augment-carrier-audit
Free, anonymous pre-tender carrier safety audit for freight brokers. Built for the
post-Montgomery v. Caribe Transport II negligent-hiring era.
Run locally
pnpm install
cp .env.example .env.local
# Fill in FMCSA_WEBKEY at minimum
pnpm dev
Open http://localhost:3000.
Deploy
# First-time: link the project to Vercel
vercel link
# Set required env vars in Vercel
vercel env add FMCSA_WEBKEY production
# (Optionally override thresholds — defaults are baked in)
# Connect Vercel KV (Storage tab → Create → KV → Connect)
# That auto-populates KV_URL / KV_REST_API_URL / KV_REST_API_TOKEN
# Deploy
vercel --prod
Architecture
- Next.js 14 App Router on Vercel Pro (60s function timeout)
- FMCSA QCMobile for live carrier safety data
- Vercel KV caches FMCSA responses 24h (saves 90%+ on repeat lookups)
- No database — submission analytics emitted as structured
console.log JSON,
Vercel forwards to Datadog if the Datadog integration is connected
- No PII — IPs hashed; carrier/load IDs not persisted
Thresholds
Defaults (override via env):
| Metric | Default | Env var |
|---|
| Crash rate per truck (24mo) | 0.20 | THRESHOLD_CRASH_PER_TRUCK |
| Driver OOS rate | 10% | THRESHOLD_DRIVER_OOS |
| Vehicle OOS rate | 40% | THRESHOLD_VEHICLE_OOS |
| Hazmat OOS rate | 5% | THRESHOLD_HAZMAT_OOS |
| Max loads per submission | 100 | MAX_LOADS_PER_SUBMISSION |
Cutoffs derived from FMCSA SMS Methodology v3.0.4 §4.6 (Crash Indicator BASIC P85
framework), rounded from a 1,356-carrier industry sample.
Statistical method
For each safety axis, the carrier's observed rate is bounded by a Wilson 95% CI
(Wilson, E.B. 1927). A flag fires only when the CI lower bound exceeds the cutoff —
small-sample noise (e.g. 1-of-1 inspections) doesn't trigger.
License
MIT
