FactVerse AI Agent Validation and Handover

Use this checklist before a FactVerse AI Agent workflow moves from pilot to regular use. Validation should confirm the technical connection, source-data readiness, output quality, approval path, audit record, and operator handover.

Acceptance gates

Gate	Required evidence
Access	Endpoint, API key policy, scopes, and key rotation owner are documented.
Tool discovery	The client can list tools at runtime and see the expected workflow tools.
Source readiness	Assets, data, documents, work records, scenes, or simulation inputs are mapped and timestamped.
Output quality	Sample outputs include evidence, assumptions, limitations, and review state.
Approval	Draft actions, write actions, compute jobs, and scenario records have responsible reviewers.
Audit	Tool calls, source references, decisions, and final feedback can be traced.
Handover	Operators or engineers know how to run, review, pause, and escalate the workflow.

Technical validation

Run these checks for each environment:

1. Connect to the expected host over HTTPS.
2. Initialize the MCP session with X-API-Key.
3. List tools and save the visible tool set for the test key.
4. Call a read-only tool for the workflow boundary.
5. Confirm expected scopes and blocked scopes behave as designed.
6. Test missing or stale source data behavior.
7. Test approval flow for draft writes or compute jobs.
8. Confirm .ai internal editing surfaces remain noindex when applicable.

Workflow validation

Workflow	Validation focus	Sample acceptance output
Facility operations	Asset mapping, alarm history, inspections, open work orders, and operator approval	A reviewed asset status summary with source timestamps and a draft inspection task.
Predictive maintenance	Signal quality, anomaly window, maintenance history, and engineer review	A health explanation with evidence, confidence limits, and a reviewed follow-up plan.
Physical AI	Scene version, SimReady metadata, simulation assumptions, and engineering validation	A scenario package with readiness status, repair items, and accepted assumptions.

Audit record

Each accepted workflow run should retain:

• request ID or task ID;
• user or client identity;
• endpoint and scopes used;
• asset, facility, equipment, or scene boundary;
• tools called and source references returned;
• output version and review state;
• reviewer identity, decision, and timestamp;
• final work order, inspection, scenario, or validation record.

Handover package

Prepare this package for the operating team:

Item	Description
Workflow owner	Business owner, technical owner, and review owner.
Run instructions	How to start the workflow, choose a boundary, and interpret output.
Review guide	How to approve, revise, reject, or escalate an output.
Failure guide	Links to troubleshooting, known source-data gaps, and escalation contacts.
Change process	How endpoint, scope, source system, prompt, or workflow changes are reviewed.
Feedback process	How completed work, false positives, repaired data, or field findings return to the knowledge and data layers.

Go-live decision

A workflow is ready for regular use when:

• read-only behavior is stable for the target boundary;
• source references are understandable to reviewers;
• missing-data and blocked-action behavior is clear;
• approval and audit records are captured;
• operating owners have accepted the handover package;
• any remaining limitations are documented with an owner and review date.