# Selva

> Selva is software for auto-parts e-commerce stores. It sells only what fits and finds the sales a store never saw: it reads what a shopper actually asked, says "fits" only when the maker's document does, asks the one question that settles the rest, gives AI shopping agents the same verdict as data, and reports what the store never knew to look for.

Contact: Ken Navarro, founder, ken@runselva.com. Wayne, Pennsylvania. Selva is a product of Selva Exploration LLC.

## What Selva does

- Fit answers: every verdict is checked against the maker's fitment document, a regulator's record or the store's own page. No guessed "fits".
- The deciding question: when one fact decides fit, Selva asks it once. Example: a 2019 F-150 can have an 8.8-inch or a 9.75-inch rear axle; a 12-bolt rear cover means the 9.75.
- Whole jobs: sets and kits are checked together, such as a lift kit with shocks sized for it, or a gear set with the install kit it needs.
- AI agents: an agent shopping the store gets the same verdict as typed data, with the deciding question and how to tell. Selva never states a price, stock level or delivery date the store does not state.
- Reports: counts losses store reports never see, such as hidden fits (the maker lists the part for the vehicle, the store says no, and the shopper leaves).

## Start

The free check measures a store's hidden-fit and false-fit rates from its public pages, with intervals, before the store shares any data. Write to ken@runselva.com.

## The 21 benchmarks

Selva is measured by 21 published benchmarks in five families. About half of each benchmark's cards, on average, sit in the two hardest of four difficulty bands. Cards are generated fresh at scoring time, shared anchors make stores comparable, verdicts are graded by code rather than a language model, and every run is recorded before it starts. Agent shoppability (B21) passes a question only when three separate runs are all right.

### Fit truth

- B1 False fits: How many of my 'fits' badges does the maker's own document contradict?
- B2 Hidden fits: Which parts does the maker list for a vehicle that my store hides?
- B3 Hedges and question debt: How many of my customers' fit questions could the maker's document already answer?
- B4 Selector depth gap: Which one selector field would clear the most wrong 'fits' badges?
- B5 Ignored warnings: Which customer misfit reviews were right, and how long has each page ignored them?
- B6 Recommendation leakage: Do my own cross-sells recommend parts my own fitment data rejects?

### Finding and promises

- B7 Search to part: When a shopper types a symptom instead of a part name, what do they find?
- B8 Self-contradictions: Where do my title, fitment table and cart disagree about the same part?
- B9 Price truth: Which charges does a shopper first see in my cart, not on the page?
- B10 Delivery promise: Does my 'ships today' agree with the date my cart actually shows?
- B11 Policy truth: Does my tuner page say 'no returns once programmed', or only my policy page?
- B12 Job completeness: Which of my kits leave a customer stuck in the driveway, missing a required part?

### Time and effort

- B15 Time to confident cart: How long does a shopper take to reach a correct cart, not just any cart?
- B16 Interaction cost and lostness: How many dead ends does a shopper hit before finding the right part?
- B17 Task-path speed: Is my search results page slow even though my homepage is fast?

### Hard shoppers and builds

- B18 Unprepared-shopper resolution: When a shopper is sure of the wrong lamp type, does my site correct them?
- B19 Build compatibility: Does my cart warn when the lift and the shocks don't match?

### Machines, law and access

- B13 Machine-readable truth: Does the price Google and AI assistants read match my page and cart?
- B14 Legal truth: Does each 'CARB legal' claim actually cover the engine and years on my page?
- B20 Accessible fit path: Can a blind shopper using a screen reader hear my fit verdict?
- B21 Agent shoppability: When an AI assistant shops my store, does it recommend the right part?

## Pages

- [Selva](https://runselva.com/): the site
- [The 21 questions](https://runselva.com/#deck-h): every benchmark, with an example and what it is checked against
- [Built for agents](https://runselva.com/#agents): agent shoppability, B21
- [The free check](https://runselva.com/#start)