Sunara Reference

The problem

A FLAC file proves nothing about what's inside it. Anyone can take a 128 kbps MP3, decode it, and re-encode it as FLAC. The result passes every integrity check, plays perfectly, and takes up lossless-sized space — but the quality lost in the MP3 stage is gone forever. These files are everywhere: in download stores, in shared libraries, in collections people have spent years building.

Why most checkers get this wrong

Nearly every existing tool checks one thing: whether the spectrum stops early. Lossy encoders cut high frequencies, so a cutoff around 16–20 kHz looks like a smoking gun.

The trouble is that plenty of genuine recordings also stop early. Old masters, analog tape sources, quiet acoustic material, deliberate mastering decisions — all of these produce natural high-frequency roll-off that a cutoff-only tool happily convicts as fake. If you've ever had a checker flag an album you ripped yourself, this is why.

What Sunara Reference does differently

Lossy codecs leave more than one fingerprint. Sunara Reference runs five detectors across three independent evidence families — where the spectrum ends, what the surviving high frequencies look like statistically, and how they behave over time — and fuses them with a simple rule: artifacts prove fake, but the absence of artifacts doesn't prove real. When the families disagree, the file is labeled Borderline instead of getting a confident guess.

Read the full design story in How it works or the technical article One cutoff is not enough.

Honest numbers

On a controlled test set — 101 genuine hi-res FLACs, plus the same 101 files transcoded through MP3 192, MP3 128, and AAC 160 and re-wrapped as FLAC — Sunara Reference scores 96.5% overall (390 of 404 files), with 93.1% on the genuine files it must not falsely accuse.

Those are controlled conditions, not a promise about every file in the wild. The score you see in the app is a confidence value, not a probability — and the app never pretends otherwise.