# Dealer Inventory Photo Evidence Scorecard

Version: 1.0
Based on: CarPixAI 2026 VDP Photo Benchmark frozen rubric
Use: internal dealership inventory QA; not a sales forecast or industry ranking

## What this scorecard does

Score one live vehicle listing from 0 to 12 across six observable dimensions. Use the same selection rule and reviewer process each time so a dealership can compare its own workflow over time.

This scorecard does **not** estimate leads, sales, days-to-sale, dealer quality, or vehicle quality. A high score does not prove that a listing will sell faster.

## Recommended audit design

1. Choose a stable inventory rule before looking at scores, such as the first 20 active used units by inventory ID.
2. Record the audit date and the exact listing URL internally.
3. Review the published hero plus enough gallery frames to judge coverage.
4. Score only observable evidence; do not infer an unshown photo exists.
5. Compare every AI-edited image with its source before publishing.
6. Keep dealer/customer identities, VINs, stock numbers, plates, and URLs out of public exports.
7. Repeat the same selection rule after a workflow change.

## The 12-point rubric

| Dimension | 0 points | 1 point | 2 points |
|---|---|---|---|
| First hero image | Placeholder, heavily obstructed, severely cropped, blurry, or not useful as a vehicle hero | Usable but affected by clutter, glare, darkness, obstruction, weak angle, or tight crop | Sharp, readable whole-vehicle presentation with a useful angle, manageable background, and crop room |
| Gallery completeness | Fewer than 12 unique vehicle images | 12–19 unique vehicle images | At least 20 unique vehicle images |
| Condition proof | Almost no observable condition evidence | Some interior, feature, odometer, wheel, cargo, or condition evidence | Interior plus at least two additional buyer-proof categories |
| Background consistency | Severe clutter, inconsistent settings, or repeated obstruction | Usable presentation with distractions or setting/framing variation | Repeatable, non-distracting setting with coherent framing |
| Mobile crop safety | Important vehicle edges are clipped or likely to disappear | Readable hero, but at least one edge is tight | Safe space remains around roof, bumpers, and tires for common card crops |
| Visible edit safety | Obvious distortion, concealment, implausible compositing, or misleading treatment | Uncertainty, aggressive overlays, or inconsistent visual cues requiring source review | No observable manipulation warning signs in reviewed images |

## Score interpretation

Use the total as a compact internal QA signal, not a grade of the dealership.

- **0–3:** evidence is missing or the listing presentation is materially unsafe.
- **4–6:** usable but several buyer-review gaps remain.
- **7–9:** solid listing with a smaller number of priority improvements.
- **10–12:** strong under this rubric; still compare edited images with their sources and inspect the full gallery.

## Action routing

| Finding | Correct next action |
|---|---|
| Blur, stale vehicle, wrong crop, or missing proof angle | Retake the source photo |
| Sharp source with a distracting environment | Test eligible background cleanup while preserving the exact vehicle and condition |
| Tight hero crop | Reframe or select a safer hero; do not invent missing vehicle pixels |
| Missing interior, odometer, wheel, cargo, or defect proof | Add truthful source photography |
| Visible AI artifact or changed vehicle detail | Reject the edit and use the original or an approved redo |
| Inconsistent order across listings | Standardize the capture and approval checklist |

## Downloads and tools

- Blank spreadsheet: `/dealer-photo-benchmark-audit-template.csv`
- Public benchmark: `/blog/used-car-photo-benchmark-report-2026`
- Photo Grader: `/tools/car-listing-photo-grader`
- Hero-image previewer: `/tools/vdp-hero-image-previewer`
- Shot-list generator: `/tools/car-photo-shot-list-generator`

## Evidence boundary

The frozen v1 benchmark reviewed 108 purposively sampled public VDPs. A separate v1.1 replication included 111 of 122 fixed candidates across seven website-platform families. Neither sample is probability-weighted or nationally representative, and neither measured commercial outcomes. Visual classifications were model-assisted and reviewer-sensitive; the published methodology and QA results disclose that uncertainty.