Disclaimer: Independent product / portfolio work by Kaushal Khodifad. Not a live commercial product.return to portfolio
Disclaimer: Independent product / portfolio work by Kaushal Khodifad.
Step 06 · Context & rigor

Methodology, market & sources

Every number here is sourced inline and fact-audited against a primary source. Where a figure is a vendor claim or an industry estimate, it is badged directional.

The P&L & liability stakes
~$91K
verified
Average large-truck crash cost

An injury crash averages ~$200K; a crash with a fatality averages ~$3.6M (FMCSA 2025 crash-cost methodology).

FMCSA - Crash Cost Methodology (2025 update) & 'Safety is Good Business'

8% / 17%
verified
PSP screening efficacy

Carriers using PSP lowered crash rates ~8% and driver out-of-service rates ~17% vs non-users (not 10%/20%).

FMCSA Pre-Employment Screening Program (PSP) study figures

90-95%
verified
Large-truckload driver turnover (annualized)

The cohort with the highest churn; ~92.7% long-run average for $30M+ carriers.

American Trucking Associations (ATA)

$8.2K-$20.7K
directional
Cost to replace one driver

Recruiting, screening/compliance, orientation/training, plus empty-truck revenue loss; quoted as a range.

Industry estimates (Centerline / ATRI-cited ranges)

~30s
directional
Sentinel CDL→report (reference product)

Spotter AI's Sentinel: a CDL upload pulls MVR/PSP/FMCSA/CDLIS into one report in as little as ~30s; claims up to ~75% screening-cost reduction.

Spotter AI / Sentinel marketing (sentinel.spotter.ai) - vendor claim

The wedge vs the incumbents
PlayerWhat they sellThe open gap
SambaSafetyMVR monitoring + a driver risk score; insurer-trusted data/score incumbent.The score is a black box - not reconstructable point-by-point or tied to a citable CFR.
FoleyDOT compliance, MVR monitoring, Clearinghouse queries, DQF management.Compliance plumbing, not a defensible auto-approve / deny decision with a reason ledger.
Tenstreet (+ DriverReach)Recruiting / onboarding / ATS + DQ-file management (acquired DriverReach, Apr 2025).Optimizes the funnel, not the underwriting decision or its auditability.
Spotter AI · SentinelAI hiring/screening/compliance; consolidated CDL→report in ~30s; up to ~75% cost cut.Fast consolidated data + a safety score; the glass-box, counterfactual, CFR-traced decision layer is the open wedge.
The open wedge · who owns which quadrant
Data pull → Defensible decisionBlack-box → ReconstructableOPEN WEDGESambaSafetyTenstreetFoleySentinelSpotter

Positioning is illustrative. Incumbents cluster on data access; the reconstructable, CFR-traced decision quadrant is the one Spotter targets.

The moat is defensible decisioning + auditability, not faster data access. The FMCSA assigns no score to PSP/MVR data, and carriers remain legally accountable for their own 49 CFR 391 qualification decisions - so a reconstructable, CFR-traced verdict is the product no incumbent owns.

Real FMCSA mechanics reproduced
  • Time weights 3 / 2 / 1 (≤6 / ≤12 / ≤24 mo), >24 mo drops out.
  • The +2 out-of-service severity uplift (legacy scale).
  • The summed severity for one inspection within a BASIC capped at 30, before time-weighting - a capped total attributed worst-first. (v1 of this demo implemented this as a proportional rescale; an external audit caught it. The fix is fuzz-tested against the preserved v1 bug in the Workbench harness.)
  • Both severity scales - legacy 1-10 and the Dec 1, 2025 revised 1-2 with the OOS-vs-Unsafe-Driving carve-out.
  • Hard gates per 49 CFR 382 (Clearinghouse) and 391 (CDLIS status).
Adapted for this demo (disclosed)
  • Carrier-level SMS BASIC mechanics applied at the driver level (the real per-driver analog is PSP events; FMCSA's public Driver SMS is retired).
  • The 0-100 index uses a fixed synthetic exposure proxy for demo stability - not an official FMCSA peer-group percentile. Real SMS produces relative percentiles.
  • The peer band shown is illustrative, to signal we know the real system is relative.
  • All data is synthetic; the decision record is an audit-trail rationale, not a statutory FCRA notice.
Regulatory citations used in the engine
49 CFR 391.11 / 391.15General driver qualifications & disqualification (drives the CDLIS suspended/revoked hard gate).
49 CFR 391.51Driver qualification file (DQF) document requirements - the completeness checklist.
49 CFR 382.501 / 382.701Removal from safety-sensitive functions + Clearinghouse query duty (the prohibited hard gate).
49 CFR 383.51Major/serious disqualifying offenses - the revised-scale Unsafe-Driving carve-out.
49 CFR 395.3Hours-of-service limits (11/14/60-70) referenced in HOS line items & incident triage.
49 CFR 382.303Post-accident drug/alcohol testing windows (incident-triage policy trigger).
15 U.S.C. 1681 (FCRA)Consumer-report adverse-action context on the decision record.
FMCSA SMS MethodologyTime weights 3/2/1, +2 OOS uplift, and the summed severity for one inspection within a BASIC capped at 30 before time-weighting (a capped total - not a proportional rescale).
FMCSA SMS (Dec 1, 2025 overhaul)1-or-2 severity scale with the OOS-vs-Unsafe-Driving / disqualifying carve-out.
Fact-audit ledger · June 2026 (cap re-verified July 2026)8 verified · 1 directional
ClaimSourceStatus
SMS time weights 3 / 2 / 1 (≤6 / ≤12 / ≤24 mo)FMCSA SMS Methodologyverified
+2 out-of-service severity uplift (legacy)FMCSA SMS Methodologyverified
Per-inspection-per-BASIC severity capped at 30 before time-weighting (a capped sum, not a rescale - v1 got this wrong, now corrected)FMCSA SMS Methodologyverified
Dec 1, 2025 1-or-2 severity overhaul (OOS all categories except Unsafe Driving; disqualifying specific to Unsafe Driving)FMCSA SMS methodology & CSA noticesverified
PSP 5-yr crash / 3-yr inspection windowFMCSA PSP sourcesverified
8% crash / 17% OOS PSP screening efficacyFMCSA PSP studyverified
Crash cost ~$91K / ~$200K / ~$3.6MFMCSA 2025 crash-cost methodologyverified
Driver turnover 90-95%American Trucking Associationsverified
Replacement cost + Sentinel ~30s / ~75% claimsVendor / industry estimatesdirectional
Why the checks are worth believing

The highest-trust moment in the product: an external audit caught v1's two “machine-verified” proofs as tautologies. Here is exactly what changed.

v1 · tautologycould never fail

The counterfactual check compared the engine’s delta against a value pre-set to the toggled item’s own points - so the two sides were the same number by construction. An external audit called it out.

expected := item.points ✕ self-referential
now · independentfails on a real bug

Expected values now come from an independent recompute of the written spec - no shared aggregation code - and toggles re-run the whole pipeline before the cap is allocated.

expected := predict.ts(spec) ✓ no shared code
Independent spec recompute
predict.ts, no shared aggregation code
Counterfactual re-run
removal before cap allocation
Fuzz vs broken variants
incl. v1's proportional-cap bug

↑ Independent product by Kaushal Khodifad. Spotter AI is a personal project / portfolio piece in the commercial driver onboarding & fleet safety space. Not a live commercial product. Data is illustrative.

Return to portfolioOpen the live demo