field-service-readiness.md · Agentic workflows
Field service readiness workbench
Give dispatchers one place to check parts, qualifications and access before a visit, investigate missing evidence and see when an earlier approval becomes stale.
- Next.js
- TypeScript
- Fastify
- PostgreSQL
- pg-boss
- LangGraph
- Claude
- Docker Compose

A booked repair visit can still lack the right part, a valid qualification or confirmed access. The facts sit in different records and can change after the dispatcher reviews them.
This workbench gives the dispatcher one view of those prerequisites, their evidence and their expiry. Its purpose is to make a missing requirement visible before the visit, and to withdraw a local approval when the facts no longer support it.
Editorial archive date: 1 October 2026. Implementation, screenshots and results were reviewed on 24 October and come from later October work.

The packaged demo's visit queue, captured on 23 October. Synthetic records, rules-mode assessment and a fixture clock; no real visits are dispatched.
Follow the wrong-kit case
PR-204 needs KIT-B, but the truck carries KIT-A. The workbench explains the mismatch. A warehouse receipt and reservation make the case ready for review, allowing the dispatcher to approve one exact assessment. A later stock withdrawal invalidates that release and blocks the case again.
An optional Claude investigator reads evidence and proposes a cited brief and follow-up requests. It cannot change readiness or release a visit. An answered request still needs a new source observation and reassessment before it changes the result.
Results and boundaries
The committed evidence reports 4,071 passing assertions across 189 synthetic scenarios, including eleven process-kill recoveries. The later model comparison accepted three of six grounded briefs from each of Sonnet and Haiku, with one attempt per case. Remaining attempts stopped for review; these small tests do not establish field effectiveness. Evidence and evaluation
The application runs locally as one container beside PostgreSQL. It has fixture login and no live operational connector. No dispatcher has validated it, backup and restore remain unproven, and a failed agent investigation starts again rather than resuming. Readiness is an administrative review, not a safety certification or repair guarantee.
Run the public repository, or read the article on expiring readiness.