Skip to content
VINly
Download on the App Store

METHODOLOGY

How VINly reads a listing.

Six stages. Six source categories. Every model-level claim is grounded against a public source and tagged with a confidence level.

THE PIPELINE

OCR + EXTRACT

01

Read the listing

Pull year, make, model, trim, price, mileage, location, title, VIN, and seller text from screenshots.

IDENTITY

02

Decode the VIN

When a VIN is present, query NHTSA vPIC for canonical year/make/model and confirm against the listing.

KNOWLEDGE

03

Research the model

Anthropic Sonnet 4.6 surveys known failures, service intervals, regional comps — grounded against public sources.

SYNTHESIS

04

Compute the verdict

Anchor math (mileage + title status) produces an opening offer and a walk-floor. Confidence + risk are scored.

CRITIC

05

Self-check

A second pass critiques the report for unsupported claims, missing evidence, and unsafe certainty.

COMPOSE

06

Build the report

Bucket the findings into the six sections. Tag each claim verified / likely / possible. Cite sources.

SOURCES WE GROUND AGAINST

Every claim has a source.

Each known-issue, service-interval, and price-comp claim is tied to a class of public sources. Each is tagged in the Evidence Log as verified, likely, or possible.

NHTSA vPIC

VIN decode against the federal vehicle database.

NHTSA Recalls & Complaints

Open recalls and complaint counts per make/model/year.

Owner forums

Generation-specific known failure patterns (ih8mud, clublexus, tacomaworld, civicx, etc).

RepairPal-grade ranges

Typical repair cost ranges by job, anchored to regional labor rates.

Manufacturer service intervals

OEM service manuals + maintenance schedules where public.

Consumer Reports / Edmunds

Reliability deltas and long-term ownership cost patterns.

VERDICT MATH

How we get to an offer.

1 — Anchor against listing class.

We start with the regional median for the model + trim + mileage band. Mileage past 150k discounts by a heuristic (typically −15% from clean comps); past 200k by more.

2 — Title penalty.

Salvage stacks an additional 30–50% off clean-title comps. Rebuilt stacks 15–30%. Branded titles raise risk regardless of asking.

3 — Walk-floor.

Anchor minus a deal-friction discount produces the walk-floor — the price below which you simply leave. Both numbers are model-typical; not live market data.

WHAT WE DON'T DO

  • ×

    OBD-II / Bluetooth scanning / live diagnostics. None of that — VINly works from listing text and screenshots only.

  • ×

    Pull live KBB or Marketcheck data. The offer is a heuristic anchored to mileage + title status, not real-time comps.

  • ×

    Pull Carfax accident history. That's their job, not ours. We work from what a buyer can see in the listing.

  • ×

    Scrape listings or log into marketplaces. We see only what you upload.

  • ×

    Promise accuracy. AI can be wrong. Every claim is tagged with a confidence level so you can decide what to trust.

See the methodology applied to a real listing.

Open the full sample report → Download on the App Store