What gets embedded
When a fan buys a book or a track here, their copy isn't identical to every other copy: a unique, cryptographically-signed identifier is woven into the file at the moment of sale, tied to that specific transaction. This is the subject of our pending U.S. provisional patent — here's what sets it apart, without walking through the mechanism itself. Text and audio carry a real embedded mark today; video, games, and immersive content are on the roadmap as the catalog expands into those formats.
Built together, not bolted on
Conventional forensic watermarking ties a leaked file to the account that downloaded it — standard practice for over a decade. We built that same per-sale mark alongside two more pieces most platforms keep separate or skip entirely: a tamper-evident, hash-chained receipt log that makes "this was a genuine sale" an auditable fact instead of a platform's word for it, and a verified-consumption gate that only lets someone rate or review a title after their account has actually opened a copy they paid for. Three purpose-built pieces working together today, for text and audio — piracy tracing and review/payout-fraud resistance — with video, games, and immersive content the next formats in line, each at a strength appropriate to its format, detailed below.
Robustness today, honestly scoped
Most watermarking is all-or-nothing: strip enough and the identifier is gone. Ours starts from a redundancy floor rather than a single point of failure, and how far that goes depends on the format. Text carries the mark at two points in the file, so either surviving fragment can recover it — though we haven't yet tested it against reflow, paste-into-another-editor, or OCR. Uncompressed audio (WAV) carries a real sample-level mark that survives plain copying and repackaging, but needs the whole file, not an excerpt, and doesn't survive lossy re-encoding, resampling, or normalization. MP3 currently carries a metadata tag only, which any re-encode strips. Making every format hold up to a crop or a format conversion the way the full patent design calls for is hardening work ahead of us, not something already shipped.
Where detection is heading
Today's extraction is deterministic and format-specific — a fixed pattern match for text, a majority-vote decode for uncompressed audio, a metadata read for MP3 — not a system that has learned to recognize transformations it's never seen. Broader, adaptive resilience across every format — surviving crops, re-encodes, and remixes more reliably — is the next stage of this work, not something running today.
What it doesn't claim
No attribution system is unbreakable, and we won't tell you otherwise. This is a meaningfully harder target than the watermarking most platforms use today — not a promise that piracy becomes impossible.