<divclass="metric-card"title="Peak Magnitude - highest spectral amplitude in dBFS (0 dBFS = full scale). How to read: • > -1 dBFS: Heavily limited/mastered, extremely loud • -6 to -1 dBFS: Typical modern mastering • -12 to -6 dBFS: Conservative, good headroom • < -12 dBFS: Quiet or attenuated Significance: Combined with noise floor, indicates the effective dynamic range of the audio.">
<divclass="metric-card"title="Transition Bandwidth (Steepness) - frequency gap between the -20 dBFS and -60 dBFS points. Measures how abruptly high-frequencies drop off. How to read: • < 500 Hz: Brick-wall cutoff - transcode or low-bitrate lossy • 500-1500 Hz: Moderately sharp - suspicious, may be higher-bitrate lossy • 1500-3000 Hz: Gradual rolloff - typical of genuine lossless • > 3000 Hz: Very gradual - natural rolloff from analog sources Significance: The strongest differentiator between genuine lossless and transcodes. Lossy encoders produce artificially steep lowpass filters.">
<divclass="metric-card"title="Noise Floor - average magnitude in the top quarter of the spectrum (> 3/4 Nyquist), in dB relative to peak. How to read: • < -90 dB: Excellent - extremely clean, high dynamic range • -80 to -90 dB: Very good - typical of quality lossless • -70 to -80 dB: Moderate - some quantization noise present • -60 to -70 dB: Elevated - suggests lossy compression or analog hiss • > -60 dB: High - almost certainly lossy compressed Significance: Lossy codecs raise the noise floor through quantization. High noise floor + sharp cutoff = strong transcode indicator.">
<divclass="metric-card"title="Spectral Roughness (CoV) - coefficient of variation (σ/μ) of spectral magnitudes in the 60-95% region below cutoff. Measures how uneven/scalloped the spectrum is. How to read: • > 0.40: Strong spectral scalloping - clear lossy encoding signature • 0.30-0.40: Moderate scalloping - suspicious, may be high-bitrate lossy • 0.15-0.30: Mild unevenness - needs band-ratio correlation • < 0.15: Smooth spectrum - typical of genuine lossless Significance: Lossy codecs create a comb-filter pattern via frequency-bin quantization. High roughness = high probability of lossy encoding.">
<divclass="metric-card"title="Band Ratio - ratio of average spectral energy in 16-20 kHz band to 12-16 kHz band. Detects unnatural high-frequency tilt. How to read: • < 0.5: Normal natural rolloff in top octave • 0.5-0.9: Typical for most recordings • 0.9-1.0: Flat HF - possible spectral reshaping from lossy encoding • > 1.0: Inverted tilt (more energy at 16-20 kHz than below) - strong transcode signature Significance: Used as secondary criterion when cutoff is high but roughness is ambiguous. Transcodes often show unnatural energy distribution in the top octaves.">
<sectionid="decisionSection"title="Verdict - the overall classification of the audio file based on spectral analysis. NATIVE: File matches its declared lossy format at a consistent bitrate (single encode). UPSCALED: Lossy file re-encoded to a higher bitrate than the original source. GENUINE: File appears to be native lossless (no transcode artifacts detected). TRANSCODE: Lossy source re-encoded to a lossless format (e.g. 128k MP3 → FLAC). The verdict is decided by: cutoff ratio, transition bandwidth (steepness), roughness (CoV), and band-ratio analysis against adaptive thresholds.">
<divclass="confidence-container"title="Validity Confidence - how strongly the evidence supports the verdict, from 0% (random chance) to 100% (certain). 85-100%: Very strong evidence - multiple criteria agree decisively. 70-84%: Strong evidence - most criteria support the verdict. 50-69%: Moderate evidence - some criteria triggered but not all. 30-49%: Weak evidence - marginal, borderline detection. 0-29%: Inconclusive - possible false positive, or audio is ambiguous. For lossy files: confidence is based on how far below the expected cutoff ratio the file sits. For lossless candidates: confidence blends cutoff ratio margin, steepness margin, and (above 80% Nyquist) secondary roughness + band-ratio criteria.">
<sectionid="factors"title="Factor Analysis - step-by-step breakdown of the detection criteria used to reach the verdict. ① PRIMARY criteria (applied to ALL files): Compares transition bandwidth (steepness) to an adaptive threshold based on cutoff ratio. If steepness is BELOW the threshold, the spectrum has an artificially sharp cutoff → transcode flagged. Files with cutoff_ratio ≥ 0.99 (full spectrum) bypass this check. ② SECONDARY criteria (applied when cutoff_ratio > 0.80): Checks spectral roughness (CoV) and band ratio for additional transcode signatures. High roughness (> 0.40) alone is sufficient to confirm. Moderate roughness needs band-ratio corroboration. Lossy files: checked only for cutoff ratio vs expected range for the bitrate. ✓ = criterion met (supports transcode/upscaled), ✗ = criterion not met (supports genuine/native).">
