Source: Tier 1 repo evaluation batch (10 CTO must-read docs)

What this source is

The 10 documents a CTO (Reuben Cook) would open during a technical evaluation of the Preston-Test repo. These cover the full evaluation surface: production vs demo distinction, connector status, readiness checklist, module deep-dive, system architecture, API documentation, AI governance, key ADR, enterprise features, and the developer guide.

Total: ~90,000 chars across 10 files (~8,900 lines). All read directly from the canonical Preston-Test repo.

Key claims (new content, grouped by critical evaluation themes)

THEME 1: What’s production vs demo (the trust question)

Per PRODUCTION-VS-DEMO-GUIDE.md (859 lines, April 9 2026):

1. Dual-mode architecture is intentional, not a deficiency: “The presence of demo/fixture modes is NOT a sign of incomplete implementation. It is a production-grade architectural pattern designed for enterprise flexibility.” The guide explicitly reframes demo/fixture as a STRENGTH: enables exploration without API keys, cost-effective demos, air-gapped deployments, graceful degradation, A/B testing.

2. AI providers — all 4 production-ready plus mock/rule-based modes:

   • OpenAI GPT-4o/GPT-4o-mini: ✅ Production + mock simulation
   • Claude model upgraded to claude-sonnet-4-5-20250929 — this is Claude 4.5 Sonnet, not 3.5 as documented in prior sources. Model upgrade.
   • OpenRouter: ✅ Production (multi-model)
   • LMStudio: ✅ Production (always local)
   • Rule-Based Engine: ✅ Always available
   • Mock providers: 🟡 Demo only

3. Connector production status (the honest inventory):

   • NetSuite: ✅ Sandbox-Tested (production-ready)
   • Salesforce: ✅ Production (real API, full CRUD)
   • Business Central: ✅ Production (OData v4, metadata-driven)
   • Oracle: ⚠️ Beta (basic implementation, needs hardening)
   • HubSpot CRM: ✅ Production (Phase 3, January 2026)
   • ShipStation 3PL: ✅ Production (Phase 3, January 2026)
   • 20+ more connectors: 🏗️ Fixture-Based (real API implementation exists, uses fixtures in DEMO_MODE)

   Important nuance: “Fixture-Based” does NOT mean “not implemented.” It means the connector has a real API implementation but exposes test data when DEMO_MODE=1. The 6 “production” connectors have been verified against real API endpoints.

4. Features production status:

   • AI Field Mapping, Integration Wizard, AI Config Dashboard, Enterprise Features, RAG Knowledge Base (7,415 chunks indexed), Help Chat Widget, Telemetry: all ✅ Production Ready
   • Mock Dashboard APIs: 🧪 Demo Only (located in src/routes/mockDashboardAPIs.ts)

Per CONNECTOR_STATUS.md (459 lines, April 9 2026):

5. Definitive connector inventory: 6 Production, 10 Fixture-Based, 2 Beta, 200+ Planned
6. Oracle is Beta, not Production: “Basic Implementation, needs production hardening, comprehensive testing” — this contradicts the canonical 6-connector wording which lists Oracle as one of the 6 production connectors. Oracle’s connector EXISTS but is not production-grade per this doc.
7. 11/11 AI routes have governance pre-checks (100% governance coverage as of October 2025)
8. HubSpot + ShipStation added January 2026 (Phase 3 SuiteCentral Parity)

THEME 2: The Squire-specific evaluation path

Per PRODUCTION_READINESS_CHECKLIST.md (1,256 lines, April 9 2026):

9. Written explicitly for Squire: “This checklist ensures Preston-Test Integration Hub is production-ready for Squire’s SaaS integration advisory services.”
10. 4-phase validation over 2-3 days: Phase 1 (Connection, 1 hour), Phase 2 (AI validation), Phase 3 (Real data with Squire’s data), Phase 4 (Scale testing)
11. Grade: 100/100 (A+) — Ready for Squire testing
12. Specific credential requirements for NetSuite (OAuth 1.0a, account ID, consumer key/secret, token ID/secret) and Business Central (Azure AD tenant, client ID/secret, environment, company GUID)
13. AI provider setup is optional but recommended: OpenAI GPT-4o-mini at $0.001/request, Claude 3.5 Sonnet, LMStudio (free)

THEME 3: Architecture, API, governance, and enterprise features

Per ARCHITECTURE.md (549 lines):

14. Full system architecture: Alpine.js client → Express server → PostgreSQL/SQLite → DI (Inversify) → 3-tier service layer. Source paths and file structure documented.

Per API-DOCUMENTATION.md (1,271 lines):

15. Full REST API: authentication (JWT + API key), rate limiting (100 req/60s), all core endpoints including Context Sidecar CRUD, NL Action Gate propose/approve/execute, MCP native tools, AI field mapping, and predictive analytics APIs. ~1,035 routes total per the endpoint registry.

Per AI-GOVERNANCE-COMPLETE-GUIDE.md (740 lines):

16. 11 AI routes with governance pre-checks: PII detection, content moderation, compliance validation (GDPR, HIPAA, SOX, PCI-DSS), risk assessment, audit trails. Every route listed with its governance coverage.

Per ADR-004-DUAL-AI-SYSTEM-DESIGN.md (288 lines):

17. Formal ADR explaining the dual-mode architecture: System 1 (production AI via /api/ai/proxy/*) and System 2 (demo/fixture via mock providers). Rationale: cost control, zero-infrastructure demos, graceful degradation. Consequences documented.

Per 21-enterprise-features-guide.md (1,081 lines):

18. 6 enterprise features (all shipped):
    • BC OData Metadata Client (metadata-driven schema discovery)
    • Delta Sync Cursor Store (SHA-256 content-addressable sync)
    • Approve-to-Apply Audit (hash verification, audit trail)
    • NetSuite Governance Pacer (3-tier rate limiting)
    • Golden-Set AI Evaluator (regression testing for AI accuracy)
    • X12 EDI Universal Translation (EDI support via the Universal Translation module)

Per SUITECENTRAL-16-MODULES-DEEP-DIVE.md (629 lines):

19. Full technical deep-dive on all 16 modules with dashboard locations, line counts, and feature status. Confirms the 12 core + 4 extension/platform split.

Per DEVELOPER_GUIDE.md (1,782 lines):

20. Comprehensive developer guide: full src/services/ decomposition, DI bindings, Week 1-8 feature map, architecture patterns, testing patterns, extension points for new connectors/modules. This is what a developer hired to extend SuiteCentral 2.0 would read first.

Pages updated by this ingest

Created (2 new pages):

• production-vs-demo — the trust document distilled; what’s real, what’s demo, and why dual-mode is a strength
• squire-readiness-checklist — the Squire-specific 4-phase evaluation path with credential requirements

Updated (5 existing pages):

• production-proof — Claude upgraded to 4.5 Sonnet; Oracle connector is Beta (not production per this doc); 11/11 AI routes have governance
• module-library — cross-link to the 16-module deep-dive
• suitecentral-2-overview — dual-mode architecture as a design principle; 6 enterprise features listed
• cto — cross-link to PRODUCTION_READINESS_CHECKLIST as the verification starting point
• competitive-landscape — Oracle connector is Beta, which slightly weakens the “6 production connectors” claim

Cross-references / contradictions found

• Oracle connector: “Production” vs “Beta”: 26-canonical-metrics-and-wording says “6 production connectors demonstrated (NetSuite, Business Central, Salesforce, HubSpot, ShipStation, Oracle).” But CONNECTOR_STATUS.md says Oracle is “⚠️ Beta (Basic Implementation).” This is a real discrepancy — the canonical wording overstates Oracle’s production readiness. It should be either 5 production + 1 beta, or “6 connectors demonstrated” without the “production” qualifier for Oracle. Flagged for asset owner.
• Claude model upgrade: prior sources said Claude 3.5 Sonnet. PRODUCTION-VS-DEMO-GUIDE.md environment example shows ANTHROPIC_MODEL=claude-sonnet-4-5-20250929 — that’s Claude 4.5 Sonnet (April 2026 release). The canonical provider table should be updated.
• RAG Knowledge Base with 7,415 chunks: NEW fact — the Help Chat system has a real vector store with 7,415 indexed document chunks. Not mentioned in any prior source. Adds to the production-readiness evidence.
• Test count: 9,130 in the readiness checklist — this is an OLDER vintage (October 2025 era). The readiness checklist process is current; the test-count number in it hasn’t been refreshed. Not a concern for the evaluation process itself.
• “20+ more connectors” as fixture-based: the platform has more connectors than the 6 production ones — they all have real API implementations that expose fixtures in demo mode. A reviewer should understand that “fixture-based” means “real code, test data” not “unimplemented.”

Notes

• The PRODUCTION-VS-DEMO-GUIDE is the single most important document for a technical evaluator. It answers the #1 question (“is this real?”) with full transparency: here’s what’s production, here’s what’s demo, here’s why we designed it this way. The honesty of this document is itself a production-readiness signal.
• The PRODUCTION_READINESS_CHECKLIST is the second most important document. It tells Reuben exactly what to do, in order, with specific curl commands and expected responses. 2-3 days to complete. Grade: A+.
• The DEVELOPER_GUIDE at 1,782 lines is the document a new hire would read. Its existence and depth signal engineering maturity — immature codebases don’t have 1,782-line developer guides.
• All 10 documents are dated April 9, 2026 or later — freshest documentation in the corpus.