---
type: source-summary
title: 'Source: 04-TECHNICAL-PROOF.md'
modified: 2026-04-07
tags:
  - technical-proof
  - architecture
  - feature-inventory
  - source
  - preston-test
---

# Source: 04-TECHNICAL-PROOF.md

## What this source is

The **canonical technical proof document** for SuiteCentral 2.0. Dated March 3, 2026 (last verified April 5, 2026) — very fresh. Architecture + test results + feature inventory + AI provider configuration + NetSuite integration proof + infrastructure stack. At ~4,400 chars, it is the single most content-dense technical document in the corpus so far.

This source **resolves multiple open questions simultaneously**:
- The 4 production AI providers are now named with their models and roles
- NL Action Gate is confirmed as shipping v3.3.0 with 6 live actions (resolves the "future innovation" discrepancy from the hook video)
- The Governance Pacer is confirmed as v2.4.0 with 3-tier rate limiting
- Squire's actual NetSuite sandbox ID is named (TSTDRV2698307)
- Every feature in the platform is categorized into Tier 1-4 with shipping status
- Line coverage is 64.59% (vs the 64.48% statement coverage from other sources — different metrics, both correct)

## Key claims

1. **Test results** (canonical breakdown, per [[sources/26-canonical-metrics-and-wording]] mapping): **9,286 unit + 170 integration + 20 E2E = 9,476 executed tests passed (100% executed pass rate, 34 skipped: 23 unit + 11 integration + 0 E2E) across 419 suites = 9,510 total tests, 100% pass rate, 64.59% line coverage**. Last verified April 19, 2026. → [[pages/concepts/production-proof]]
2. **4 production AI providers (NEW — fully named)**:
   - **OpenAI GPT-4o** — Primary inference
   - **Anthropic Claude 3.5 Sonnet** — Secondary/validation
   - **OpenRouter** (multi-model) — Routing/fallback
   - **LMStudio Llama 3.1 8B** — On-premise/fallback
   All four operational. → [[pages/concepts/production-proof]] and [[pages/concepts/suitecentral-2-overview]]
3. **AI accuracy — verified metrics (NEW)**:
   - Field mapping accuracy: **95-99%** (verified Oct 2025) — resolves "95% or 95+%?" ambiguity
   - Confidence calibration: 90%+ (Oct 2025)
   - Multi-provider consensus: +5-15% accuracy boost (Oct 2025)
4. **Architecture stack** (NEW): Frontend = Alpine.js + Tailwind CSS. Backend = Node.js + TypeScript (Express). Database = SQLite (dev) / PostgreSQL (prod). Deployment = Docker + Kubernetes + Terraform. → [[pages/entities/preston-test-repo]]
5. **Key component LOC counts** (NEW): NetSuite Connector (OAuth 1.0 HMAC-SHA256, 500+ LOC), Business Central Connector (OAuth 2.0 + OData, 400+ LOC), AI Field Mapping (multi-provider + RAG, 1,500+ LOC), MDM Engine (Golden Record + Survivorship, 800+ LOC), Context Sidecar (PostMessage + Context Bus, 600+ LOC). → [[pages/entities/preston-test-repo]]
6. **Tier 1 Production-Ready features (6 items, all shipping)**: AI Field Mapping v2.5.0, Context Sidecar v3.2.0, Action Islands v3.2.0, Delta Sync v2.3.0 (SHA-256 chunked, DB-backed), Governance Pacer v2.4.0 (3-tier rate limiting), Approve-to-Apply v2.4.0 (hash verification). → [[pages/concepts/suitecentral-2-overview]]
7. **Tier 2 AI Safety & Compliance (4 items, all shipping)**: Reasoning Trace Engine, Hallucination Detector (multi-method validation), DLP/PII Protection (9 patterns), Active Learning Service (learns from user corrections). → [[pages/concepts/suitecentral-2-overview]] and [[pages/role-briefs/cto]]
8. **Tier 3 Production-Ready Completed March 2026 (5 items, 4 shipping + 1 roadmap)**:
   - Golden Record MDM v3.4.0 — **SHIPPING** (DB persistence D1+D2+D3, survivorship rules) → [[pages/modules/mdm-central]]
   - **NL Action Gate v3.3.0 — SHIPPING** (6 live actions, LLM intent parsing). **Resolves the mild discrepancy from [[sources/ai-governance-layer-video]] which called it "future innovation"** — the base version is shipped, the "future innovation" is deeper integrations beyond the v3.3 baseline. → [[pages/concepts/nl-action-gate]]
   - MCP Gateway + Tenant Policy API v3.3.0 — SHIPPING (base tools + optional gateway tools + authenticated `/api/mcp/policies`) → [[pages/concepts/suitecentral-2-overview]]
   - Schema Drift Shield v3.3.0 — SHIPPING (discovery, caching, drift detection + sync blocking) — this is the fourth differentiator from [[sources/01-executive-summary]] slide 2 that previously had no detail. → [[pages/concepts/suitecentral-2-overview]] (updates the "four differentiators" list)
   - Predictive Ops v3.3.0 — **ROADMAP** (ML models need production training data). **This is the only non-shipping feature in the entire inventory.**
9. **NetSuite sandbox ID (NEW — specific)**: **TSTDRV2698307** — Squire's actual NetSuite sandbox. First concrete Squire-specific infrastructure identifier in the corpus. → [[pages/entities/squire]]
10. **NetSuite CRUD verified** against customer records, vendor records, transaction records, custom record types, saved searches — full Create/Read/Update/Delete/Search. → [[pages/concepts/production-proof]]
11. **The 12 core modules are listed again** here (SupplierCentral through ContractCentral). Confirms [[sources/22-module-library]] from a fourth source. One-liner descriptions per module are shorter than in 22-MODULE-LIBRARY but consistent. → [[pages/concepts/module-library]]
12. **Infrastructure recovery claim**: "Full rebuild from code in 10 minutes." Combined with "Version-controlled infrastructure" and "Multi-AZ high availability" — these are SOC 2 A1 Availability claims. → [[pages/entities/compliance-dashboard]] (when created)

## Pages updated by this ingest

**Created** (0 new pages — all content lands on existing pages).

**Updated** (5 existing pages):
- [[pages/concepts/production-proof]] — added the canonical test breakdown (9,244 + 146 + 20 = 9,410), 9 AI providers with names and models, AI accuracy metrics (95-99% field mapping, 90%+ confidence calibration, +5-15% consensus boost), NetSuite sandbox TSTDRV2698307, full CRUD verification
- [[pages/concepts/suitecentral-2-overview]] — added the 4-tier feature inventory framing; added Schema Drift Shield detail (filling the 4th differentiator slot); added architecture stack
- [[pages/concepts/nl-action-gate]] — **resolved the "future innovation" discrepancy**: v3.3.0 is shipping with 6 live actions. The video's "future innovation" framing was either outdated or referred to deeper post-v3.3 integrations.
- [[pages/entities/squire]] — added the NetSuite sandbox ID TSTDRV2698307 as concrete Squire-specific infrastructure
- [[pages/modules/mdm-central]] — added v3.4.0 version + "DB persistence D1+D2+D3" detail

## Notable quotes

This source is a specification document, not a pitch document. The most notable "quote" is the explicit call-out that **only one feature is on the roadmap** (Predictive Ops) while every other feature across four tiers is shipping:

> "Predictive Ops v3.3.0 — 🔧 Roadmap — ML models need production training data"

Every other row in the tier tables is marked ✅ READY. For a platform that positions itself as "production-ready," a feature inventory where only 1 of ~20 features is roadmap is strong evidence.

## Cross-references / contradictions found

- **NL Action Gate shipping status resolved**: [[sources/ai-governance-layer-video]] 03:54-04:05 called NL Action Gate "future innovation." This source names it as v3.3.0 with 6 live actions, shipping. Resolution: the base NL Action Gate (regex fast-path + LLM fallback + allowlist, per [[sources/narration-scripts]] scene6) is shipped; the "future innovation" framing in the video likely refers to deeper integrations beyond the v3.3 baseline (more LLM providers, more allowlist actions, better confidence tuning). Not a contradiction — just a scope distinction.
- **Coverage metric clarification**: this source says **64.59% line coverage** while [[sources/15-start-here-async-standalone]] says **64.48%** and [[sources/26-canonical-metrics-and-wording]] lists both. They are different metrics (lines vs statements). Both correct. Per canonical metrics: Statements 64.48%, Branches 52.34%, Functions 67.15%, Lines 64.59%.
- **DLP pattern count — CORRECTED 2026-04-07**: this source says "9 patterns (SSN, CC, email, etc.)." `DLPService.ts` has **10 regex patterns**; the compliance dashboard's JavaScript snapshot says **14** (combining DLP + governance + planned patterns). The "9" here likely counts phoneUS + phoneIntl as one "phone" category (10 → 9). See [[pages/concepts/production-proof]] "DLP pattern count — reconciled" for the full table. The "8 patterns" figure that appeared in earlier wiki analysis came from a stale NotebookLM scrape, not the actual dashboard.
- **Confidence calibration metric** (NEW, 90%+): a separate metric from field mapping accuracy. Means the AI's confidence scores are themselves calibrated — when it says "95% confidence," the actual correctness rate matches. This is a sophisticated ML quality claim and not yet reflected anywhere else in the corpus.
- **Multi-provider consensus boost** (NEW, +5-15%): the four AI providers aren't just fallbacks — they vote on ambiguous mappings and the consensus improves accuracy. This is a subtle architectural claim worth more exploration in a future AI-provider ingest.
- **Schema Drift Shield specifics** (NEW): "discovery, caching, drift detection + sync blocking" — this is the architectural implementation of the 4th differentiator from 01-executive-summary slide 2 that had no detail before. Now has a shipping version and a feature list.

## Notes

- The source is labeled as dated **March 3, 2026** in the header but "Last verified April 5, 2026" — so the content is recent and the claims have been re-verified against live state.
- The "Core SuiteCentral Modules" list has short descriptions ("Vendor management", "Payment processing", etc.) that are complementary to [[sources/narration-scripts]]' 4-cue module narrations. Neither is more authoritative — they serve different purposes (pitch short vs demo narration).
- The file path at `src/connectors/NetSuiteConnector.ts` gives future ingests a specific starting point for architectural deep-dives if needed.
- This source is best consumed alongside [[sources/26-canonical-metrics-and-wording]] — they were both ingested in the same turn, and the canonical metrics file is the style guide that dictates how the numbers in this technical proof should be cited in executive-facing materials.