The Examiner does not ask an AI whether a track sounds artificial. It measures the physics of the performance itself: pitch drift (a human voice or hand never sits perfectly on a note), micro-timing (people land slightly off the grid, machines land on it), note-duration variance, spectral flatness (a recorded room versus a reconstructed signal), and stereo phase correlation (real space versus synthetic width). Each measurement is published with the weight it carried, combined into a 0–100 human-performance index, and stated as a plain verdict: Claimable, Partial claim, or Must exclude.

It is deterministic — the same file always produces the same numbers — and it runs entirely inside the browser, so no audio ever leaves the device. Nothing here is a black-box score. Every figure can be checked, repeated, and put in front of an examiner at the Copyright Office, which requires AI-generated material to be disclaimed and decides every application itself.

A modern record is rarely all-human or all-machine — it is a human vocal over generated drums, or a played bassline under synthesized keys. Stem mode splits the track into vocals, drums, bass and a residual using a real source-separation model (LALAL.ai), then runs the full Examiner on each part independently. The result is a per-part verdict: exactly which layers a person made, which were generated, and therefore exactly what can be claimed and what must be disclaimed — the filing language, written from the evidence.

Straight talk, because investors deserve it: separation quality is the model's, bleed between stems is normal and shows up in the numbers, and no system on earth can identify which person played a part. What this does is tell human from machine, layer by layer, with measurements you can show anyone.

The console the Examiner lives in is a full mastering chain running in 32-bit float: valve saturation, three-band shelving EQ, mid/side stereo width, glue compression, a 3D spatial field with convolution studio rooms, and ten audio-reactive visualizers including a WebGL 3D core. Around that core sit the Surreal stages — restoration and finishing tools that are true-bypass by design: a denoiser that learns a track's own noise floor, a split-band de-esser, Haas cross-feed and early reflections, a transient shaper, a sub bump, codec restore for crushed streaming rips, and a true-peak loudness maximizer with a 4×-oversampled detector that holds a streaming-safe ceiling exactly.

Two guarantees make it institutional rather than hobbyist. First, every stage is verified bit-identical when off: a master you already trust never changes unless you change it. Second, every export — the four-valve float32 prints and the original / master / mono bounce — is SHA-256 hashed into a forensic ledger at the moment it is made, with live true-peak, RMS and clip telemetry on the output. The engine does not just make the record sound better; it produces the receipt.

The Master Trust settles money on the basis of who owns what. That claim is only as strong as the evidence beneath it. By examining, restoring, hashing and sealing every record at the point of entry, the engine gives the protocol what no royalty system has ever had: a provenance record created before the first dollar moves, attached to the waveform itself. Registrable works get registered. Generated layers get disclosed and paid under the Fair Rules split. Nothing is guessed, and nothing is hidden. That is the difference between a tool and infrastructure.

The move: make the Examiner the default step before a music copyright registration — the tool you run a record through before you file.

Who we approach: not the majors. DistroKid — the largest independent distributor, moving new releases at the volume where the disclosure problem bites hardest — as the first distribution partner, with Bradley Spalter's own catalog as the founding proof case. The proof exists before the first outside conversation: a demonstration, not a promise.

Take the square: the founder's catalog examined, filed and sealed end to end; DistroKid engaged on a pilot; the methodology published for audit. Real filings, real proof.

The offer: a launch-partner window — first and alone with the Examiner and sealed-authorship filing for a defined period, their edge over every other distributor while it lasts. When the window closes it opens to the industry as the standard. First-mover advantage, not a permanent lock.

The move: turn the working tool into an open, documented protocol and bind it to the rails the industry already trusts — C2PA for provenance, ISRC/ISWC and DDEX for identity. Stand up neutral governance.

Who we approach: standards bodies (C2PA, DDEX), the PRO and MLC world, and a second-vertical adopter to prove it travels.

Take the square: Spalter is the reference implementation and the trusted operator of an open standard — the name on it, not the cage around it.

The move: extend the same provenance layer into film, VFX and video games — where AI-authorship and asset provenance is a bigger, hotter, more unsolved problem than in music.

Who we approach: a game studio or engine, and a film/VFX post house or a guild-aligned partner (synthetic-performer and SAG-AFTRA AI terms are the opening).

Take the square: cross-industry proof — one standard spanning music, film and games, carried in on music proof, never a cold pitch.

The move: once provenance is trusted and adopted, the settlement layer — the SSP ledger, split escrow, and the fiat off-ramp — becomes the default rail, because it is already attached to the records everyone references.

Who we approach: the installed base won in Phases I–III — every rights-holder, distributor and platform already citing a Spalter-proven record.

Take the square: recurring infrastructure revenue on top of an installed base whose trust we already own — the durable business.

The rounds are not a hope for runway — each one funds a named phase and produces the proof that unlocks the next. Capital is fuel for a campaign already mapped, not a bet on one yet to be drawn.

Investors are wired into the distribution waterfall: capital returns programmatically as protocol revenue is realized — as an advance recouped by the same rails that pay everyone else, never as a perpetual cut of creators' earnings.