| Document Control | |
|---|---|
| Document Title | Artificial Intelligence Systems Management & Governance SOP |
| Document ID | HWB-QMS-7.7 |
| Version | 1.0.0 |
| Governing Standards | ISO/IEC 42001:2023, ISO 9001:2026 (Clause 7.1.6), SOC 2 CC6.8 |
| Author & Lead Auditor | George (Systems Architect, mbB, Senior ISO 9001 Auditor) |
| Approval Authority | Humberto Dominguez, Chief Executive Officer |
| Effective Date | 10/02/2026 |
Standard Operating Procedure: Artificial Intelligence Governance
Rules and quality safeguards governing autonomous AI agents, commercial takeoff calculations, and lead synchronization under international standard ISO/IEC 42001:2023.
1.0 Purpose
This standard operating procedure guarantees that all artificial intelligence models and autonomous background scripts used by HWB Cleaning Services operate with mathematical accuracy, ethical transparency, zero synthetic data, and absolute human executive oversight. It provides verifiable proof to commercial clients, school superintendents, and general contractors that automated bids and schedules are rigorously controlled.
2.0 Scope
This procedure applies to all autonomous agents, machine learning algorithms, and automated estimating scripts across the HWB ecosystem, including:
- Yamamoto Moto: Lead AI Estimator responsible for architectural blueprint takeoffs, cleanable square footage calculations, production rates, and bid proposal spreadsheets.
- George: Autonomous Systems Architect responsible for codebase integrity, ISO/QMS alignment, and system documentation.
- Silas Sync: Autonomous database and backoffice synchronization daemon.
- Tessa: Automated regression testing agent and platform quality gatekeeper.
- Lead & GIS Daemons: Autonomous Texas daycare and public records lead synchronization scripts.
3.0 Operational Rules & Mandatory Controls (ISO 42001 Annex A)
Control 1: The Human-in-the-Loop Sovereign Gate
Under no circumstances may an AI agent or script automatically send an unreviewed proposal, execute a legally binding contract, or push uninspected production deployments to Azure. Permanent Approval Authority rests solely with CEO Humberto Dominguez. Every takeoff estimate generated by Yamamoto Moto must be staged in the executive operations review queue before dispatch.
Control 2: Blueprint Discrepancy & Heuristic Tagging Mandate
Whenever an architectural blueprint lacks an explicit room finish schedule, contains contradictory lot dimensions, or relies on an estimator assumption:
1. Database records in ConstructionBids must be tagged with [DATA DISCREPANCY: Heuristic Estimate - Field Verification Required].
2. Generated proposal spreadsheets must insert an Amber/Gold ESTIMATOR NOTICE: Data Discrepancy / Calibrated Footprint warning block on Page 1.
3. The empirical delta must be reported in plain words directly to the CEO.
Control 3: Physical Truth & Zero Synthetic Data
Synthetic, fabricated, or "make-belief" data is strictly prohibited across all operational databases and executive dashboards. Every lead, bid record, customer interaction, and square-foot takeoff must represent empirical facts verified through physical walkthroughs, public records, or transactional receipts.
Control 4: Automated Pre-Flight Regression Verification
Before any software release, updating bidding formulas, or modifying automated workflows:
• scripts/yamamoto_bid_test_suite.py must execute and pass 100% of mathematical stress tests.
• scripts/tessa_regression_suite.py must achieve a Grade A+ across all 11 operational modules.
• scripts/test_site_security_rack_09.py must confirm zero security regressions.
Control 5: Immutable WORM Audit Trails
All AI takeoff commits and administrative inspections must trigger forensic records in SecurityAuditLogs under TAKEOFF_COMMITTED or AUDIT_LOG_INSPECTED. These logs are protected by database triggers that reject any modification or deletion, satisfying ISO 27001 Clause A.12.4 and SOC 2 CC6.8.
4.0 Verification & Zero-Defect Audit Checklist
- [ ] Has every automated takeoff proposal been reviewed and confirmed by CEO Humberto Dominguez prior to client delivery?
- [ ] Are architectural discrepancies visibly flagged with amber warning banners on exported Excel proposals?
- [ ] Have all automated test suites (Yamamoto, Tessa, Security) achieved 100% passing scores?
- [ ] Are all AI actions logged to PostgreSQL
SecurityAuditLogswith unalterable WORM timestamps? - [ ] Has the SQL Brain verified 0 synthetic data rows in the live production database?