NAICS injury-rate outliers, state concentration, year-over-year anomalies.
wc-engine ingests BLS Survey of Occupational Injuries reference rates and runs three deterministic checks per engagement. Same kernel that ships every other JIL vertical: customer-profile gated on lob = 'workers_compensation_carrier', sealed CREB on every finding, FRE 902(14) self-authentication. One kernel, 8 industry verticals, 175 production checks, 273 production services.
What this POC shows.
If you're a workers' comp insurer SIU, a state DOI fraud bureau, or a self-insured employer, this is the short answer for what's being detected on BLS-shaped industry injury data.
What's the dataset?
BLS-shaped industry injury rate data, plus deterministic outlier rules. 1.8K records modeled. Public, free, reproducible.
What did JIL find?
207 findings: NAICS injury-rate outliers (employers reporting injury rates inconsistent with their industry baseline), state concentration (geographic clustering of high-cost claims), year-over-year anomalies (sudden injury-rate jumps without operational change).
Why does this matter?
WC fraud (employer mis-reporting + claimant fraud) is a $30B+ annual loss line. Public industry data shows the patterns; nobody runs systematic outlier detection across the full employer universe. JIL does.
What this is NOT
Not a fraud determination. Not a coverage decision. 'Flagged' = 'industry data shows an anomaly worth review.' The SIU still owns the human judgment + investigation.
How do I run this on my book?
If you're an insurer, we run the catalog over your loss-run alongside the BLS public surface to surface employer outliers + claim patterns. Turnaround: 5-10 days.
What wc-engine fires on.
NAICS-by-year cohort, 2.0x national mean.
NAICS-by-year cohort whose incidence_rate exceeds 2.0x the national mean for that year. Material indicator of carrier-level loss-ratio exposure and OSHA inspection priority. Regulatory basis: BLS SOII methodology, OSHA 29 CFR 1904, NCCI Class Code framework.
Single state holds 60%+ of cohort DAFW.
NAICS cohort where a single state accounts for 60%+ of days-away-from-work cases nationwide in a given year. Signals geographic-cluster underwriting risk or fraud-ring concentration. Regulatory basis: NCCI Experience Rating Plan, state WC statutes.
Year-over-year incidence change beyond +/- 50%.
NAICS-state cohorts with year-over-year incidence_rate change exceeding +/-50%. Bidirectional: large drops trigger underreporting investigations; large spikes trigger experience-modification review. Regulatory basis: BLS SOII, OSHA 29 CFR 1904.
What the WC carrier takes to NCCI or the state DOI.
finding_id : e2a8f0c1-...-wc-naics-outlier-001 check_id : wc_naics_outlier subject_type : naics_year subject_id : 56|2023 severity : high incidence_rate : 6.76 per 100 FTE national_mean : 2.65 per 100 FTE ratio : 2.55x peer cohort regulatory_basis: BLS SOII, OSHA 29 CFR 1904, NCCI Class Code code_version : wc-engine@2026.05.01-wc-1
Deterministic, replayable, court-defensible.
Same kernel as the other 7 verticals. NAICS-outlier is a SQL aggregate against the per-year national mean; state-concentration is a per-NAICS DAFW share computation; YoY-anomaly self-joins prior-year rates. No stochastic LLM in the verdict path. Ava (the in-house agentic AI) groups, narrates, and routes findings; it never produces the underlying flag.
One kernel. Eight industries. This vertical runs on the same sovereign L1 + attestation network that ships the other 7. Kernel age: 18+ months. Adding a vertical: ~1 week. Competitor moat: build the kernel first.