Proof

How a Verdict is built, audited, and reproducible.

You wouldn’t hand your car to a mechanic who couldn’t show their work. The same standard goes the other way. Here’s how every Verdict is constructed — the data sources, the math, the audit pass, and the receipt you can hand to anyone for a second opinion.

Audit trail per Verdictengine v3.8.14prompt Engine version 41

NHTSA-anchored

Every Verdict pulls live data from NHTSA's vPIC, recalls, complaints, and TSB endpoints — keyed off your VIN, not someone else's anecdote. No forum scraping, no generic averages.

You can verify any TSB / recall number from the Verdict directly at nhtsa.gov.

730+ failure patterns

Platform-specific failure-mode database covering 730+ documented patterns across the makes and platforms most car owners deal with. Subaru CVT shudder, F-150 EcoBoost intake carbon, Honda V6 timing-tensioner — each pattern carries mileage band, repair cost range, and the diagnostic tells that distinguish it.

When a finding cites a pattern, it carries the specific make/model/engine + mileage band it applies to.

Math, not vibes

The 0–100 score is computed from a 13-stage deterministic pipeline: VIN decode, mileage-band clustering, Weibull hazard math per system, documented mitigation crediting, regional cost adjustment. Same inputs always produce the same number.

The audit_trail JSON stored with every Verdict shows every stage’s input, output, and the deltas that moved the score.

Independent pass

After the prose draft is written, an independent 6-lens audit pass re-reads it: factual grounding (do TSB / recall citations resolve?), anti-punt (any vague hedge?), mechanical reasoning (does the diagnostic chain hold up?), customer experience (legible to a non-mechanic?), output discipline (does the prose stay in voice?), and credibility of claims (counter-arguments + diagnostic tells present?). Findings drive a revisor pass, and if anything material remains it’s flagged on the report.

The audit_log records every lens’ findings and the cycles run before delivery.

Audit-trailed

Every Verdict carries an immutable audit_trail_id and inputs_hash. We can reproduce the exact same Verdict from the exact same inputs on any day, forever. You can hand the link to your mechanic, your spouse, your insurance adjuster — they can read the same report you did.

The audit_trail_id is visible in the footer of every Verdict and survives every reissue.

Independent of shops

One transaction with you — $79, $129, or $199. No subscription, no commission, no affiliate fees on parts, no referral payments from shops, no data sales to insurers. If anyone other than you is paying us, the product is compromised. So nobody else does.

Read the refund policy. If a Verdict doesn't earn what you paid, Devon refunds it — he reads every refund request himself.

The 13-stage pipeline

Every stage runs every time. No skips. No vibes.

The deterministic pipeline does the math. The language model writes the prose from the audit trail — it doesn’t negotiate the score.

1
IntakeVIN + mileage + symptoms + (Full only) your shop quote / inspection report. PDFs get OCR’d via Claude.
2
VIN decode (vPIC)40+ structured vehicle fields hydrated from NHTSA — make, model, year, trim, engine code, transmission, drive type, plant, body class.
3
Recall + TSB + ODI pullActive recalls, manufacturer technical-service bulletins, ODI investigations — all keyed to your VIN.
4
Failure-mode joinCross-reference your platform + mileage against the 730+ pattern failure-mode database.
5
Mileage-band clusterWhere your vehicle sits within its peer cohort — early life, mid-life, late life, beyond-design-life.
6
Weibull hazard mathPer-system probability that a documented failure mode is the cause of your symptom at your mileage.
7
Documented mitigation creditRecent service records, warranty extensions, recall remedies completed — each one reduces the relevant system’s hazard.
8
Coverage triageOpen recalls, customer-satisfaction programs, emissions warranty (8yr/80k), powertrain extensions, manufacturer goodwill — anything that could cover the work at no cost.
9
Regional cost adjustmentYour ZIP modifies labor-rate assumptions and parts pricing — Bay Area labor is not Birmingham labor.
10
Repair-to-value mathMost-likely repair cost vs. your vehicle’s trade-in and private-party values. The number where the math flips.
11
Score + band0–100 score with explicit confidence floor per rollup. Aligned with the 7-band customer-facing vocabulary (Strong → Replacement Window).
12
Prose renderClaude (Sonnet 4.6) writes the report from the deterministic audit trail. No score-fudging — the LLM renders prose only.
13
Independent audit pass6 lenses re-read the draft. Findings drive revisor passes. Up to 3 cycles. Failures are surfaced, not hidden.

Audit trail for every VerdictEach Verdict carries a permanent audit_trail_id. We can reproduce any score we shipped, byte-for-byte, on demand — and prove which inputs and which engine version produced it.

13-stage deterministic engineSame VIN + mileage + symptom in, same score out. The pipeline is mechanical, not vibes — reproducible on a different day, on a different server, by anyone with the inputs.

NHTSA-grounded by VINRecalls, complaints, and technical service bulletins pulled live from NHTSA against your exact VIN — not generic year/make/model lookups.

Powered by Anthropic ClaudeClaude Sonnet 4.6 renders the prose; Claude Opus 4.7 powers the deeper research passes. The deterministic pipeline produces the score — the model writes the explanation.

Built by a former restoration shop ownerDevon Smith ran a large vehicle restoration shop until a motorcycle accident closed it. He built RepairVerdict to give car owners the veteran-mechanic analysis he watched them go without.

One-time price · no upsellsNo subscription. No commission from shops. No affiliate referrals. We sell Verdicts; that is the entire business model.

Devon reviews every concern personallyDevon reviews every refund request personally against the Verdict's audit trail. No blanket calendar window — outcomes depend on the specifics, and Devon explains the decision either way.

Reproducibility, in plain language

Same inputs, same Verdict. Every time.

A Verdict isn’t a vibe-check. It’s a deterministic function of your VIN, your mileage, your symptoms, any documents you submit, and the snapshot of NHTSA data the engine pulled at run time. We store all of it under your audit_trail_id. If you run the same inputs tomorrow, you get the same Verdict — same score, same band, same probability ranking, same cost spread, same recommendation.

That sounds boring, and it’s the entire point. You’re not getting a roll of the dice. You’re getting a position you can defend — to your mechanic, to your spouse, to the dealer, to yourself.

Want to see a real Verdict before you spend a dollar?The sample gallery has complete Verdicts on a 2014 Land Rover LR4 and a 2020 RAV4 Hybrid — same format, same data sources, same audit trail.

Read them

Want to inspect the database the engine draws from? Browse the failure-mode encyclopedia — 730+ documented patterns by make, engine code, and mileage band.