Skip to content

Documentation / THOR Data

14 / 17

Participation

Private beta with qualified firms, then permissionless annotation.

The private pilot is THOR's annotation beta, starting with qualified, whitelisted professional firms. After beta, annotation will open to permissionless participation.

Why start with a whitelist?

The beta qualifies firms for the contracts THOR needs to deliver. The team assesses relevant skills, available capacity, quality-control processes and the ability to handle approved material before assigning work. Each submission remains subject to the buyer's acceptance criteria.

Why start with a private pilot? explains the demand-first rationale, funded worker obligations and transition to permissionless annotation after beta.

Pilot contributors

Each participating firm can operate multiple miner identities (UIDs) and contribute to project batches under agreed briefs and commercial terms. Access to source material remains specific to the assignment.

THOR will publish eligibility rules and explain how additional firms can qualify and when the whitelist will be reviewed. The workflow sets out the acceptance, payment and dispute information published before work starts.

Professional annotation firms can apply to mine. The form asks about company details, production and QA leads, capacity, relevant experience and evidence of past work. Admission and assignments follow qualification and the applicable project terms.

The action-captioning example illustrates the initial work. Challenge-specific thresholds, production tooling and onboarding instructions are not specified here.

Workers and operators

The mechanism separates a worker account from a registered miner UID. An account records the worker's assignments, review history and earned balance. Workers can contribute directly or through an operator, subject to participation rules. Operators coordinate workers through registered UIDs; the mechanism supports worker accounts with or without their own UID.

Treasury-funded payments are recorded for each worker. Assignments and earned balances stay with the worker account if an operator's UID is replaced. During beta, work remains restricted to approved firms and their authorised workers.

Researchers and buyers

Buyers define the result they need and agree how it will be accepted. Researchers can shape useful tasks and evaluate whether delivered data supports the intended learning goal.

Producers and reviewers have different responsibilities. Quality and verification explains the review methods; Data rights and access explains permissions for private material.

Preparing to participate

  • Annotation firms: prepare relevant work samples, available capacity, a review process and arrangements for restricted material, then use the application above.
  • Buyers: use the brief checklist to define the task, available material, required output and acceptance criteria.
  • Researchers: bring a concrete research question, an evaluation method and the intended public output, including access or licensing constraints.

Permissionless annotation after beta

The beta establishes the production and verification workflow for permissionless annotation within the Data stage.

Permissionless participation still requires following task specifications. Access to private customer datasets remains governed by each project's permissions.

What must be defined as access expands

Each job needs clear eligibility conditions, assignment rules and a way to address errors or disputes. The mechanism uses gold-task accuracy, buyer acceptance history and capacity to allocate future work, including opportunities for new contributors. Quality and verification explains the approach. Governance and transparency addresses responsibility for the rules.