5.2 Attribution and Royalty System
Royalty Formula Mechanics
The multi-generation waterfall formula distributes revenue across ancestor models proportionally to their distance and contribution.
The formula takes the form:
Each parameter serves a specific purpose.
BaseRate (default 5%) establishes the initial royalty percentage, aligned with patent licensing (3-5%) and music streaming royalties (5-10%). Creators can customize this rate or set it to zero for fully open-source models where attribution suffices without financial compensation. DecayRate (default 0.5) reduces royalty by half for each generation step, preventing excessive burden on models with deep lineage while maintaining meaningful attribution to base model creators. The exponent N represents generation depth, starting at 0 for immediate parent, incrementing to 1 for grandparent, 2 for great-grandparent, and so forth. ContributionWeight (range 0.0-1.0) splits royalties for merged models with multiple parents, summing to 1.0 across all parents to ensure total royalty matches single-parent scenarios. Revenue represents total income from the derivative model through licensing fees, API usage charges, or commercial deployment contracts.
A total cap of 20% prevents unbounded royalty obligations.
If a model has 20 generations of ancestry, the formula might calculate 25% total royalty burden. The cap enforcement stops at 20%, protecting derivative creators from excessive obligations while still rewarding recent ancestors substantially through the bulk of the 20% pool. This cap applies after summing all ancestor royalties but before distribution, ensuring derivative creators never face surprise obligations that exceed the advertised maximum.
The exponential decay reflects diminishing influence as derivatives move further from original work.
When Model C derives from Model B (which itself derives from Model A), Model B contributed directly to Model C through its weights and capabilities. Model A contributed indirectly through Model B, making its influence real but attenuated. The 0.5 decay rate means Model B receives 5% while Model A receives 2.5% when Model C generates revenue. If Model D then derives from Model C, the chain extends further: Model C gets 5%, Model B gets 2.5%, Model A gets 1.25%. Each generation receives half what the previous generation received, naturally limiting total burden while preserving attribution chains that can stretch across many derivative steps.
Calculation Examples
Example 1: Simple Linear Chain
Model A represents a base model trained from scratch. Model B fine-tunes Model A for medical applications.
Model C further specializes Model B for radiology diagnostics.
When Model C generates $10,000 in licensing revenue, the distribution works as follows: Model B's creator (generation 0 from C's perspective) receives $10,000 × 5% × 0.5^0 × 1.0 = $500. Model A's creator (generation 1 from C's perspective) receives $10,000 × 5% × 0.5^1 × 1.0 = $250. Total royalties equal $750 (7.5%), leaving 92.5% ($9,250) for Model C's creator who added the final specialization.
Example 2: Multi-Parent Merge
Model A specializes in medical knowledge. Model B specializes in radiology imaging. Model C merges them through weight averaging (60% Model A, 40% Model B).
Model D fine-tunes this merged model for pediatric radiology.
When Model D generates $10,000 revenue, the distribution becomes: Model C's creator (direct parent) receives $10,000 × 5% × 0.5^0 × 1.0 = $500. Model A's creator (grandparent through C) receives $10,000 × 5% × 0.5^1 × 0.6 = $150, with the contribution weight reflecting Model A's 60% influence on the merge. Model B's creator (grandparent through C) receives $10,000 × 5% × 0.5^1 × 0.4 = $100, proportional to their 40% contribution. Total royalties equal $750 ($500 + $150 + $100), which is 7.5%, leaving Model D's creator with $9,250 (92.5%).
Example 3: Deep Lineage with Cap
A model with 6 generations of ancestors would calculate royalties as: Gen 0: 5%, Gen 1: 2.5%, Gen 2: 1.25%, Gen 3: 0.625%, Gen 4: 0.3125%, Gen 5: 0.15625%.
Summing these gives 9.84375%.
All ancestors receive their calculated amounts since the total stays below the 20% cap. If the model had 10 generations, the uncapped sum would approach 10% (the geometric series limit), still remaining under the cap. Only with custom parameters (higher base rates or slower decay) would the cap trigger, at which point the smart contract proportionally reduces all payments to fit within 20% while preserving relative attribution weights.
Open Source Opt-Out and Comparisons
Creators embracing open-source principles can set BaseRate to zero during registration.
This makes the model free to use and modify commercially without royalty obligations. Meta's Llama 2 exemplifies this approach: release it with zero royalty rate, enabling unrestricted derivative works while maintaining lineage visibility for discovery and attribution. Developers see that their fine-tuned Llama models descend from Meta's work, providing attribution and discovery benefits without financial obligations.
The zero rate is immutable once set, providing certainty to downstream users that royalties won't be imposed retroactively.
Comparison to Traditional Royalties
Music streaming services like Spotify distribute approximately 70% of revenue to rights holders, then split this among songwriters, performers, and labels according to negotiated contracts. Multi-level attribution exists but requires manual tracking and industry-specific databases (ASCAP, BMI, PROs).
Origyn automates this with smart contracts and transparent calculation rules.
Patent licensing typically charges 3-5% royalties for technology use, with sublicensing clauses allowing derivative patents to reference original patents. However, tracking sublicense usage requires contractual mechanisms and trust in self-reporting, creating friction and disputes. Disputes arise when companies use patented technology in derivative works without proper licensing or deliberately obscure the lineage to avoid royalty obligations.
Origyn makes this transparent and automatic.
Derivative model registration explicitly declares parent relationships as a protocol requirement. Royalty calculations execute on-chain according to immutable formulas. Payment happens atomically when revenue is distributed, with smart contracts enforcing attribution without requiring trust between parties. This reduces friction, eliminates disputes over calculation methods, and enables micro-royalty payments that would be uneconomical through traditional legal agreements where legal fees might exceed the royalty amounts themselves.
Last updated