PRO rights-reconciliation & catalog pipeline
✓ Client data normalization, discrepancy rules, and publisher review workflow
Role & contribution: Sole full-stack contract engineer collaborating directly with Unisound product and music-publishing stakeholders.
Status & usage: Used internally by Unisound stakeholders to ingest ASCAP/BMI/SESAC catalogs and triage registration discrepancies before publisher review.
Problem & workflow context
Music publishers auditing song registrations across ASCAP, BMI, and SESAC had to manually reconcile inconsistent society exports—where writer IPI identifiers, split percentages, and alternate titles differed by catalog—causing missed ownership mismatches and delayed royalty collection.
Decision & rejected alternative
Built a deterministic per-PRO Factory/Adapter ingestion pipeline writing immutable, content-hashed catalog snapshots to PostgreSQL/Prisma, invoking structured LLM fallback strictly when malformed rows failed deterministic parsing. Rejected an LLM-first extraction pipeline (non-deterministic across recurring scans and slow/costly on full catalogs) and rejected mutable in-place row overwrites (which erased audit history between publisher review cycles).
Failure encountered & diagnosis
Early cross-PRO runs generated noisy false-positive discrepancy flags on valid works because societies formatted writer names differently ("LAST, FIRST" vs "FIRST M. LAST") and rounded 3-way splits ("33.33%" vs "33.34%"). Diagnosed by diffing flagged false positives against publisher ground-truth spreadsheets, then added canonical title/IPI normalization and integer basis-point split tolerances prior to hashing.
Verification, role & current status
Verified each society adapter, basis-point split rule, and snapshot hash transition against real multi-PRO publisher catalog fixtures in Dockerized CI/CD, iterating on flag severity directly with stakeholder reviewers before contract handoff.
