How can we help?

MelonSpank works by analyzing the acoustic response of the melon when you tap it. A ripe, sweet melon produces a different resonance profile than an underripe or overripe one.

The microphone captures the tap sound, and the on-device AI extracts spectral and temporal features from that signal to estimate the Brix score.

No audio is ever recorded to disk or transmitted. The raw audio is processed entirely on your device and discarded immediately after feature extraction.

Location is completely optional and not required to use the app. If you grant access, we use a coarsened version of your coordinates (approximately 100 m precision) for regional model calibration only. Precise GPS is never stored or transmitted.

You can revoke this permission any time in iOS Settings, Privacy & Security, Location Services, MelonSpank.

MelonSpank is in beta and the model is still being trained. A few things to keep in mind:

• The model was bootstrapped with synthetic and laboratory data. Real-world accuracy improves as more users submit tap-and-rate sessions.
• Tap technique matters significantly. See the "How do I get better results?" question below.
• Refrigerated melons behave differently acoustically than room-temperature ones.

Submitting your star rating after tasting is the single most useful thing you can do to improve future predictions.

No. Audio is processed entirely on-device. The raw audio buffer is discarded immediately after feature extraction. It is never written to disk and never uploaded.

What gets sent to our servers (if you submit a rating) is a numerical SHA-256 hash of aggregate acoustic features, not audio, not a recording, and not anything that can be reversed into sound.

• Tap the equator — the middle band around the widest part of the melon.
• Medium force — firm but not hard. Think "knocking on a door."
• 5 taps — the app collects 5 taps and averages the response.
• Quiet environment — background noise interferes with signal quality.
• Room temperature melon — refrigeration changes resonance characteristics.

MelonSpank is trained specifically for the six melon varieties listed in the app: watermelon, cantaloupe, honeydew, galia, charentais, and cavaillon. Other hollow fruits such as pumpkins or squash may produce outputs but the calibration will not be accurate. Use results on non-melon produce for curiosity only, not purchasing decisions.

File a support ticket using the form above and select "Privacy" as the category. Include your anonymous UUID (visible in the app under Settings). We will delete all associated records within 30 days.

Yes, completely free during beta. In exchange, we ask that you submit anonymous tap-and-rate sessions to help train the community model. There are no ads, no subscriptions, and no in-app purchases. The model improves as the community grows.