Problem
Large AI-era tasks need more than generation: they need planning, decomposition, tool use, verification, expert review, revision, and accountable integration. The total token and compute budget can exceed one user or lab.
Shared token economics for useful work
Compose idle tokens, agents, tools, and people into verifiable task graphs.
TokenMosaic is a research protocol for turning distributed AI capacity into auditable public-good computation: decompose, assign, verify, merge, recover, and settle.
Paper Core
From proof-of-work to proof-of-useful-work.
Resource
Personal devices, local GPUs, model credits, browser tools, human annotators, and user-owned agents often sit idle. TokenMosaic treats this spare capacity as a shared resource pool.
Protocol
A large request becomes a task tree or DAG. Atomic units are routed to contributors, locally validated, globally verified, recursively merged, and recorded in a contribution ledger.
Protocol View
Every accepted answer carries its route, evidence, and settlement trail.
Task Graph
Decomposition makes the work parallel; verification makes it safe to merge.
The protocol does not assume one universal solver. Each task family supplies a decomposition function, executor contract, verifier, and merge rule. The shared layer handles leases, retries, provenance, canonical output selection, and settlement.
Incentive Loop
Rewards attach to verified contribution, not mere participation.
Clients commit to assigned work, submit artifacts and metadata, and receive rewards when their outputs pass verification and merge checks. Failed, late, or malicious outputs can be rejected, reassigned, or penalized.
Design Desiderata
Five constraints define when a task fits TokenMosaic.
01
Decomposable
Clear task units, dependencies, and merge relations.
02
Verifiable
Each unit has a checker, quorum, rubric, or formal validator.
03
Extendable
Task-specific logic stays in plugins, not the protocol core.
04
Scalable
More contributors improve speed, quality, redundancy, or recovery.
05
Autonomous
Routine assignment, retry, merge, and settlement follow rules.
Repository
The public code release is a local protocol prototype.
Included
Python protocol engine, event ledger, SQLite projections, artifact hashes, factorization plugin, Lean proof plugin, tests, and experiment runners.
Not Included Yet
The paper PDF is not uploaded in this release. Production chain, wallet, marketplace, and public worker network code are outside the current prototype.