01
Data collection
Every report starts with independent sources — VIN decode, live listings, provider valuations, recall and ownership-cost data. No single provider is trusted alone.
Methodology
How Carlarity grades confidence
Carlarity Pre-Signature Intelligence™ — the trust layer behind every report.
Most car-pricing tools give you a single confident answer regardless of how much data they actually have. Carlarity calibrates every recommendation to the evidence behind it. When data is strong, we tell you decisively. When it's thin, we tell you to verify. We chose this trade-off deliberately.
6 min readRevised August 2026
The promise
One verdict. Backed by evidence.
Every report reduces a high-pressure decision to four things you can hold us to.
Verdict
A plain-English recommendation you can act on — not a wall of numbers.
Confidence
Every verdict carries a grade that tells you how strongly to trust it.
Evidence
The sources behind each number — pricing, identity, and ownership — shown inside the report itself.
Action
What to verify, negotiate, or walk away from — before you sign.
The pipeline
The five pillars behind every Carlarity report
Raw provider data becomes a decision in five deliberate steps — and the evidence grade constrains the verdict.
02
Normalization
Raw provider payloads disagree on format and meaning. We reconcile trim, mileage, condition, and locality so every source describes the same vehicle before any comparison happens.
03
Analysis & scoring
Pricing, financing structure, equity, depreciation exposure, and ownership cost are scored on a common 0–100 scale — higher is always better — so unrelated signals can be weighed together. Fees are itemized and classified rather than scored.
04
Evidence rating
Before a verdict is allowed to sound confident, the evidence behind it is graded. Thin, stale, or conflicting data caps how strongly the report may speak.
05
Verdict generation
Only then does the recommendation form — calibrated to the evidence grade, with softer language whenever confidence is limited.
What we analyze
Five dimensions of the same deal
A fair price can still be a bad deal. Every report grades the whole transaction, not just the sticker.
Price fairness
The asking price against market evidence, with labeled anchors and a confidence band — an estimate, and labelled as one. The out-the-door total it is measured against is computed from codified state tax and registration rules.
Fees & add-ons
Doc fees, add-ons, and dealer extras, read from the deal you entered — flagged against an out-the-door total computed from codified state tax and registration rules.
Financing terms
APR, term, and payment structure. Your monthly payment and total interest are closed-form arithmetic — exact for the terms you enter. Never forced onto cash purchases.
Equity outlook
What you'll owe at any month is exact from your loan terms. When that exceeds what the car is worth — and when you break even — depends on the value estimate.
Ownership cost
A five-year cost-of-ownership forecast — the cost of keeping the car, not just buying it. Fuel figures come from FuelEconomy.gov; maintenance and repairs are projected.
Trust Level
How we grade confidence
Every Carlarity report carries a Trust Level™ label. The label tells you how strongly to trust the recommendation.
HIGH
Decisive
Multiple sources, strong agreement. Our recommendation is well-supported. Verdicts at this level use decisive language.
MEDIUM
Cautious
Solid pricing data with some variance. Verify trim and options before relying on the exact number for major decisions.
LOW
Limited
Limited pricing data. The estimate is indicative, not authoritative. Verdicts use softer language and recommend verification.
INSUFFICIENT
Verify first
Not enough data to make a confident recommendation. Verdicts point to verification rather than action.
Principles
Built to sit on your side of the table
Four commitments the methodology is designed around — the reasons a Carlarity verdict can afford to be honest.
No sales incentives
We don't take dealer commissions, lender referral fees, or lead-resale revenue. Reports are paid for by buyers — that's the whole business model.
Source transparency
Every report names the providers that contributed in its Data Provenance section. We don't hide our sources.
Computed, not guessed
The cost side of your report is arithmetic. Your monthly payment and total interest are closed-form; your out-the-door total is computed from codified state tax and registration rules. The market valuation is an estimate, and it is labelled as one.
Conservative bias
When evidence is thin, we understate rather than overclaim. A cautious range you can verify beats a confident guess.
Buyer-first design
Every surface is written for the person signing the paperwork — plain language, visible caveats, and next steps you can act on.
Data sources
Where our data comes from
Carlarity uses multiple independent sources for every report and tells you which ones contributed.
NHTSA vPIC
Vehicle identity decode (free government API). Confirms make, model, year, body style from your VIN. Open-recall checks come from NHTSA's recall API.
VehicleDatabases
Multi-endpoint vehicle data (Market Value, Sales History, Title Check). An input to our Fair Price algorithm, weighted as an external benchmark; provides history corroboration.
MarketCheck
Live dealer listing data + market-value estimates. Cross-checks pricing against active inventory.
FuelEconomy.gov
Government-curated ownership cost data. Underpins our 5-year cost-of-ownership forecast.
We attribute which providers contributed to each report in the Data Provenance section. We don't hide our sources.
Carlarity Fair Price
Carlarity computes its own Fair Price signal: aggregated MarketCheck listings with asking prices corrected to estimated transaction prices, plus statistical outlier removal. Where it reaches MEDIUM confidence it is the primary pricing signal, shown as “Carlarity Fair Price” in the Data Provenance section; otherwise the multi-provider consensus is used. Confidence tier capped at MEDIUM until contribution data flows. Buyers post-report can help make future reports more accurate by sharing their outcomes anonymously.
Tier 2 evidence — Open-Web Corroboration
How we use the open web
When the primary pricing providers don't have enough data to support a confident answer, our second-tier engine widens the evidence surface — but only up to a capped ceiling, so an open-web supplement can never substitute for a confirmed market signal.
When it fires
Only when the API-tier confidence resolves to LOW or INSUFFICIENT — that is, when the primary providers couldn't produce a strong enough signal on their own. On reports where the API tier is already MEDIUM or HIGH, the engine never runs and never appears in the report.
What we cap
Open-web evidence can elevate confidence from INSUFFICIENT to LOW, or from LOW to MEDIUM — but never to HIGH. A HIGH-confidence verdict always requires primary provider agreement; the open web alone cannot earn it.
Sources we use
Ten whitelisted aggregators handle deep listing extraction: Autotrader, Carfax, CarMax, Cars.com, Carvana, CarGurus, Edmunds, Kelley Blue Book, TrueCar, and Vroom. Lighter-weight comps read from search snippets on other sites can also contribute, at reduced weight. Adding a source to the whitelist requires a code change and human review — the list is pinned by automated tests.
How extraction works
When a value is extracted from a listing page, price and mileage must be grounded in a verbatim quote — both are substring-checked against the page text, and anything the language model can't ground in an exact quote is rejected, not surfaced. Comps read from structured listing markup or search snippets carry those source values directly instead.
How we label it
When open-web evidence contributed, your report shows a Search-Augmented chip — in the pricing evidence section on the web report, and on the cover of the PDF — and lists the comps used (price, mileage, trim, source, and date where available; up to ten shown) so you can verify the underlying evidence yourself.
How accuracy is tracked
We measure the Carlarity Fair Price two ways: against publicly listed dealer asking prices, and against internal cross-provider convergence — how closely MarketCheck and VehicleDatabases agree on the same vehicle. Tier-elevation rules are tuned on that observed convergence across vehicle classes before any further loosening. We're building that into a published monthly figure.
We don't publish a headline accuracy percentage until enough real outcomes have flowed to stand behind one. We validate against verifiable market evidence, not against another vendor's estimate.
Three outcomes — what we tell you when
Open-web supply varies by vehicle. Some vehicles have abundant structured comps; others have almost none. Our engine handles all three cases explicitly — corroboration is visible on every eligible report, not just the ones where it elevates confidence.
Search-Augmented
Fired (3+ open comps)
Corroboration found at least three independent open-web comps for your vehicle. When the comp cluster is tight enough, confidence elevates per our methodology table (capped at MEDIUM, never HIGH). The comps used are shown — price, mileage, trim, source, and date where available.
Partial Corroboration
Supplemental (1–2 open comps)
We found one or two open-web data points, shown transparently as additional context. A single data point never moves confidence. Two tightly-clustered points can support an indicative, search-augmented price range and lift confidence at most from INSUFFICIENT to LOW — always labeled as an indicative estimate, never presented as confirmed market value. The amber chip is intentional: this is a partial signal, not a confirmed one.
Open-Web Supply Limited
Dark (0 open comps)
For some vehicle classes — luxury, rare, very-low-volume — structured listing data isn't reliably available on the open web. We say so rather than fabricate signal. Confidence stays at whatever the primary providers reported.
Why this matters. The cleanest way to be honest about thin pricing data is to widen the evidence surface without overclaiming. Open-Web Corroboration adds independently-sourced comps to your report when the primary providers can't carry it alone — and stops short of letting those comps masquerade as confirmed market value.
The systems
How Carlarity Pre-Signature Intelligence works
Twelve proprietary systems power every report. Each one has a specific job in turning raw provider data into a trust-honest decision.
Deal Intelligence Engine™
What it doesThe Deal Intelligence Engine converts pricing, financing, and ownership-cost signals into a single trust-honest decision.
Why it mattersIt's the system that turns raw provider data into a recommendation calibrated to the evidence behind it.
Fair Market Signal Engine™
What it doesCarlarity's Fair Market Signal Engine compares the asking price against live market evidence, provider-backed ranges, and confidence-adjusted fallback logic.
Why it mattersIt tells you how the price you're being shown stacks up against what comparable listings are actually asking.
Provider Consensus Engine™
What it doesWhen two pricing providers are available, Carlarity compares them and reports how closely they agree. When only one is available, we cap confidence accordingly.
Why it mattersAgreement across independent sources is what earns a confident verdict; disagreement is what triggers caution.
Signal Integrity Layer™
What it doesThe Signal Integrity Layer prevents Carlarity from overstating certainty when provider evidence is thin, conflicting, or incomplete.
Why it mattersIt's the guard that keeps the report honest when the data behind it isn't strong enough to support a confident answer.
Deal Verdict Engine™
What it doesThe Deal Verdict Engine converts pricing, financing, fee, equity, and depreciation signals into a plain-English recommendation.
Why it mattersIt's the system that produces the one-line verdict you act on — and it softens that language when confidence is low.
Report Guard™
What it doesReport Guard prevents low-confidence reports from using overconfident language like 'proceed with confidence' or 'confirmed market data.'
Why it mattersIt enforces the trust-honest rule at the surface level: the words you read always match the evidence behind them.
Transaction Context Engine™
What it doesThe Transaction Context Engine makes sure cash buyers don't receive loan advice and finance buyers receive accurate APR, payment, and term analysis.
Why it mattersYour report's recommendations are tailored to how you're actually paying, not a generic one-size-fits-all template.
Vehicle Identity Resolver™
What it doesThe Vehicle Identity Resolver confirms when a VIN has enough data, when manual entry is needed, and how much confidence the report can safely claim.
Why it mattersIf we don't know exactly which vehicle you're looking at, we can't match it to the right comparable listings — this is the system that catches that.
Indicative Estimate Engine™
What it doesWhen live provider evidence is incomplete, the Indicative Estimate Engine provides a cautious range without pretending it is confirmed market value.
Why it mattersIt's how Carlarity stays useful when data is thin — a range you can verify against, not a single number presented as truth.
Decision Report Composer™
What it doesThe Decision Report Composer turns all backend intelligence into a buyer-safe overview, full report, and PDF.
Why it mattersEverything you see in a Carlarity report — the overview, the full breakdown, the PDF — flows through this composer with consistent confidence labels and provenance.
Negotiation Intelligence Engine™
What it doesThe Negotiation Intelligence Engine turns pricing and financing findings into specific scripts a buyer can use before signing.
Why it mattersKnowing the numbers is one half of the negotiation; having the exact words to use at the dealership is the other half.
Open-Web Corroboration Engine™
What it doesThe Open-Web Corroboration Engine supplements the API valuation with structured open-web listings — full listings from a whitelisted set of aggregators, and price snippets from other sources at reduced weight — when the API-tier evidence alone is too thin to support a confident answer.
Why it mattersWhen the primary pricing providers don't have enough data, the engine widens the evidence surface — but only up to a capped confidence ceiling, so an open-web supplement can never substitute for a confirmed market signal.
Philosophy
Why we say what we don't know
Trust-honest by design
Most car-pricing tools give you a single decisive answer regardless of how much data they actually have. That's good for confidence — and bad for accuracy. We chose a different path: when we have strong data, we'll tell you decisively. When we don't, we'll tell you the data is limited and what to verify. We'd rather show limited confidence than pretend our data is stronger than it is.
This is what we mean by 'trust-honest.' Every verdict you see is calibrated to the evidence behind it. Read the confidence label first; read the verdict in that context.
And what we do know exactly
Your total interest, monthly payment and the loan side of the equity timeline are closed-form arithmetic. Your out-the-door cost is computed from codified state tax and registration rules, and detected fees, title status and open recalls come from the deal you entered and from public registries. The market valuation is an estimate, and it is labelled as one.
Verdicts
What changes verdicts
Five verdict labels, and only five. At LOW confidence a Buy or Negotiate reads VERIFY FIRST; at INSUFFICIENT confidence the report says INSUFFICIENT EVIDENCE. The buying mode changes the advice, never the word.
- BUY
- ADJUST
- VERIFY FIRST
- WALK
- INSUFFICIENT EVIDENCE
Handling disagreement
When sources disagree
Spread thresholds
When two pricing providers are live for a vehicle and return different price estimates, the Provider Consensus Engine measures the spread. Tight agreement (under 5% spread) supports HIGH confidence. Wide disagreement (over 15%) demotes the report to LOW confidence — we don't pick a winner and pretend it's the truth. We tell you the sources disagree. When a single pricing provider is active — the common case in the current configuration — consensus doesn't apply, confidence is capped per the single-source rules below, and the provenance line names the provider that backed the report.
When pricing data is too thin to compute a meaningful estimate, the Indicative Estimate Engine provides a cautious range without presenting it as confirmed market value.
Single-source confidence
Why a single source can support MEDIUM confidence
Class-aware tier caps
MEDIUM tier can be supported by either (a) multi-source consensus between two live pricing providers (MarketCheck and VehicleDatabases, when both are active), OR (b) a single high-quality VehicleDatabases response — when VDB is live — for the vehicle classes well-served by its pricing model: mainstream sedans, EVs, near-new vehicles, and trucks. Which providers are live varies by configuration; in every single-source case, the provenance line on the report shows the active provider so customers see exactly what backed the tier.
We accept single-source MEDIUM for these classes because VDB's spec-precision signal — exact trim, drivetrain, locality, and mileage band — carries enough discriminating power on its own. The customer-visible attribution is the trust contract: we never claim multi-source consensus we don't have.
For vehicles in the rare-low-liquidity class — older vehicles, salvage titles, very-low-volume models — confidence is floored at LOW regardless of source quality. Comp pools are inherently thin for these vehicles, so consensus on a thin pool isn't evidence of trustworthy data.
Open-web corroboration can ELEVATE a confidence tier (e.g. INSUFFICIENT to LOW) when it surfaces three or more independent comps — it never demotes. A “dark” corroboration outcome (no usable open-web evidence) does not downgrade a single-source MEDIUM.
Coverage
What vehicles we analyze
Scope
Our current coverage is the mainstream models we have tested end to end, model years 2015–2026 — not a make-and-year grid. Electric and plug-in hybrid vehicles, luxury and import brands, heavy-duty trucks, gray-market imports and model years before the tested floor fall outside that scope. Within the covered list, thin evidence lands a vehicle at LOW confidence and the verdict points to verification rather than a confident answer — and when the evidence can't support a confident price at all, the report is held before checkout and you aren't charged.
Put it to work
See the methodology at work
Run it on a real deal — or read a full sample report first and judge the evidence for yourself.