Cross model validator

Cross-Model Validator

This skill validates consistency across multiple canonical service domain models. While canonical-model-validator checks a single model in isolation, this skill detects mismatches between models — broken cross-company interactions, inconsistent authority levels, misaligned reference data, and missing reciprocal endpoints.

When to Use

• After creating or updating models for multiple companies in the ACID supply chain
• Before generating the collaborative network view (acid-collaborative.json)
• When adding a new company to the supply chain and connecting it to existing models
• When cross-company interactions are modified in any model
• As part of the bsa-pipeline-orchestrator pipeline

Instructions

1. Load All Models

Load all canonical models from: docs/architecture/bsa/canonical-models/*.json

Build a master index of:

• All services across all models (keyed by model + service ID)
• All AI Personas across all models
• All external interactions across all models
• All SOPs across all models

2. Cross-Company Interaction Validation (Errors)

For every externalInteraction in each model:

• Reciprocal endpoint exists: If Model A declares an external interaction with "SpinnyThings", Model B (SpinnyThings) must have a corresponding service or external interaction that references Model A's entity
• Service name alignment: The externalServiceName in Model A matches an actual service name in the target model
• Domain alignment: The externalDomain matches the target model's modelMetadata.organisation

For cross-company interactions in the collaborative model (acid-collaborative.json):

• Source exists in source company model: The sourceId maps to a real service in the source company's individual model
• Target exists in target company model: The targetId maps to a real service in the target company's individual model

3. Agent Authority Consistency (Errors)

When an AI Persona from Company A operates in Company B's node:

• Authority model declared in both models: The persona's authority level in the home model matches or is compatible with the authority level granted in the host model
• Host authority <= Home authority: A persona should never have MORE authority at a host node than at its home node
• SOP access alignment: SOPs the persona expects to invoke at the host node are actually exposed by the host node's model

4. Reference Data Alignment (Warnings)

Check shared reference data consistency:

• Material codes: Material identifiers used in one model map to recognised references in others (e.g., Ti-6Al-4V / AMS 4911 / R56400 consistency)
• Quality standards: AS9100, NADCAP, FAA references are consistent across models
• Company identifiers: Company IDs used in interactions match across all models
• SOP naming: SOPs referenced in cross-company interactions use consistent naming

5. Collaborative Model Reconciliation

Validate the collaborative model (acid-collaborative.json) against individual company models:

• Every company model has a corresponding L0 service in the collaborative model
• L1 services in collaborative model align with L0 services in individual company models
• Cross-company interactions in collaborative model are consistent with external interactions in individual models
• AI Persona deployments shown in collaborative model match persona definitions in individual models
• No orphaned companies: Every company in the collaborative model has an individual model file

6. SOP Cross-Reference (Warnings)

For SOPs that appear in multiple models (e.g., a persona in Model A invokes SOPs exposed by Model B):

• SOP IDs match: Same SOP ID used in both models
• SOP names match: Same human-readable name
• MCP server alignment: The MCP server name in the invoking model matches the server name in the hosting model
• No SOP ownership conflicts: Each SOP has exactly one owning domain (One Server, One Owner doctrine)

7. Supply Chain Topology Validation (Warnings)

• No circular supply chains: Company A supplies B supplies C does not loop back to A
• Tier consistency: If A is Tier 0 and B is Tier 1, interactions flow appropriately (OEM doesn't supply to itself)
• Complete paths: Every material flow has both source and destination
• Demo scenario coverage: The three demo scenarios (Quality Escape, AOG, Schedule Slip) have supporting interactions in the relevant models

8. Report Format

# Cross-Model Validation Report

**Models Validated:** [count]
**Models:** [list of model files]
**Validated:** [date]
**Result:** [PASS / PASS WITH WARNINGS / FAIL]

## Summary
- Cross-model errors: [count]
- Cross-model warnings: [count]
- Models with issues: [list]

## Cross-Company Interaction Errors
| # | Source Model | Target Model | Interaction | Issue |
|---|------------|-------------|-------------|-------|
| 1 | aerovista | spinnythings | ext-003 | No reciprocal endpoint for "Engine Assembly Status" |

## Authority Consistency Errors
| # | Persona | Home Model | Host Model | Issue |
|---|---------|-----------|-----------|-------|
| 1 | ap-quality-agent | aerovista | spinnythings | Authority L3 at home but L1 at host (acceptable) |

## Reference Data Warnings
| # | Data Type | Model A Value | Model B Value | Issue |
|---|-----------|--------------|--------------|-------|
| 1 | Material Code | Ti-64 | R56400 | Different identifier for same material |

## Collaborative Model Reconciliation
| # | Issue Type | Details |
|---|-----------|---------|
| 1 | Missing company model | PowerCell appears in collaborative but has no individual model |

## Cross-Model Entity Index
| Model | Services | Personas | External Interactions | SOPs |
|-------|----------|----------|----------------------|------|
| aerovista | 12 | 3 | 5 | 8 |
| spinnythings | 10 | 2 | 4 | 6 |
| acid-collaborative | 11 | 0 | 0 | 0 |

Relationship to Other Skills

• Upstream: canonical-model-validator (each model should pass individual validation first)
• Parallel: bsa-cross-reference-index (builds navigable index; this skill validates correctness)
• Downstream: bsa-pipeline-orchestrator (includes this in the pipeline), methodology-compliance-check (cross-model issues affect compliance)

Output

A cross-model validation report identifying mismatches between canonical models, with categorised errors (must fix) and warnings (should fix), and a reconciliation check against the collaborative model.