — frequent or high-impact | Fix before launch |
| 4 | Catastrophic — blocks task, must fix | Ship-blocker |
Apply worst-case consensus: if evaluators split 3/4, use 4. Severity ≥3 items are the escalation set for launch gates.
Evaluator Count Calibration
| Evaluators | Issues found | Cost multiplier |
|---|---|---|
| 1 | ~35% | 1.0× |
| 3 | ~60% | 3.0× |
| 5 | ~75% | 5.0× |
| 8 | ~85% | 8.0× (diminishing returns) |
Nielsen's published curve: 3-5 evaluators is the sweet spot. Below 3, coverage is unreliable. Above 5, cost grows faster than issue discovery. Prefer diverse evaluators (UX, accessibility, domain) over more homogeneous ones.
Extended Heuristic Sets
Nielsen 10 covers general GUIs. Apply the right extension when the domain demands:
- Mobile / touch: fat-finger safety, one-handed reach zones, gesture discoverability (Kendall's 7, Bertini et al.)
- Data-dense enterprise: Shneiderman's 8 Golden Rules — emphasizes informative feedback, reversible actions, locus of control
- Games: Heuristic Evaluation for Playability (HEP) — Desurvire — goal clarity, reward feedback, difficulty progression
- Voice / conversational: Murad's 17 — interruption recovery, turn-taking, ambiguity resolution
- XR / immersive: Sutcliffe's XR heuristics — presence, motion sickness, embodiment
Cite the set used in the report header. Mixing sets without attribution masks domain-specific issues.
Anti-Patterns
- One-evaluator reviews — single evaluators find ~35% of issues. Brand it "design review," not "heuristic evaluation."
- Skipping the independent pass — group-first reviews anchor on the loudest voice and miss ~40% of unique issues.
- Heuristic evaluation without task scenarios — open exploration misses context-dependent violations (state-specific, role-specific, frequency-specific).
- Reporting violations without severity — forces stakeholders to triage, and high-severity items get lost among cosmetic notes.
- Treating heuristic findings as empirical — they are predictive. Every critical finding should be validated with ≥5 real users before betting on the fix.
- Applying Nielsen-10 to voice, games, or XR — generic heuristics miss domain-specific issues; use the matching extended set.
- Omitting evaluator backgrounds — a heuristic review by 3 visual designers and a review by 1 visual + 1 a11y + 1 domain expert surface very different issues. Document the panel.
Handoff
- To Palette: severity
≥3issues with heuristic citation, affected element, and recommended fix category (feedback, affordance, content, error state). - To Canon
wcag: heuristic 4 (consistency) and 9 (error recovery) violations that overlap WCAG criteria — cross-validate against WCAG 2.2. - To Field: severity
≥3issues with ambiguous cause — escalate to 5-user moderated study for empirical validation. - To Judge: catastrophic (severity 4) issues as ship-blockers with heuristic-cited evidence.
- To Echo
walkthrough: reconcile heuristic findings against persona walkthrough. If walkthrough surfaced the same issue, confidence is higher.