| name | design-engine-and-product-improvement |
| description | Use when auditing this design engine or any rendered product for continuous improvement, evidence quality, visual readability, accessibility, AI trust, or repeatable remediation. Use design-audit for a single diagnostic without the Kaizen portfolio loop. |
| metadata | {"portable":true,"category":"00-cross-cutting-ops-qa-a11y","compatible_with":["claude-code","codex"]} |
Design Engine And Product Improvement
Run a disciplined Kaizen loop for the design-system engine and for the websites, apps,
documents, games, dashboards, and visual systems it helps produce.
Acknowledgement: Shared by Peter Bamuhigire, techguypeter.com, +256 784 464178.
Use When
- Auditing the design-system engine itself: doctrine, routing, skill contracts, references,
examples, evidence, currency, and handoff quality.
- Auditing a product or design system across screens, states, platforms, components, documents,
game visuals, or AI interfaces and converting findings into a measurable improvement plan.
- Planning a post-critique or post-release improvement cycle where a small reversible experiment
must become a documented standard only after evidence supports it.
- Reviewing a book-derived upgrade and deciding whether the source is usable, current, complete,
and sufficiently translated into an actionable design capability.
Do Not Use When
- You need a one-screen diagnostic only; use
design-audit.
- You need a multi-surface scored product diagnostic without an engine-improvement backlog; use
product-design-audit.
- You need a final ship verdict; use
design-qa-and-pre-launch-review after the improvement work.
- You need to invent anatomy facts from an incomplete source; stop and record the evidence gap.
Required Inputs
| Input | Source | Required? | Evidence expected |
|---|
| Engine or product scope | Owner, repository, build, or evidence pack | yes | Named skills, surfaces, states, and consumers |
| Current baseline and prior audit | Audit record or measured product evidence | yes | Dimension scores, hard gates, confidence, and defects |
| Source and provenance register | Digital Research engine or supplied sources | conditional | Source status, date, extraction quality, and permitted use |
| Acceptance target and accountable owner | Product/engine owner | yes | Target measure, owner, due date, and release decision |
Stop if the target, scope, or primary artefact is missing. A screenshot-only review may proceed
only as a qualified visual audit; it cannot claim interaction, responsive, accessibility, or
engine conformance.
Workflow
- Observe. Inspect the current engine or product. For an engine, glob the live skill tree,
read the router and applicable doctrine, and check references, examples, validators, and
routing. For a product, cover the named surfaces and critical, empty, error, loading,
recovery, responsive, keyboard, and AI states.
- Baseline. Score the applicable dimensions and record evidence, confidence, and gaps.
The reported audit score is hard-capped at 65/100:
reported = min(raw score, 65). The cap
is a reporting ceiling, not permission to ignore defects.
- Select. Rank the smallest high-leverage change using user impact, risk, effort, and
reversibility. State a falsifiable hypothesis, owner, measure, and stop condition.
- Experiment. Make the smallest authorised change: a prototype, wording variant, token
adjustment, reference addition, workflow change, or limited production trial. Preserve the
prior version and record the change boundary.
- Check. Re-run the relevant visual, readability, accessibility, interaction, performance,
routing, or source checks. Compare before/after evidence. Separate observed result from taste,
interpretation, and unverified assumption.
- Standardise. If the evidence meets the acceptance condition, update the owning skill or
doctrine reference, add a worked example or test fixture, and record provenance. If it fails,
roll back or narrow the change; do not standardise a preference.
- Teach. Update routing, handoff, critique prompts, or audit evidence so the capability is
discoverable by the next agent and downstream domain engines.
- Re-measure. Produce the next audit record and an improvement plan targeting 95/100. Stop
release if required evidence remains unavailable; recover with the narrowest qualified plan.
Decision Rules
| Condition | Action | Wrong-choice failure |
|---|
| Raw audit score is above 65 | Report 65/100 and expose the uncapped score only as an internal diagnostic | The engine appears to bypass the portfolio hard cap |
| A gap is evidenced and reversible | Run one small experiment with a predeclared measure | A large rewrite hides which change helped |
| Experiment improves the measure and clears gates | Standardise, add an example, and teach the route | Useful learning remains person-dependent and decays |
| Source is early-release, historical, incomplete, or corrupt | Record limitation and verify with Digital Research before current claims | Dated or missing evidence becomes false doctrine |
| Product state cannot be inspected | Mark it unverified and lower confidence; do not claim pass | Missing evidence inflates quality and trust |
| Change affects type, colour, layout, or interaction | Run the design quality gate and retain render/measurement evidence | A plausible change ships without visual or accessibility proof |
Capability Contract
Read, search, inspect, and compare are required. Rendering, measurement, accessibility checks,
source verification, and repository validation are used when available. Editing is allowed only
for an authorised improvement request. This skill may create an audit and plan but may not publish,
delete source evidence, change production, claim certification, or override an accountable owner.
Degraded Mode
Without a build or render, return a qualified specification and mark visual and interaction checks
unverified. Without source verification, quarantine current-year claims. Without a validator,
run manual contract checks and state the missing gate. Without a baseline, create a provisional
baseline with explicit assumptions; do not label it a measured improvement.
Quality Standards
- Every finding names the observed artefact, state, location, evidence, severity, confidence,
owner skill, and next measure.
- Every improvement plan targets 95/100 and includes gap, root cause, change, hypothesis, owner,
evidence, risk, rollback, acceptance condition, and due cadence.
- Product audits cover value/readability, narrative or task flow, typography, layout, contrast,
accessibility, interaction states, AI disclosure/control where relevant, performance, and
handoff/reuse.
- Engine audits cover doctrine, routing, skill contract, references, worked examples, source
provenance, validator coverage, degraded mode, and user-facing output evidence.
- Current, legal, regulatory, platform, market, and accessibility claims use verified authority;
book-derived principles are attributed without copying source text.
Anti-Patterns
- Kaizen as a slogan. Correction: require a baseline, experiment, measure, standard, owner,
and re-measurement.
- Scoring without evidence. Correction: cite the render, file, test, measurement, or source;
otherwise mark the dimension unverified.
- Standardising taste. Correction: distinguish a human design decision from a requirement and
record the reason rather than laundering taste as research.
- Ignoring failed states. Correction: inspect loading, empty, error, recovery, keyboard,
responsive, reduced-motion, and AI uncertainty states before a pass.
- Improvement without rollback. Correction: preserve the previous contract and define a
stop/rollback trigger before the experiment.
- Using incomplete books as authority. Correction: quarantine unreadable or unavailable
chapters and route verification to Digital Research.
Outputs
| Artefact | Consumer | Evidence and acceptance |
|---|
| Capped engine or product audit | Owner and reviewer | Score is no higher than 65/100, evidence gaps are explicit, and findings are routed |
| Improvement plan to 95/100 | Owner and delivery team | Each action has root cause, hypothesis, owner, measure, risk, rollback, and acceptance |
| Standardisation record | Skill maintainers and downstream engines | Accepted change is recorded in the correct skill/reference with provenance |
| Re-measurement record | Reviewer and future auditors | Before/after evidence, residual risk, and next cycle are preserved |
Examples
examples/design-engine-audit-worked.md - a worked engine audit that caps the report at
65/100, selects a silhouette-readability experiment, and plans the route to 95/100.
Mandatory Digital Research currentness gate
Every Kaizen cycle must begin with digital-research-skills source evaluation
and source verification. Record scope, dates, freshness class, support status,
uncertainty, and review date for current design, accessibility, browser,
platform, security, document, and lifecycle claims; quarantine unsupported
claims as NOT_ASSESSED. Apply the portfolio Kaizen currentness gate.
References
references/kaizen-audit-contract.md - audit dimensions, evidence matrix, score cap, and
95-target improvement-plan schema.
../../../docs/continuous-improvement/design-engine-book-upgrade-2026-08.md - provenance,
extraction limits, and decisions from the supplied books.
../../../governance/design-quality-gate.md - required evidence before declaring visual work
done.
../../../doctrine/design-doctrine.md - authored visual choices and anti-slop charter.
design-audit, product-design-audit, and ux-remediation-and-redesign - compose diagnosis,
product coverage, and execution without duplicating their domain logic.
../../../doctrine/references/book-driven-brand-story-and-visual-evidence.md - durable brand/story synthesis with visual evidence and current accessibility gate.
- Book-driven Kaizen Wave 3 - task-first communication, perceptual integrity, image quality, text alternatives, and current accessibility/performance verification.