Safeguard Work-Order Agent Ecosystem
Four AI agents (classify → route → validate → action) process work orders end-to-end, reducing human-in-the-loop to exception handling only. Go/gRPC + Oracle/Postgres tie-ins are contract-first disclosed seams.
At a glance
The four agents
1 · Classifier
Trade category, priority, and field extraction with calibrated confidence. Deterministic lexicon engine; LLM engine is a disclosed seam.
2 · Router
Table-driven routing to queue, crew, vendor tier, region, and SLA — inspectable business policy.
3 · Validator
The safety boundary: required fields, confidence floors, cost cap, duplicate detection. Anything uncertain goes to a human.
4 · Actioner
Auto-dispatch vs. human exception. Idempotent gRPC dispatch, append-only audit, no silent drops.
Deliverables
Overview
What it is and how to run it
Architecture
Pipeline, principles, module map
Outcome Contract
Scope, seams, success criteria
Spec · Classifier
Classifier agent contract
Spec · Router
Router agent contract
Spec · Validator
Validator (safety boundary)
Spec · Actioner
Actioner + dispatch contract
Routing Policy
Queues, SLAs, regions
Verification Report
17/17 checks + benchmark
Proof Report
Certification basis + seams
Executive Evidence
The 10 IRS_AUDITOR questions
Limitations
Disclosed seams
Auditor Objections
Hostile objections answered
Auditor Challenge
Generated interrogation
Evidence Grade
Grade + basis
Reproduce
Exact commands
Verify
What each check asserts
User Guide
Run batches, single orders, console
Run & Deploy
Local run + live integration
Release Notes
v1.0.0 summary
Honest scope
Verified on a synthetic, seeded corpus of 600 work orders: classification 98.9%, exception recall 100%, false-auto-action 0%, and 62.2% actioned automatically (human-in-the-loop reduced to 30.8%). The Go/gRPC dispatch and Oracle/Postgres persistence are contract-first disclosed seams (simulated), and the LLM classifier engine is a documented seam. No real Safeguard data was processed. See the proof report and limitations.
Delivery metrics
Tokens, elapsed time, and cost for producing this outcome. Basis: Estimated — reproducible model over this outcome’s published artifacts. metrics.json
Cost and tokens are estimates derived deterministically from published artifacts and representative list pricing; actual billing may differ. Model basis: claude-sonnet-class (representative).