---
# METADATA
id: Meta-RefineOutput
name: Meta - Refine Output through Iterative Self-Critique
version: 1.2 # Aligned with consolidated strategy (Schema v6.0, Catalog v2.0, AIOpsProtocols v1.1)
status: Active
description: >
A meta-process template guiding an AI to iteratively refine a generated output
(e.g., a document, a plan, another template, a complex response) through cycles of
self-critique (including mandatory checks from AIOperationalProtocols), adversarial analysis,
simulation (if applicable), and feedback incorporation. The goal is convergence on a stable,
robust, and optimized output.
type: Meta_Process
domain: AI Self-Improvement, Quality Assurance, Generative AI
keywords: [self-critique, iterative refinement, continuous improvement, adversarial analysis, simulation, feedback loop, convergence, meta-cognition, AI safety, AI alignment, data integrity]
# RELATIONSHIPS
invoked_by:
- "User"
- "[[ProjectOrchestrator]]" # For refining complex drafts (plans, charters, deliverables)
- "[[archive/templates/projects/LiteContentCreator]]" # For refining drafted content
- "[[InvokeAISkill]]" # Internal logic within ProjectOrchestrator, for complex skill executions
- "System (Self-Triggered)" # E.g., when AI generates new/updated templates
references_schema: "[[ProjectStateSchema]]" # v6.0
uses_skills_from: "[[archive/templates/projects/AISkillsCatalog]]" # v2.0 (May use skills like CritiqueArtifact)
uses_knowledge_artifacts:
- "[[AIOperationalProtocols]]" # For mandatory critique checks
# USAGE
instructions_for_ai: |
**Objective:** To take an initial AI-generated output (the "Draft Output") and subject it to a rigorous, iterative process of self-evaluation and refinement until it reaches a stable state of high quality, robustness, and alignment with specified goals or criteria. This process emphasizes deep feedback loops, diverse critique methods, and adherence to `AIOperationalProtocols`.
**CRITICAL PREREQUISITES (AI MUST VERIFY AT START of this process invocation):**
1. **`ProjectStateSchema`:** Confirm loaded. If not, HALT and request.
2. **`AISkillsCatalog`:** Confirm loaded. If not, WARN (critique depth may be limited).
3. **`AIOperationalProtocols` Definition/Instance:** Confirm loaded/active. If not, WARN (mandatory checks may be based on general principles).
**Interaction Note for AI:** Maintain concise, action-oriented "machine voice." Filename references must NOT include extensions. Output for template files must be complete and non-truncated. Adhere to DATE/TIMESTAMP/DURATION handling critical directive.
**Input:**
1. `draft_output_content`: The actual content of the AI-generated output to be refined.
2. `draft_output_reference`: string (Optional, if output is in `project_state` or a file).
3. `refinement_goals_and_criteria`: object or string (Specific goals for this refinement cycle).
4. `max_iterations`: integer (Optional, default: 3).
5. `convergence_threshold`: string (Optional, description of convergence).
6. `user_feedback_integration_mode`: string (Enum: "AfterEachIteration", "AfterConvergenceAttempt", "OnDemand". Default: "AfterConvergenceAttempt").
7. Access to `project_state` (if relevant).
**Meta-Process Steps (Iterative Loop):**
4. **Initialization & Pre-Generation Constraint Review:**
a. Verify prerequisites. If criticals missing, HALT. Issue WARNs if non-criticals missing.
b. Store `draft_output_content` as `current_version_output`.
c. Initialize `iteration_count = 0`, `refinement_log`: list of objects.
d. AI internalizes inputs.
e. **AI performs "Pre-Generation Constraint Review Protocol"** (from active `AIOperationalProtocols` KA or its definition) to compile `active_constraints_checklist` for *this refinement task on the draft_output_content*.
f. State: "Meta-Refinement initiated. Goals: [summarize goals]. Active constraints checklist compiled."
5. **Iteration Start:**
a. Increment `iteration_count`. If `iteration_count > max_iterations`, proceed to Step 6.
b. Log in `refinement_log`: "Starting refinement iteration [iteration_count]."
6. **Multi-Perspective Self-Critique & Analysis (Adhering to `AIOperationalProtocols`):**
AI performs these on `current_version_output`, referencing `active_constraints_checklist`:
a. **MANDATORY CHECKS (from `AIOperationalProtocols` - Data Integrity & Self-Correction Protocol, incorporating TID017):**
i. YAML/Output Completeness (No placeholders/truncation). Log Pass/Fail.
ii. Data Sourcing (Traceability, No Invention). Log Pass/Fail/PartiallySourced.
iii. Placeholder Protocol Adherence (If output is a template, does it guide AI correctly on input placeholders?). Log Pass/Fail.
iv. Adherence to all other items in `active_constraints_checklist`. Log Pass/Fail per constraint.
b. **Goal Alignment Critique:** Evaluate against `refinement_goals_and_criteria`.
c. **Adversarial / Contrarian Analysis.**
d. **Simulation / Scenario Testing (If Applicable).**
e. **Comparative Analysis (If Applicable).**
f. **Clarity, Conciseness, Usability Review.**
g. **(Optional) Invoke Specific AI Skills for Analysis** (e.g., `CritiqueArtifact` from `AISkillsCatalog`).
h. All findings logged in `refinement_log`.
7. **Synthesize Critique Findings & Propose Revisions:**
a. AI analyzes `refinement_log`, prioritizing failures from MANDATORY CHECKS.
b. Identify critical issues and high-impact areas for revision.
c. Generate specific, actionable list of proposed changes.
d. Log proposed changes.
8. **Implement Revisions (Generate Next Version):**
a. AI applies proposed changes to create `next_version_output`.
b. AI ensures revisions address issues from Step 3, especially mandatory check failures.
c. Log: "Revisions implemented for iteration [iteration_count]."
9. **Assess Convergence & Loop Control:**
a. Compare `next_version_output` with `current_version_output`.
b. Evaluate `next_version_output` against `convergence_threshold` AND results of mandatory checks (from Step 2, as if applied to `next_version_output`).
Convergence requires: All MANDATORY CHECKS pass; No significant new critical issues; Output stabilizing.
c. If converged: Set `current_version_output = next_version_output`. Log convergence. Proceed to Step 7.
d. If NOT converged AND `user_feedback_integration_mode == "AfterEachIteration"`: Set `current_version_output = next_version_output`. Proceed to Step 6.
e. If NOT converged AND `user_feedback_integration_mode != "AfterEachIteration"`: Set `current_version_output = next_version_output`. Loop to Step 1.
10. **User Intervention / Feedback Point:**
a. State: "Refinement iteration [iteration_count] complete." (Or max iterations message).
b. Present `current_version_output` (NO TRUNCATION).
c. Present summary of `refinement_log` (key critiques, changes, mandatory check status, convergence assessment).
d. Ask user:
"1. Does current version meet refinement goals: [summarize goals]? (Yes/No/Partially)
2. Provide feedback/directives for further refinement, or confirm acceptance." (Freeform).
e. If user confirms acceptance: Proceed to Step 7.
f. If user provides feedback: Incorporate into `refinement_goals_and_criteria`. Log feedback. Reset `iteration_count` if major. Loop to Step 1 (or Step 4).
11. **Present Final Converged Output:**
a. State: "Meta-Refinement process converged after [iteration_count] iterations (or with user approval)."
b. Present final `current_version_output` (NO TRUNCATION).
c. Present final summary from `refinement_log`.
d. Instruct user on saving/using output (e.g., "Refined content ready. Save as `[SuggestedFilename]` in `[Path]`)."
e. Terminate this meta-process.
**Output:**
* The refined output content.
* (Optionally) The `refinement_log`.
# OBSIDIAN
obsidian_path: "templates/projects/support/MetaRefineOutput"
---