User Guide — Forge PM Work Order Simulation

FICTIONAL / SYNTHETIC DEPLOYMENT MODEL. No real customer data is used.

Prerequisites

• Node.js >= 20.
• The sibling package ../work-order-agents with its dependencies installed

(@electric-sql/pglite, @grpc/grpc-js, @grpc/proto-loader). This simulation imports the ecosystem from there and reuses its node_modules at runtime.

Run the full simulation

node verify.mjs

This will:

1. Generate 100,000 deterministic synthetic work orders (seed 20260625) and write

the full corpus to data/enterprise-work-orders.jsonl (git-ignored, ~60 MB) plus a 1,000-row sample to datasets/sample-1000.jsonl.

1. Boot the real ecosystem (PGlite Postgres + gRPC dispatch).
2. Process every order through classify → route → validate → action.
3. Write verification-report.json, verification-report.md, and

evidence/{simulation-results.json, roi.json, audit-trace-sample.json}.

1. Print a 25-check MUST_PASS summary and exit non-zero if any check fails.

A full run takes roughly 15–20 minutes (the ecosystem performs real SQL and a real gRPC round trip per order). For a fast iteration loop, scale it down:

WO_SIM_N=2500 node verify.mjs

The synthetic mix is proportional, so small runs approximate the full-scale numbers.

Build the reports

node build-deliverables.mjs    # regenerates report/impact-study.md + certification-report.md

These read verification-report.json so the narrative always matches the latest run.

Publish the case-study page

node publish.mjs

Generates ../../published/forge-pm-work-order-sim/ (landing page, rendered docs, downloadable package, and copied Proof Layer decision).

Where the numbers come from

Interpreting the certification

The Proof Layer assigns the authoritative state. For this outcome it is not PRODUCTION_VALIDATED by design (fictional customer + synthetic data). The simulation-tier label FICTIONAL_DEPLOYMENT_MODEL_CERTIFIED means the simulation is reproducible and internally validated — it is not a claim about a real deployment.