Cascading Quality Gates
Quality Gates ensure no mediocre output leaves your studio. Every piece of content passes through an automated evaluate-revise-verify cycle powered by specialized lims.
How It Works
- Evaluate — A quality-assurance lim reviews the output, scoring it 0-100 on multiple dimensions
- Revise — If the score is below the threshold (default: 75), an editorial-lead lim revises the weak points
- Verify — A signal-analyst lim verifies the improvements meet the quality bar
- Repeat — Up to 3 revision cycles before escalating to human review
Quality Gate Configuration
- Minimum Score — the threshold to pass (default: 75/100)
- Max Revisions — maximum cascade attempts before escalation (default: 3)
- Evaluator — which archetype evaluates (default: quality-assurance)
- Revisor — which archetype revises (default: editorial-lead)
- Verifier — optional third archetype for verification (default: signal-analyst)
Revision Chains
Each quality gate creates a revision chain — a full audit trail of every evaluation pass, score, feedback, and revision. You can inspect any chain to see exactly how content was improved.
Escalation
If a piece of content fails quality gates after the maximum number of revisions, it's escalated with a "task_stuck" event. You'll see it in your dashboard and can decide to approve, provide feedback, or reassign the task.