i.
Spectral feature extraction.
Your recording is decomposed into frequency-domain features — harmonic content, transient signatures, periodic patterns. The acoustic fingerprint of your engine, captured in numbers.
Revly · The Manual
Edition No. 04 / 2026
Section 02
No. 02
Procedure
Procedure
From Hood Open to Diagnosis
The Manual.
Five steps. Roughly five minutes. Performed in your driveway, in a parking lot, or in your seller's driveway before you sign anything.
Step
i.
Find quiet ground.
A residential street, a half-empty parking lot, or your driveway. Revly tolerates moderate ambient noise but rewards a clean recording with sharper confidence scores. The app rates your environment before it begins.
Step
ii.
Open the hood. Hold the phone close.
About twelve inches above the engine bay. The app walks you through three captures: a cold idle, a warm idle, and a controlled rev to the cue. About ninety seconds end to end.
Step
iii.
Submit for analysis.
Recordings are processed on-device for noise isolation, then uploaded for the heavier analysis. Spectral features are compared against the reference set for your engine family — not a generic average, your family.
Step
iv.
Receive the report.
Under three minutes, in most cases. The report names detected conditions, their severity, the specific acoustic features that triggered each finding, and a typical cost range for repair. You may also receive a verdict of "all clear," which is its own kind of useful.
Step
v.
Repeat at intervals.
Monthly is reasonable. Before a long drive is wise. Before buying a used vehicle is, frankly, transformative. Each scan refines your engine's baseline and improves the precision of future findings.
No. 03
Apparatus
Apparatus
Three layers, working in concert.
A brief account of what the model is actually doing while you wait for your coffee to cool.
ii.
Reference comparison.
Your fingerprint is matched against the curated reference set for your engine configuration — same displacement, same arrangement, similar mileage band. Apples to apples, not apples to engine averages.
iii.
Failure mode classification.
A trained classifier maps any detected anomalies to specific named failure modes, each with a calibrated confidence score and a recommended urgency. The report is built from these classifications.
No. 04
A Word on Honesty
A Word on Honesty
A diagnosis you don't trust is worse than no diagnosis at all.
So we have built the system to admit uncertainty out loud. Every finding ships with a confidence score on a calibrated scale, every finding describes which sound features triggered it, and every report is willing to deliver the most useful possible result: "we don't hear anything obviously wrong."
A confident diagnosis the driver can verify is more valuable than a hidden one the model is sure of. — Revly engineering principle, No. 1
We do not gamify the report. We do not invent findings to justify the scan. We do not claim certainty we don't have.
No. 05
Next
Next
Ready for your first scan?
Enrollment in the beta is free, by application, and currently has space. We onboard in small cohorts and reply to every application personally within a week.