Preserve a claim that can actually be reviewed
This workflow is for collectors, editors and library maintainers reviewing a received FLAC/WAV. Record the supplier's claim separately from facts measured in the file. Keep the original, its SHA-256, acquisition notes and intended use. A matching hash identifies bytes, not authorship, recording history or a trustworthy master. Probe metadata only if uploading is permitted; declared codec and depth describe the current file, not its ancestors.
Screen the original locally
Open the supported original in the local checker without conversion. Limits are 100 MiB, 60 minutes, mono/stereo and 8–192 kHz. Native FLAC needs known length, metadata at most 2 MiB and blocks at most 16,384 samples; RF64 is unsupported. A rejected format is not evidence of lossy origin. The server endpoint runs synthetic demos only and must not receive your recording.
Save the PNG and JSON with settings and revision. Up to 384 evenly spaced 8192-sample Hann windows are inspected; reported coverage is sampled duration, not confidence. A persistent sharp cutoff can follow legitimate filtering, an older master or sample-rate conversion. Silence, tones, short material and low rates may provide insufficient evidence. A clean label cannot exclude gradual rolloffs, high-bitrate encodes or edits between sampled windows.
Follow clues without modifying evidence
If authorized, inspect a time-frequency spectrogram and listen around suspected transitions in a separate player at safe volume. Record the timestamp and whether the clue persists or is local to an instrument or section. Different scales and windows can change the picture; do not force a universal cutoff threshold or equate darkness with compression.
The integer bit analyzer accepts PCM/FLAC up to 100 MiB and 180 seconds, mono/stereo at 8–96 kHz. Float and lossy inputs are not applicable, and silent channels are indeterminate. Low zero bits describe occupancy, not historical recording depth or ENOB; processing and dither can occupy low bits. For longer recordings prepare separately identified excerpts externally, retaining offsets and untested ranges. Do not silently claim a short excerpt covers the whole file.
Use a reference only when its role is known
No trusted reference means no reference-based verdict. The codec artifact tool compares a known original master with its lossy encode; it does not authenticate an arbitrary FLAC by being given an unrelated release. Inspect alignment, event timestamps and listening evidence. Its energy indicators are not a perceptual certificate.
Null testing is a separate sample-equality question. Inputs must be at most 30 seconds with identical rates and channel counts; no automatic alignment, gain matching or resampling occurs. Document the integer offset and overlap, compared and excluded frames. Zero residual applies only to compared decoded samples, not all file bytes or recording history. Nonzero residual may reflect gain, timing or a different master, not necessarily lossy coding. Use hashes for byte identity and leave unsuitable comparisons unassessed.
Deliver evidence and an honest decision
Compare prior and new JSON only with matched file identity, settings and report structure. Explain changed parameters instead of ranking verdict labels. Hand off original identities, PNG/JSON, authorized helper reports, listening notes, reference provenance, excerpt mappings and final hashes. Keep private source information in an access-controlled local ledger, not public tool inputs.
Write one bounded conclusion: no obvious traces under this method, investigate specified evidence, or insufficient evidence. List what would resolve uncertainty, such as the supplier's original master or export history. Do not denoise, upsample or convert to make provenance appear stronger. Noise/SNR intake, DSP regression and codec-delivery preflight belong to the related workflows; none turns this screen into authenticity certification.