Outcome-as-a-Service · PROTOTYPE · CERTIFIED

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.

Verification 100% (23 / 23) Evidence Grade B Trust 80/100 External Postgres + Go gRPC v1.0.0

At a glance

62.2%actioned automatically
100%exception recall (safety)
0%false auto-action rate

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

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

~269.9k
Tokens used
~1h 29m
Elapsed time
~$1.21
Cost (USD)

Cost and tokens are estimates derived deterministically from published artifacts and representative list pricing; actual billing may differ. Model basis: claude-sonnet-class (representative).