Initial commit for claims-gpt-governance-pack

standards/naming_conventions.md

@@ -1,32 +1,28 @@
 # Naming Conventions
 
 ## Prompts
-
-Format: `{INTENT}_{CATEGORY}_V{NUMBER}`
-
+Format: `{DOMAIN}_{PURPOSE}_{VERSION}`
 Examples:
 - `CLAIM_TRIAGE_V1`
 - `FRAUD_DETECTION_V2`
+- `COMPLIANCE_CHECK_V1`
 
 ## Models
-
-Format: `{author}_{model}_{version}`
-
+Format: `{name}_{date}`
 Examples:
 - `kimi_0905`
-- `mixtral_8x7b_1`
+- `mixtral_8x7b`
+- `mistral_v0.1`
 
 ## Branches
-
-- Feature: `feature/{issue-id}-{description}`
-- Bugfix: `fix/{issue-id}-{description}`
-- Release: `release/v{YYYY-MM-DD}`
+- `main`: Production code
+- `staging`: Pre-prod
+- `develop`: Development
+- `feature/{name}`: New features
+- `fix/{name}`: Bug fixes
+- `release/{version}`: Release branches
 
 ## Tags
-
-- Staging: `staging/v{YYYY-MM-DD}`
-- Production: `prod/v{YYYY-MM-DD}`
-
----
-**Version**: 1.0
-**Last Updated**: 2026-02-22
+- `staging-v1.0`: Staging release
+- `prod-v1.0`: Production release
+- `hotfix-v1.0.1`: Production hotfix

workflows/model_approval_workflow.md

@@ -1,37 +1,49 @@
 # Model Approval Workflow
 
-**Owner**: AIReviewer
-**SLA**: < 3 days draft  approval
-
-## Stages
-
-### 1. Draft
-- Developer creates model.json
-- Commits to feature branch
-
-### 2. Automated Evaluation
-- GitHub Actions runs tests
-- Golden set evaluation
-- Risk score calculated
-
-### 3. Risk Assessment
-- System calculates risk score
-- Low risk (< 0.3): Auto-approve
-- Medium risk: AIReviewer review
-- High risk: Executive approval
-
-### 4. Compliance Review
-- Legal/audit verifies policies
-- Approves or requests changes
-
-### 5. Approval
-- AIReviewer approves PR
-- All checks must pass
-
-### 6. Release
-- Tag as v{date}
-- Notify stakeholders
+## 6-Stage Process
+
+### Stage 1: Draft (Submitter)
+- Developer creates model variant
+- Runs local tests
+- Prepares documentation
+- **SLA**: None (internal)
+
+### Stage 2: Automated Eval (System)
+- Runs evaluation suite
+- Checks against schema
+- Tests hallucination, bias
+- **SLA**: 2 hours
+
+### Stage 3: Risk Assessment (System)
+- Calculates risk score
+- Determines approval path
+- **SLA**: 10 minutes
+
+### Stage 4: Compliance Review (Governance team)
+- If risk = High: Legal review
+- Checks CBK/DFSA rules
+- Approves or denies
+- **SLA**: 24 hours
+
+### Stage 5: Approval (Appropriate authority)
+- Low risk: Auto-approved
+- Medium risk: AI reviewer approve
+- High risk: Executive approve
+- **SLA**: 24 hours (medium), 4 hours (high)
+
+### Stage 6: Release (DevOps)
+- Deploy to production
 - Monitor for issues
-
----
-**Timeline**: < 3 days total
+- Rollback plan ready
+- **SLA**: 1 hour
+
+## Timeline
+- Simple model: 24 hours
+- Complex model: 72 hours
+- SLA: < 3 days
+
+## Rollback Criteria
+- Error rate > 5%
+- Fraud detection < 30%
+- Compliance violations
+- P95 latency > 3.0s

workflows/prompt_approval_workflow.md

@@ -1,29 +1,28 @@
 # Prompt Approval Workflow
 
-**Owner**: AIReviewer  
-**SLA**: < 2 days draft  approval
+## 4-Stage Process (Simplified)
 
-## Process
+### Stage 1: Draft
+- Developer writes prompt
+- Tests locally
+- Preparation: 2 hours
 
-### 1. Draft
-- Create CLAIM_TRIAGE_V2.json
-- Test locally
-- Commit to feature branch
-
-### 2. Validation
+### Stage 2: Validation
+- Automated testing
 - Schema validation
-- Prompt tests pass
-- Example execution works
+- Duration: 30 minutes
 
-### 3. Review
-- AIReviewer reviews for safety
-- Check KPIs
-- Approve or request changes
+### Stage 3: Review
+- Technical review
+- Compliance check
+- SLA: 24 hours
 
-### 4. Merge & Deploy
+### Stage 4: Merge
 - Merge to main
-- Tag as prompt-v{version}
-- Update routing config
+- Auto-deploy to staging
+- Deploy to prod (manual)
+- SLA: < 2 days total
 
----
-**Timeline**: < 2 days total
+## SLA
+Total time: < 48 hours
+Blocking issues: 4-hour response