DOMAIN:CREATIVE — CREATIVE FEEDBACK PROTOCOL¶

OWNER: leah ALSO_USED_BY: rick, christel, ilian, felice, alexander (receiving agents), faye, sytske (PM awareness) UPDATED: 2026-04-03 SCOPE: structured improvement briefs for BLOCKED/CONDITIONAL creative deliverables

PURPOSE¶

RULE: every BLOCKED or CONDITIONAL scoring result produces a structured improvement brief RULE: briefs are machine-parseable — receiving agent acts on them directly RULE: briefs replace vague "rejected, redo" with specific, prioritized, actionable fix instructions RULE: maximum 3 improvement cycles before human escalation

IMPROVEMENT_BRIEF_FORMAT¶

HEADER (YAML frontmatter)¶

---
brief_id: IB-{YYYYMMDD}-{seq}
deliverable_id: {reference to original deliverable}
deliverable_type: copy | brand_identity | motion | ui_design | production_asset
originating_agent: {who produced the deliverable}
scoring_result: BLOCKED | CONDITIONAL
total_score: {0-100}
threshold_required: 85
cycle: {1 | 2 | 3}
previous_scores: [{cycle 1 score}, {cycle 2 score}]
---

FAILED_DIMENSIONS (ordered by impact — fix highest priority first)¶

FOR EACH dimension below threshold:

## DIMENSION: {name}
SCORE: {current} / TARGET: {minimum acceptable for this dimension}
WEIGHT: {from scoring matrix}
EVIDENCE: {specific element(s) that failed — cite location, quote text, name file}
FIX_GUIDANCE: {concrete action — not "improve this" but "replace X with Y because Z"}
RUBRIC_REF: {which scoring criterion or pattern was violated}
PRIORITY: {1 = fix first, 2 = fix second, ...}

AI_DETECTION_FAILURES (if applicable — only for text-containing deliverables)¶

FOR EACH detected AI pattern:

PATTERN: {pattern ID from ai-detection-patterns.md, e.g., ST-03}
CATEGORY: {Banned Vocabulary | Structural Tell | Style Tell | Filler | Communication Tell}
LOCATION: {paragraph/section/line where detected}
DEDUCTION: {points deducted}
SPECIFIC_TEXT: "{the flagged text}"
FIX: "{specific rewrite suggestion}"

PASSING_DIMENSIONS (context — what worked, don't break it)¶

LIST dimensions that passed with brief note on why they scored well. RULE: this section prevents the receiving agent from accidentally degrading strong dimensions while fixing weak ones.

DEADLINE_AND_ROUTING¶

TURNAROUND: {standard | expedited | FINAL}
  cycle 1 → standard (next available slot in agent's queue)
  cycle 2 → expedited (front of queue — agent is iterating)
  cycle 3 → FINAL — if this fails, escalate to human (Dirk-Jan)
ROUTE_TO: triggers.{originating_agent} via ge-orchestrator
PM_NOTIFIED: {faye | sytske} — BLOCKED status logged
BRIEF_STORED: /ge-ops/clients/{client}/projects/{project}/reconciliation/creative/briefs/

RECEIVING_AGENT_PROTOCOL¶

WHEN triggered with an improvement brief from Leah:

READ the entire brief before starting any rework
CHECK cycle number — cycle 3 = FINAL chance before human escalation
ADDRESS failed dimensions in PRIORITY order (1 first, then 2, etc.)
DO NOT modify passing dimensions — they scored well, leave them alone
APPLY specific FIX_GUIDANCE for each failed dimension — don't reinterpret
FOR AI_DETECTION_FAILURES: rewrite flagged text using the suggested fix, then self-scan the rewrite against ai-detection-patterns.md
SELF_CHECK against ALL failed dimensions before resubmitting
RESUBMIT via ge-orchestrator to triggers.{team}.leah with note: "Improvement brief IB-{id} addressed"

RULE: Leah re-scores the ENTIRE deliverable on resubmission — not just the fixed parts RULE: if you disagree with a finding, note your rationale in the resubmission — Leah will re-evaluate RULE: never ignore an improvement brief — even LOW priority items should be addressed or explicitly justified

ESCALATION¶

IF cycle 3 score < 85: - Leah escalates to human (Dirk-Jan) with all 3 briefs + score progression - Human decides: accept as-is, provide direction, or reassign to different agent - Leah does NOT block indefinitely — 3 cycles is the maximum

IF score does not improve between cycles (e.g., cycle 1: 62, cycle 2: 60): - Leah escalates IMMEDIATELY — agent is stuck, more cycles won't help - Include analysis of WHY score didn't improve (same patterns? new failures introduced?)

EXAMPLE_BRIEF¶

---
brief_id: IB-20260403-001
deliverable_id: RICK-COPY-0042
deliverable_type: copy
originating_agent: rick
scoring_result: BLOCKED
total_score: 64
threshold_required: 85
cycle: 1
previous_scores: []
---

DIMENSION: ai_humanization¶

SCORE: 52 / TARGET: 75 WEIGHT: 0.30 (effective 0.45 with 1.5x multiplier) EVIDENCE: 6 banned vocabulary violations, 2 structural tells in body copy FIX_GUIDANCE: rewrite paragraphs 2-4 using ai-detection-patterns.md remediation protocol. Replace "leverage," "streamline," "comprehensive" with specific language. Break parallel 3-item list structure in paragraph 3. RUBRIC_REF: ai-detection-patterns.md — BV (leverage, streamline, comprehensive), ST-01 (parallel lists) PRIORITY: 1

DIMENSION: clarity_impact¶

SCORE: 65 / TARGET: 75 WEIGHT: 0.20 EVIDENCE: CTA in section 3 is vague ("Learn more about our approach"). No specific next step. FIX_GUIDANCE: replace with specific CTA that tells the reader exactly what happens next ("Book a 15-minute architecture walkthrough — we'll show you how 59 agents handle your stack") RUBRIC_REF: Rick CORE Quality Standards — "action-oriented, specific CTAs" PRIORITY: 2

AI_DETECTION_FAILURES¶

PATTERN: BV-02 (leverage) CATEGORY: Banned Vocabulary LOCATION: paragraph 2, sentence 1 DEDUCTION: -2 SPECIFIC_TEXT: "leverage our multi-agent architecture" FIX: "use our multi-agent architecture" or "build on our multi-agent architecture"

PATTERN: BV-08 (streamline) CATEGORY: Banned Vocabulary LOCATION: paragraph 3, sentence 2 DEDUCTION: -2 SPECIFIC_TEXT: "streamline your development workflow" FIX: "cut 3 steps from your development workflow" or "simplify your development workflow"

PATTERN: ST-01 (parallel lists) CATEGORY: Structural Tell LOCATION: paragraph 3 DEDUCTION: -3 SPECIFIC_TEXT: "Three key benefits: automate X, improve Y, reduce Z" FIX: rewrite as prose with varied structure. Lead with the most surprising benefit. See ai-detection-patterns.md ST-01 example.

PASSING_DIMENSIONS¶

brand_alignment: 82 — messaging pillars clearly expressed, strategic brief well-integrated
audience_resonance: 78 — good use of audience-specific language, technical terms appropriate for CTO audience
craft_quality: 71 — decent rhythm, some strong headlines, but body copy needs tightening

DEADLINE_AND_ROUTING¶

TURNAROUND: standard ROUTE_TO: triggers.alfa.rick PM_NOTIFIED: faye BRIEF_STORED: /ge-ops/clients/acme/projects/rebrand/reconciliation/creative/briefs/IB-20260403-001.md

READ_ALSO: domains/creative/expert-panel-scoring.md READ_ALSO: domains/content/ai-detection-patterns.md READ_ALSO: domains/creative/creative-acceptance-criteria.md