build.thiri.ai
Solana · Seeker · Cadenza
Whitepaper · v0.1 Draft

T.H.I.R.I. — Whitepaper

The first cryptographic provenance layer for AI-generated music on Solana. Full architectural and economic specification.

Source: docs/WHITEPAPER.md

T.H.I.R.I. — Whitepaper

The First Cryptographic Provenance Layer for AI-Generated Music on Solana.

Authors: Dennison Blackett, Blues Prince Media Version: v0.1 (Draft) · 2026-04-30 Status: Living document — updated weekly through the 12-week sprint

[!IMPORTANT] This is a living draft. The architecture sections track CADENZA_SPEC.md — when the spec changes, this doc changes. The market analysis and economic sections will deepen post-mainnet (Week 0+).


Abstract

AI music generation reached commodity scale in 2025. Suno crossed 12M users; the platform generated 300M tracks in a single year. None of those tracks carry cryptographic provenance. None pay the artists whose recorded work — through public training data, scraped catalogs, and “in the style of” prompts — informed the model’s outputs.

This is not a music problem. It is a settlement infrastructure problem. Existing on-chain music platforms (Sound.xyz, Audius, Royal) optimize for distribution — moving tracks to listeners. None solve the layer underneath: a per-track, per-derivative, per-license audit trail with automated royalty disbursement that is both legally enforceable and rent-economically viable.

We propose Cadenza — a Solana Anchor program providing on-chain mint, split, and license primitives purpose-built for music — and T.H.I.R.I., a 12-app ecosystem of consumer dApps that write into Cadenza without exposing the chain to the user. The wallet is the login. The blockchain is the receipt. The musician sees only their tools.

This whitepaper documents the architecture, economic model, and 12-week ship plan for the first such infrastructure to deploy on the Solana Seeker dApp Store.


1. Problem

1.1 The provenance gap

When a Suno user prompts “soulful R&B with jazz chords in the style of D’Angelo,” the model outputs a track that:

  1. Has no cryptographic record of which artists’ work informed the output.
  2. Has no on-chain owner — the prompter holds a license from Suno; the artists hold nothing.
  3. Cannot be licensed downstream by anyone other than the platform.
  4. Earns the same $0.003-per-stream rate on Spotify as a song that took three weeks to write.

1.2 The settlement gap

Existing music-on-blockchain platforms either (a) ignore derivative provenance entirely, or (b) treat each work as an isolated NFT with no compositional lineage. Neither addresses the core question: when an AI-generated track is created, listened to, remixed, or sampled, who gets paid, in what proportion, and how is that proven?

1.3 Why this is now a tractable problem

Three Solana-specific developments make this viable in 2026:

  • Metaplex Core reduced per-asset rent ~80% versus Token Metadata — making per-track on-chain provenance economically sensible.
  • The Solana Seeker ships hardware-backed wallet signing (Seed Vault) and a curated dApp Store, removing both the security and distribution barriers that historically kept on-chain music apps off mobile.
  • 400ms finality lets a “Mint Provenance” tap during playback feel instantaneous — closing the UX gap with web2.

2. Cadenza — the on-chain rail

Cadenza is a single Solana Anchor program with five PDA types and fourteen instructions. It is not a marketplace, not a token, and not a competitor to Metaplex Core — it adds a music-specific metadata and settlement layer on top of Core.

2.1 The five accounts

AccountPurpose
PublisherNFTOne per developer/label; identity for the entity authoring mints
CadenzaRecordOne per track; carries music-specific metadata (key, tempo, fingerprint hash, license terms, royalty bps)
SplitConfigUp to 10 recipients with basis-point shares summing to 10,000; lockable; permissionless distribute-crank
LicenseGrantOne per (asset, licensee); records on-chain license terms and payment
FingerprintRefOne per artist wallet; aggregated harmonic fingerprint hash with session count

Full schemas, instruction set, security model, and SDK surface in Cadenza Spec.

2.2 Why basis points, not percentages

The original prototype (Genesis) used u8 percentages summing to 100 — making fractional splits like 12.5% impossible. Cadenza uses u16 basis points summing to 10,000, the music industry standard since the 1976 Copyright Act.

2.3 Why Cadenza does not record streams on-chain

Each StreamRecord PDA in the original prototype cost ~0.0018 SOL of rent. At Spotify’s $0.003/stream payout, the rent burned the royalty 600× over. Cadenza records provenance events (mint, license grant, derivative creation) on-chain, and leaves stream play counts to off-chain indexers. The cryptographic guarantee lives where it is economically tractable.


3. T.H.I.R.I. — the consumer surface

The 12-app ecosystem is structured in three phases over a 12-week ship sequence (May 5 → August 11, 2026).

Phase 1 · Prove Execution (Weeks 0–3)

WeekAppFunction
0Mainnet FoundationDeploy Cadenza · mint Publisher NFT · publish SDK
1Mint StudioArtist mints a music NFT with Cadenza metadata
2Royalty ConsolePWA dashboard for earnings + portfolio
3Collab SplitMulti-wallet royalty splits (≤10 recipients)

Phase 2 · Intelligence Surfaces (Weeks 4–8)

WeekAppFunction
4Stem SplitterOn-device 4-stem separation (Demucs ONNX)
5Key Finder10s audio → detected key
6Chord AnalyzerReal-time chord progression timeline
7Ear TrainerInterval/chord recognition drills
8 ⭐Player THIRIMobile player with real-time WoodShed analysis + Cadenza provenance — grant submission target

Phase 3 · Marketplace + Identity (Weeks 9–12)

WeekAppFunction
9Key TransposerAudio + sheet music transposition
10Setlist PlannerKey/tempo flow analysis across tracks
11Fan MarketTwo-sided marketplace; license purchase → SplitConfig disbursement
12Skill LoggerCross-app identity capstone; aggregates FingerprintRef

Full week-by-week ship gates and dependencies in Seeker Execution Plan.


4. Economic model

4.1 Fees

OperationFeeRecipient
mint_cadenza_record0.01 SOLBPM treasury
distribute_split1% of disbursementBPM treasury
grant_license2.5% of priceBPM treasury
Publisher init · split create · fingerprint registerFree (rent only)

4.2 Conservative revenue projection (Month 1 post-mainnet)

  • Mint Studio: $500–$1,200 MRR at 50–100 mints
  • Collab Split: incidental
  • Fan Market: not live until Week 11
  • THIRI freemium upgrades: $4.99/mo

Month 4 target (post-grant): $5,000–$12,000 MRR.

4.3 Why this is sustainable beyond grant funding

The fee structure compounds with usage. Once Cadenza has 10,000 active CadenzaRecords, even 1 license grant per record per year at $1 average = $250/year in protocol fees with effectively zero marginal cost. The grant accelerates the cold start; it does not subsidize the long-term operation.


5. Open source commitment

ComponentLicense
Artist Fingerprint SDKMIT
Cadenza TypeScript SDKMIT
Cadenza IDL (Anchor)MIT
WoodShed EngineProprietary (Blues Prince Media)
App source codeProprietary

Anyone can build apps that read/write Cadenza accounts and compute artist fingerprints. The intelligence engine (WoodShed) and consumer apps remain proprietary — this is the sustainable business model.


6. Risks and mitigations

6.1 Solo-builder bandwidth

12 apps in 12 weeks is aggressive. Mitigation: 7 of the 12 are thin SDK wrappers around Cadenza; the heavy lifts (Cadenza itself, Stem Splitter, Player THIRI) are time-boxed and de-scoped if they slip.

6.2 dApp Store review queues

Mitigation: submit Phase 1 apps for review at Week 3, not Week 8 — review queue ≠ build queue.

6.3 On-device ONNX inference may not fit Seeker hardware

Mitigation: server-side fallback path documented; v1 ships with whatever runs on device, v2 expands.

6.4 Brand collision with prior app-cadence-live repo

Mitigation: full rename to Cadenza completed pre-mainnet (April 2026). See Genesis for the lineage.


7. References


Blues Prince Media · Building T.H.I.R.I. · 2026