| name | kaizen-improvement-system |
| description | Use when auditing or improving the social-media engine or any campaign, content system, strategy, report, training asset, or marketing product it produces. |
| metadata | {"portable":true,"compatible_with":["claude-code","codex"]} |
Kaizen Improvement System
Acknowledgement: Shared by Peter Bamuhigire, techguypeter.com.
Use When
- Auditing this engine or a social/digital-marketing product after planning, production, publication, or reporting.
- Turning audience feedback, platform changes, performance evidence, or client review into a tested improvement.
Do Not Use When
- Only one creative asset needs a normal anti-slop gate.
- Current platform, market, legal, or policy claims lack source-register verification.
Required Inputs
| Artefact | Source/provider | Required? | Purpose | If absent |
|---|
| Brief, audience/market, channel, objective, content/campaign/report, evidence, current score, and constraints | Client brief, platform data, and engine | yes | Set audit scope and improvement target | Stop or mark unassessed |
Workflow
- Read the local adoption plan, router, anti-slop gates, and portfolio standard.
- Inventory strategy, platform, content, AI, analytics, reporting, training, policy, and campaign routes.
- Score applicable dimensions and output types. Publish
min(raw score, 65) and record legal, rights, safety, and evidence blockers.
- Audit audience value, cultural fit, evidence, message/story, accessibility, channel execution, ethical persuasion, AI provenance, measurement, handoff, and learning.
- Build a 95/100 plan with exact skill/reference/fixture, owner, experiment, metric, acceptance evidence, and rollback.
- Run a small content, channel, CTA, or measurement experiment. If the evidence or safety gate fails, pause, recover the safe version, and correct it.
- Standardise only what the evidence supports, rerun the relevant gates, and schedule the next review.
Outputs
| Artefact | Consumer | Acceptance condition |
|---|
| Capped audit, evidence/legal/quality blockers, 95/100 plan, experiment result, and standardised learning | Strategist, client, reviewer, and release owner | Evidence, owner, decision rule, guardrail, and re-audit date are explicit |
Evidence Produced
| Evidence | Consumer | Acceptance condition |
|---|
| Brief-to-output map, source register, creative/legal review, measurement result, and before/after decision record | Reviewer and release owner | Another reviewer can reproduce the decision and verify the learning |
Two-level product contract
Run the loop once for the engine (routes, skills, references, validators, templates,
handoffs, and East Africa defaults) and once for the product (campaign, content,
calendar, report, training asset, policy, or website/social output). At each level,
the first analysis is capped at min(raw_score, 65). Then select one root cause and
one reversible improvement toward 95/100. Use the product adapters in
references/product-type-adapters.md; do not transfer a campaign metric directly
to a report, training asset, or website without its own evidence.
Capability
Read and search are required. Audits are read-only by default; edits, publishing, or external communication require explicit authority and permission. Route current facts to Digital Research and visual work to design-system-skills.
Degraded Mode
If the brief, audience evidence, platform data, source register, rights, or reviewer is unavailable, return a qualified result, mark the check not assessed, and withhold release readiness.
Decision
| Condition | Action | Failure or risk avoided |
|---|
| A performance claim lacks a source or baseline | Remove or qualify it | Fabricated proof |
| A trend improves reach but harms trust, safety, or cultural fit | Reject, pause, or narrow the test | Harmful optimisation |
| A test passes outcome and guardrail checks | Standardise the learning and re-audit | Lessons lost |
Quality Standards
Do not invent platform benchmarks, audience statistics, or performance results. Use British English and East African defaults unless the brief says otherwise.
Mandatory 65-to-95 gate
The first review is an initial analysis: calculate raw findings, publish only
min(raw_score, 65), and keep evidence, rights, safety, and measurement gaps visible.
Do not improve the score by adding copy alone. After the capped baseline, target
95/100 through one small reversible change at a time, with a root cause, owner,
primary measure, trust/accessibility guardrail, stop/rollback rule, acceptance
evidence, standardisation decision, and re-audit date.
Anti-Patterns
- Optimising vanity metrics without a decision. Fix: define outcome, guardrail, and action.
- Copying a trend without audience fit. Fix: test the local hypothesis.
- Treating a content calendar as strategy. Fix: connect content to funnel and learning.
- Using AI output without provenance or cultural review. Fix: run AI, legal, and cultural gates.
- Closing a test without a learning record. Fix: standardise or reject explicitly.
Worked Example
If a short-form video gets reach but no qualified action and triggers cultural concerns, retain the result as a failed experiment, pause the variant, document the evidence, and test a better-fit message with a guardrail.
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 platform, market, legal, policy,
technology, and lifecycle claims; quarantine unsupported claims as
NOT_ASSESSED. Apply the portfolio Kaizen currentness gate.
References
- Local adoption plan
- Portfolio standard: resolve
digital-research-skills through the global engine-routing table, then read docs/continuous-improvement/portfolio-kaizen-standard-2026-08.md.
skills/meta-analytics-ops/meta-testing-framework/skills/ai-marketing/anti-ai-slop/- Book-driven campaign learning and retention - audience, story-to-action, ethical experimentation, retention, and currentness.
