Work Order QMS Module — Integration & API Strategy

Classification: Internal — Architecture & Engineering
Date: 2026-02-13
Artifact: 67 of WO System Series
Status: Proposed
Source Artifacts: 12-sdd.md §External Systems, 13-tdd.md §1 APIs, 14-c4-architecture.md §C1/C2, 25-agent-orchestration-spec.md, 27-coditect-impact.md

---

1. Integration Tier Classification

Not all integrations are equal. Each external system connecting to the WO module is classified by strategic importance, which determines investment level, SLA requirements, and ownership model.

1.1 Tier Matrix

Tier	Classification	Coupling	SLA	Ownership	WO System Examples
T1: Core	Platform-defining, revenue-critical	Bidirectional, real-time, deep	99.99%	Build + own	PostgreSQL (state store), NATS (event bus), AI model providers (Claude/OpenAI)
T2: Strategic	Competitive advantage, customer-facing	Bidirectional, near-real-time	99.9%	Build adapter + maintain	EHR/LIMS (clinical data), CMMS (Maximo/SAP PM), IdP (Auth0/Okta/Azure AD), Vault (secrets)
T3: Standard	Expected capability, replaceable	Unidirectional or webhook	99.5%	Adapter pattern, community	Slack/Teams notifications, SIEM (Splunk/Sentinel), Email (SMTP), Calendaring
T4: Commodity	Undifferentiated utility	Configuration-only	Best effort	Config, not code	SMS (Twilio), File storage (GCS), CDN

1.2 Classification Decision Criteria

For each external system, ask:

1. Revenue dependency: If this integration breaks, do we lose customers?
   YES → T1 or T2

2. Compliance dependency: Is this required for regulatory compliance?
   YES → Minimum T2

3. Replaceability: Can we swap providers in < 1 sprint?
   YES → T3 or T4

4. Competitive moat: Does this integration create switching costs?
   YES → T1 or T2

5. Data flow: Does regulated data (L3/L4) cross this boundary?
   YES → Minimum T2 (per 63-data-architecture.md §1)

1.3 Current Integration Inventory

System	Tier	Direction	Data Classification	Protocol	Status
PostgreSQL	T1	Bidirectional	L4 (all WO data)	SQL/Wire	✅ Implemented
NATS	T1	Bidirectional	L2–L4 (events)	NATS protocol	✅ Implemented
Anthropic Claude	T1	Outbound (API calls)	L2 (prompts, no PHI)	HTTPS/REST	✅ Implemented
OpenAI (fallback)	T1	Outbound (API calls)	L2 (prompts, no PHI)	HTTPS/REST	✅ Implemented
Auth0/Okta (IdP)	T2	Bidirectional (OIDC)	L3 (user identity)	OIDC/SAML	🔧 Architecture ready
HashiCorp Vault	T2	Bidirectional	L3 (secrets)	HTTPS/REST	⚠️ Gap G01
ServiceNow	T2	Bidirectional (sync)	L2–L3 (tickets, assets)	REST API	🔧 Adapter designed
Maximo / SAP PM	T2	Bidirectional (sync)	L2 (asset data)	REST/OData	📋 Planned V2
EHR/LIMS	T2	Read-only import	L4 (regulated records)	HL7 FHIR / REST	📋 Planned V2
Slack/Teams	T3	Outbound (notifications)	L1 (notification text)	Webhook	🔧 Plugin ready
SIEM (Splunk/Sentinel)	T3	Outbound (streaming)	L2 (security events)	Syslog/CEF	⚠️ Gap G13
Email (SMTP/SendGrid)	T4	Outbound	L1 (notifications)	SMTP	🔧 Config ready

---

2. API Design Philosophy

2.1 Principles

api_principles:
  style: REST (JSON:API for external, internal events via NATS)
  versioning: URL path (/api/v1/, /api/v2/) — explicit, discoverable
  authentication: OAuth 2.0 + JWT (external); mTLS (service-to-service)
  pagination: Cursor-based (opaque cursor, not offset) — consistent under writes
  filtering: Query parameter expressions (field[op]=value)
  sorting: sort=field:asc,field2:desc
  rate_limiting: Token bucket per tenant (burst + sustained)
  idempotency: Idempotency-Key header on all POST/PATCH
  error_format: RFC 7807 Problem Details
  content_type: application/json (default), application/vnd.coditect.v1+json (versioned)

2.2 Error Response Format (RFC 7807)

{
  "type": "https://api.coditect.ai/errors/wo/invalid-transition",
  "title": "Invalid State Transition",
  "status": 422,
  "detail": "Cannot transition WO-2024-0042 from DRAFT to COMPLETED. Required path: DRAFT → PLANNED → APPROVED → ASSIGNED → IN_PROGRESS → COMPLETED.",
  "instance": "/api/v1/work-orders/wo-2024-0042/status",
  "extensions": {
    "current_state": "DRAFT",
    "requested_state": "COMPLETED",
    "valid_transitions": ["PLANNED"],
    "correlation_id": "req-a1b2c3d4",
    "compliance_rule": "STATE_MACHINE_GUARD_T1"
  }
}

Every error includes: machine-parseable type URI, human-readable detail, correlation_id for support, and contextual extensions enabling clients to show actionable recovery.

2.3 Pagination

# Cursor-based (preferred — stable under concurrent writes)
GET /api/v1/work-orders?limit=25&cursor=eyJpZCI6MTIzfQ

Response:
{
  "data": [...],
  "pagination": {
    "cursor": "eyJpZCI6MTQ4fQ",      // Opaque, base64-encoded
    "has_more": true,
    "total_count": 1247                // Only when ?include_count=true (expensive)
  }
}

Offset pagination is explicitly not supported — it breaks under concurrent inserts/deletes and creates compliance risks when audit trail records shift between pages.

2.4 Rate Limiting

Tier	Requests/sec (sustained)	Burst	Concurrent	Response on Limit
Starter	10 req/s	30	5	429 + Retry-After header
Professional	50 req/s	150	25	429 + Retry-After header
Enterprise	200 req/s	600	100	429 + Retry-After header
Agent (internal)	500 req/s	1500	200	Circuit breaker + backpressure

Rate limit headers on every response:

X-RateLimit-Limit: 50
X-RateLimit-Remaining: 42
X-RateLimit-Reset: 1708300800

2.5 Idempotency

All mutating operations accept Idempotency-Key header:

POST /api/v1/work-orders
Idempotency-Key: client-generated-uuid-v4
Content-Type: application/json

{ "title": "Upgrade HPLC Firmware", ... }

Behavior: first request processes normally; subsequent requests with same key within 24 hours return the original response without re-execution. Idempotency keys stored in Redis with 24-hour TTL. Critical for agent retry patterns and network failure recovery.

---

3. API Versioning & Lifecycle

3.1 Versioning Strategy

Level	Mechanism	Scope	Example
Major	URL path	Breaking changes	/api/v1/ → /api/v2/
Minor	Request header	Additive changes	API-Version: 2026-03-01

Breaking change definition for the WO API:

• Removing a field from response body
• Changing a field's type or validation rules
• Removing an endpoint
• Changing authentication requirements
• Modifying state machine transitions (compliance impact)

Non-breaking (always safe):

• Adding optional request field
• Adding response field
• Adding new endpoint
• Adding new enum value to non-exhaustive enums
• Adding new event type

3.2 Lifecycle Stages

ALPHA → BETA → GA → DEPRECATED → SUNSET
  ↓       ↓      ↓        ↓          ↓
 Internal  Partner  All    12-month   410 Gone
 only      access   SLA    warning    + migration
                    full   Sunset:    guide
                           header

Stage	Stability	SLA	Documentation	Support
Alpha	May break without notice	None	Internal only	Engineering only
Beta	Additive changes only	Best-effort	Public with beta banner	Standard
GA	Versioned, backward compatible	Full tier SLA	Full reference docs + examples	Full tier support
Deprecated	No new features, bug fixes only	Maintained	Migration guide published	Migration support
Sunset	Endpoint returns 410 Gone	None	Archived	None

3.3 Deprecation Policy

• Minimum notice: 12 months from deprecation announcement to sunset
• Communication: Sunset header on every response, documentation banner, email to all API consumers, in-app notification for dashboard users
• Migration support: Published migration guide with before/after examples, SDK migration helpers, automated compatibility checker
• Compliance constraint: State transitions and audit trail endpoints may never be sunset — they can only be superseded by newer versions that maintain full audit trail continuity

HTTP/1.1 200 OK
Sunset: Sat, 01 Mar 2028 00:00:00 GMT
Deprecation: true
Link: <https://docs.coditect.ai/migration/v1-to-v2>; rel="successor-version"

---

4. Webhook & Outbound Event Architecture

4.1 Event Catalog

Event Type	Trigger	Payload Summary	Compliance Relevance
wo.created	New WO created	WO ID, type, creator, regulatory flag	Audit trail start
wo.transition	State machine transition	WO ID, from_state, to_state, actor, guards passed	Core audit event
wo.approval.requested	Approval gate reached	WO ID, required approvers, deadline	§11.10(g) checkpoint
wo.approval.completed	Approval submitted	WO ID, approver, decision, signature hash	§11.50 signature event
wo.blocked	WO enters BLOCKED state	WO ID, reason, blocking dependencies	Escalation trigger
wo.completed	WO reaches COMPLETED	WO ID, duration, all approvals summary	Lifecycle completion
wo.compliance.violation	Guard function rejects action	WO ID, violation type, rule, actor	§11.10 violation event
wo.agent.dispatched	Agent assigned to WO task	WO ID, agent type, model, token budget	Agent audit
wo.agent.completed	Agent finishes WO task	WO ID, agent type, result, tokens consumed	Agent audit
wo.dependency.resolved	Blocking dependency cleared	WO ID, dependency WO ID, resolution type	DAG progression

4.2 Webhook Envelope

interface WebhookPayload {
  id: string;                    // UUID v7 (time-ordered)
  type: string;                  // e.g., 'wo.transition'
  version: '1.0';                // Event schema version
  timestamp: string;             // ISO 8601 UTC
  tenant_id: string;
  source: 'wo-lifecycle-engine' | 'compliance-engine' | 'agent-orchestrator';
  data: Record<string, unknown>; // Event-specific payload
  metadata: {
    correlation_id: string;      // Request trace ID
    causation_id: string;        // ID of event that caused this event
    actor_id: string;            // Person ID or agent session ID
    actor_type: 'PERSON' | 'AGENT' | 'SYSTEM';
    wo_id: string;               // Always present for WO events
    regulatory: boolean;         // Whether this WO is regulatory
  };
}

4.3 Delivery Guarantees

Guarantee	Implementation
At-least-once	Retry on failure; receivers MUST be idempotent (use id for dedup)
Ordered per WO	Events for same wo_id delivered in sequence; cross-WO ordering not guaranteed
Signed	X-Coditect-Signature: sha256=HMAC(webhook_secret, raw_body)
Retry policy	1s, 2s, 4s, 8s, 16s, 32s, 64s, 128s, 256s, 512s (10 retries, ~17 min)
Dead letter	After max retries: stored in tenant's dead letter queue for 30 days; manual replay available
Timeout	10 second response timeout; 2xx = success, anything else = retry

4.4 Webhook Management API

POST   /api/v1/webhooks              # Register webhook endpoint
GET    /api/v1/webhooks              # List registered webhooks
PATCH  /api/v1/webhooks/:id          # Update (URL, events, active status)
DELETE /api/v1/webhooks/:id          # Deactivate webhook
GET    /api/v1/webhooks/:id/deliveries     # Delivery history (30 days)
POST   /api/v1/webhooks/:id/deliveries/:did/retry  # Replay failed delivery
POST   /api/v1/webhooks/:id/test     # Send test event

---

5. Plugin & Extension Architecture

5.1 Extension Point Registry

The WO system exposes five extension categories, each with defined boundaries and security constraints:

Category	Extension Point	Mechanism	Sandboxing	Audit
Custom Fields	Additional WO/JobPlan metadata	JSONB + tenant JSON Schema	Schema validation	Field changes in audit trail
Approval Chains	Custom approval logic and routing	Tenant config rules engine	Rule evaluation sandbox	All rule evaluations logged
Job Plan Templates	Reusable task definitions	Template library (JSONB)	Schema validation + param bounds	Template usage tracked
Notification Channels	Custom delivery (Slack, PagerDuty, etc.)	Event subscriber plugins (NATS)	Outbound-only; no data read access	Delivery attempts logged
Resource Matching	Custom prioritization algorithms	Strategy pattern interface	Read-only access to resource data	Match decisions logged

5.2 Plugin Security Model

Plugin Permissions:
  READ:    Can read specific entity types (declared in manifest)
  EVENT:   Can subscribe to specific event types (declared in manifest)
  NOTIFY:  Can send outbound notifications (scoped endpoints)
  
  NEVER:   Cannot write to regulated entities (WO, Approval, AuditTrail)
  NEVER:   Cannot bypass compliance guards
  NEVER:   Cannot access other tenants' data
  NEVER:   Cannot modify state machine transitions

Plugin Lifecycle:
  INSTALL → REVIEW (tenant admin) → ACTIVATE → MONITOR → DEACTIVATE
  
  All plugin actions logged in tenant audit trail
  Plugin failures isolated by circuit breaker (per 64-security-architecture.md §5)
  Plugin resource consumption metered (API calls, compute time)

5.3 Future: Marketplace Architecture (V3)

Planned for product roadmap Phase 3 (per 11-product-roadmap.md):

Extension Marketplace:
  ├── Compliance Packs (FDA, ISO 13485, ICH Q10)
  ├── Integration Connectors (Veeva, SAP, ServiceNow)
  ├── Report Templates (audit reports, compliance dashboards)
  ├── Agent Skill Packs (specialized AI agent capabilities)
  └── Industry Templates (pharma, meddev, CRO, fintech)
  
Revenue model: 70/30 split (developer/platform)
Review: Security audit required before listing
Isolation: Each plugin runs in isolated WASM sandbox or container

---

6. Integration Adapter Architecture

6.1 Adapter Pattern

All T2 and T3 integrations use a common adapter pattern to ensure consistency:

interface IntegrationAdapter<TConfig, TInbound, TOutbound> {
  readonly tier: 'T1' | 'T2' | 'T3' | 'T4';
  readonly name: string;
  readonly version: string;
  
  // Lifecycle
  initialize(config: TConfig): Promise<void>;
  healthCheck(): Promise<HealthStatus>;
  shutdown(): Promise<void>;
  
  // Inbound (external → WO system)
  transformInbound(external: TInbound): Promise<WOEntityPartial>;
  validateInbound(data: WOEntityPartial): ValidationResult;
  
  // Outbound (WO system → external)
  transformOutbound(internal: WOEntity): Promise<TOutbound>;
  send(data: TOutbound): Promise<DeliveryResult>;
  
  // Error handling
  handleError(error: IntegrationError): Promise<ErrorResolution>;
  getRetryPolicy(): RetryPolicy;
  
  // Observability
  getMetrics(): IntegrationMetrics;
}

// Example: ServiceNow adapter
class ServiceNowAdapter implements IntegrationAdapter<
  ServiceNowConfig,
  ServiceNowIncident,
  ServiceNowChangeRequest
> {
  readonly tier = 'T2';
  readonly name = 'servicenow';
  readonly version = '1.0.0';
  // ... implementation
}

6.2 Data Mapping Rules

Rule	Rationale	Enforcement
WO system is source of truth for WO state	Prevent external systems from corrupting state machine	Inbound writes rejected if state mismatch
External IDs stored as references, never as primary keys	Decouple from external system ID schemes	external_ref field on relevant entities
Data classification preserved across boundary	L4 data must maintain controls when synced	Adapter validates classification before send
Conflict resolution: last-write-wins with audit	Simple, auditable, deterministic	Version field + audit trail entry
Sync failures never block WO transitions	WO lifecycle is primary; sync is secondary	Async sync with retry queue

6.3 Sync Patterns

Pattern	Use Case	Implementation	Example
Event-driven push	Real-time outbound notifications	NATS → Adapter → External API	WO status → ServiceNow ticket update
Polling pull	Import external data on schedule	Cron → Adapter → WO API	Asset inventory sync from Maximo
Webhook receive	Real-time inbound notifications	External → API Gateway → Adapter → NATS	ServiceNow closure → WO status check
Batch sync	Bulk data reconciliation	Scheduled job → Adapter → Diff → Apply	Nightly asset reconciliation

---

7. Migration Playbooks

7.1 Migration from MasterControl QMS

Phase	Activities	Duration	Success Criteria
Assessment	Map MasterControl change control → WO model; identify custom fields; inventory approval chains; count validated systems	2 weeks	Gap analysis complete, data mapping doc approved
Data Migration	Export change records → transform → import as WOs (historical, read-only); map users → Person entities; map equipment → Asset entities	4 weeks	100% records migrated, validation checksums match
Parallel Run	Both systems active for new changes; CODITECT gets copy of all new change requests; compare outcomes	4–8 weeks	95% outcome parity on parallel changes
Cutover	Freeze MasterControl writes; final sync; redirect users; DNS/SSO update	1 day	All users on CODITECT, no new MasterControl entries
Decommission	MasterControl to read-only; archive after retention period; update validation docs	4 weeks	MasterControl access revoked, data archived

Compliance considerations: FDA-regulated migration requires a validation protocol (VP), execution evidence, and a deviation report for any discrepancies. The migration itself is a change that requires a WO in the new system.

7.2 Migration from ServiceNow ITSM

Phase	Key Differences from MasterControl	Duration
Assessment	ServiceNow data model is ticket-centric (not WO-centric); map incident/change/task → WO hierarchy	2 weeks
Data Migration	ServiceNow has rich CMDB; map CI → Asset; map assignment groups → Teams	3 weeks
Parallel Run	ServiceNow likely stays for non-QMS tickets; integration adapter handles coexistence	6–12 weeks
Cutover	Only QMS-related change control moves; ServiceNow remains for IT operations	1 day

7.3 Greenfield (No Existing QMS)

Phase	Activities	Duration
Discovery	Inventory validated systems, current paper/spreadsheet processes, approval chains	1 week
Configuration	Set up tenant, roles, approval chains, asset registry, job plan templates	2 weeks
Validation	Execute IQ/OQ with representative WO scenarios	2 weeks
Go-Live	First production WO, monitored closely	1 week
Optimization	PM automation setup, agent training, vendor portal activation	4 weeks

7.4 Data Migration Toolkit

interface MigrationJob {
  id: string;
  source: 'mastercontrol' | 'servicenow' | 'veeva' | 'csv' | 'custom';
  target_tenant_id: string;
  status: 'PENDING' | 'VALIDATING' | 'MIGRATING' | 'VERIFYING' | 'COMPLETE' | 'FAILED';
  config: {
    field_mapping: Record<string, string>;      // source_field → wo_field
    value_transforms: Record<string, Transform>; // field → transformation function
    skip_rules: SkipRule[];                      // conditions to skip records
    dedup_strategy: 'SKIP' | 'UPDATE' | 'VERSION';
  };
  progress: {
    total_records: number;
    processed: number;
    succeeded: number;
    failed: number;
    skipped: number;
  };
  validation: {
    checksum_match: boolean;
    record_count_match: boolean;
    field_coverage: number;       // % of source fields mapped
    compliance_verified: boolean; // audit trail continuity verified
  };
}

---

8. Cross-Reference to Other Artifacts

Topic	Primary Artifact	Integration Strategy Section
API endpoint list	13-tdd.md §1.1	§2 (philosophy layer on top of endpoint list)
Event topics	13-tdd.md §1.2	§4 (webhook delivery on top of internal events)
Extension points	13-tdd.md §1.3	§5 (security model + marketplace vision)
External system context	12-sdd.md §1, 14-c4-architecture.md §C1	§1 (tier classification)
Agent message contracts	26-agent-message-contracts.md	§4 (agent events in webhook catalog)
Security boundaries	64-security-architecture.md §5	§5.2 (plugin security)
Data classification at boundary	63-data-architecture.md §1	§6.2 (classification preserved across boundary)
Competitive migration	08-competitive-moat-analysis.md	§7 (migration playbooks per competitor)

---

Integration strategy is not a technical appendix — it's a business capability. Every new customer asks: "Does it work with our existing systems?" Every retention conversation includes: "How hard would it be to leave?" The answers to both questions live in this document.

---

Copyright 2026 AZ1.AI Inc. All rights reserved. Developer: Hal Casteel, CEO/CTO Product: CODITECT-BIO-QMS | Part of the CODITECT Product Suite Classification: Internal - Confidential