used-car-sentinel

Use when you are shopping for a used car and must decode a vehicle listing (VIN, mileage, price, age) into a real risk picture before you drive anywhere or pay a deposit. VIN validation + decoding (WMI country/manufacturer, model-year code, check-digit verification, sequential-number red flags), mileage-plausibility analysis against age and class (detects odometer rollbacks and clocked clocks), holistic price-vs-market fair-offer model, deal-breaker screening (salvage/rebuilt/flood/title brands, TSI taxi/fleet patterns, untold collision age gaps), and a structured 30-minute test-drive inspection sheet. All offline — works when you are standing in a parking lot with one bar of signal.

Used Car Sentinel — Listing Decoder & Deal Screener

Overview

Used-car listings are adversarial documents. The seller — professional or private — presents the car in the best light the truth allows and some light it doesn't. The buyer has one photo set, a price, an age, a mileage, and (if lucky) a VIN, and must decide: is this car worth a 90-minute round trip and a deposit?

Every year, buyers lose billions to:

• Rolled-back odometers — still common (interstate commerce makes prosecution rare); a 5-year-old car "with 68,000 km" that actually has 168,000.
• Salvage/rebuilt titles laundered state-to-state or simply not mentioned.
• Fair-price blindness — asking prices on private-marketplace listings routinely run 10–25% over real transaction values; buyers negotiate $500 off a $3,000 markup and feel victorious.
• Wasted weekends — the listing was junk; the photos were old; the "one owner" was a rental fleet.

This skill is a pre-contact screening tool. It takes what the listing gives you and returns a risk-ranked verdict before you ever send a message:

1. VIN Intelligence — structural validation (17 chars, no I/O/Q, correct check digit per ISO 3779), WMI decode (country + manufacturer), model-year code extraction with a 30-year cycle disambiguation, plant code. A VIN that fails the check digit is a forgery flag, full stop.
2. Mileage plausibility — km/year vs class norms (city runabout ~12k, average ~15–20k, highway cruiser ~25–35k, taxi/fleet 50k+); flags both suspiciously low (rollback or garage queen rot) and unusually high (fleet or engine near end-of-life). Age-band tables: what mileage is normal for a 3/5/8/12-year-old car.
3. Fair-offer model — asking price vs computed market band from age, mileage, and class depreciation curves; suggests an opening offer, a walk-away price, and the exact negotiation talking points derived from the car's own risk flags.
4. Deal-breaker screen — the patterns that end the analysis: title brands, missing VIN plate, mileage inconsistent with service history gaps, price too good to be true (>30% below modeled market = almost certainly a scam or salvage).
5. Pre-purchase protocol — what to demand before travel (VIN photos, service-book pages, title photo), the 30-minute inspection sheet, the test-drive route that exposes head gasket, transmission, and suspension faults, and when to pay for an independent inspection ($100–200 that saves $4,000).

All offline. No API keys. Runs in a parking lot on a phone.

When to Use

• Screening any private-party or small-dealer used-car listing before contact
• Validating a VIN the seller sent (or refusing to travel until they do)
• Sanity-checking mileage vs age before believing "little old lady, weekend use"
• Building a negotiation sheet from objective risk flags
• Running the 30-minute inspection at the viewing (printable checklist)
• Don't use for: new cars (different economics), collector/enthusiast vehicles (value is condition+provenance, not depreciation curves), or as a substitute for a professional pre-purchase inspection, a vehicle-history report (Carfax/Autocode/local equivalent), or legal advice. This narrows the odds; it doesn't replace diligence that costs money.

How It Works

1. vin decode <VIN> — validate structure + check digit, decode WMI (country/manufacturer), extract year/plant, flag mismatches (e.g. "2019 model year" VIN on a listing that says 2021).
2. screen --year 2018 --km 68000 --price 11500 --class compact — the full pre-contact workup: mileage plausibility, market band, offer strategy, risk list, verdict (GREEN / YELLOW / RED).
3. inspect — prints the 30-minute inspection sheet + test-drive protocol.
4. negotiate --year 2018 --km 68000 --price 11500 --class compact — talking points derived from the risk flags, with dollar values attached.

Commands

bash

# VIN first (get it from the seller before traveling)
python3 scripts/sentinel.py vin decode 1HGCM82633A004352
# → check digit OK, Honda JP/US plant, year 2003 — structure valid

python3 scripts/sentinel.py vin decode 5YJ3E1EA7KF317000
# → Tesla, 2019, valid structure

# Full screening of the listing
python3 scripts/sentinel.py screen --year 2018 --km 68000 --price 11500 --class compact
# → mileage PLAUSIBLE-LOW 11.3k km/yr ... verdict YELLOW: negotiate from 9800

# Inspection sheet for the viewing
python3 scripts/sentinel.py inspect

# Negotiation sheet
python3 scripts/sentinel.py negotiate --year 2018 --km 68000 --price 11500 --class compact

VIN Check-Digit Math (why it catches forgeries)

Position 9 of the VIN is a check digit computed from the other 16 via a published letter→number table and weight loop (ISO 3779). Legitimate manufacturers compute it; counterfeit plates, typo'd listings, and cloned identities frequently get it wrong. vin decode recomputes it and says VALID / MISMATCH / NOT-COMPUTABLE (EU VINs sometimes skip it — noted, not flagged red by itself).

Common Pitfalls

1. Trusting the listing year over the VIN year. The VIN's position-10 code is the model year; a mismatch means the listing (or the seller) is wrong. Always reconcile.
2. Believing round mileage. 68,000 exactly on an 8-year-old car is either true or rolled back to a number humans like. Ask for the service book and cross-check every stamp.
3. Negotiating from the asking price. Negotiate from your computed band. Anchoring on their number is how a $3,000 markup becomes a $2,500 "win".
4. Skipping the check digit. A VIN with a bad check digit means the plate is counterfeit or mistyped — either way, walk.
5. Confusing "no accidents reported" with "no accidents". History reports only show what was reported to insurance/police. The inspection sheet covers panel gaps, paint texture, and weld beads for a reason.
6. Testing the stereo before the cold start. The inspection sequence is ordered so the expensive faults (cold-start knocks, head gasket, sludging) surface first. Follow the order.
7. Paying a deposit to "hold" a car you haven't inspected. Deposits on GREEN/YELLOW cars are refundable only in your imagination.
8. Buying flood cars by smell alone. Modern flood cars are detailed professionally. The sheet checks under-seat bolts, harness connectors, and spare-wheel wells for silt lines and corrosion.

Verification Checklist

• vin decode validates 17-char structure, rejects I/O/Q, computes the ISO 3779 check digit
• WMI table covers major markets (US, JP, DE, KR, CN, FR, IT, GB, SE, ES, CZ, TR, MX, BR, IN, RU)
• screen outputs: mileage verdict, market band, opening/walk-away offer, risk flags, GREEN/YELLOW/RED
• Price >30% under modeled market triggers a SCAM/SALVAGE warning, not excitement
• inspect sheet covers cold start, fluids, panel gaps, OBD port, test-drive route, paperwork
• Everything runs offline with stdlib only

References

• references/vin-math.md — check-digit computation worked through, WMI table excerpt, year-code cycle, market-band model assumptions, inspection rationale