Overall design principles
The roadmap follows one principle: every phase must stand on its own. Later phases do not subsidize earlier ones. Each phase produces real revenue or real growth, so that even if the external environment shifts, the network does not collapse.
- There are clear capability dependencies between phases, but a phase does not have to be fully mature before the next one begins piloting;
- Every phase sets verifiable quantitative milestones rather than a vague promise to "scale";
- Commercialization capability is built early, to avoid depending on funding too soon.
Phase 01 (2026): Content-incentive foundation
Goal: use the lightest entertainment scenarios to accumulate a schedulable pool of human users at scale.
Key moves: launch novels / short dramas / mini-games; build a unified points and account middle layer; connect the AD Reward incentive loop; validate unit economics in the first regional markets (Southeast Asia, Latin America).
Pass criteria: content-side active scale, node conversion rate and daily schedulable task throughput all hit target.
Phase 02 (2026–2027): The AI human judgment core
Goal: ship Human Judge and build the scarce, network-wide capability of human preference judgment.
Key moves: launch the human preference evaluation system; open the AI task SDK and build an in-app task center; establish the node qualification system and gold-standard quality checks; sign the first preference-data partnerships with AI companies.
Pass criteria: judgment-task consistency, response latency and unit cost all reach an enterprise-usable line.
This is the phase where the moat forms. Whoever first builds a large-scale, high-quality, low-cost supply of human judgment occupies the AI alignment layer.
Phase 03 (2027): Human tasks as a service
Goal: package human execution capability as a standard cloud API and close the commercialization loop.
Key moves: ship four standard APIs (answer ranking / quality evaluation / data labeling / human verification); establish SLA, audit logging and compliance systems; run the pay-per-use model for AI companies and agent vendors; accumulate enterprise case studies.
Pass criteria: API revenue share, enterprise customer retention and per-customer call volume.
Phase 04 (2028): Global compute ecosystem
Goal: expand global nodes, serve global customers, and become a stable presence at the AI infrastructure layer.
Key moves: compliant node deployment across multiple global regions; multilingual and cross-cultural preference adaptation; native integrations with major AI platforms; institutional funding to support infrastructure-level investment.
Pass criteria: number of regions covered, cross-language task consistency and overall network availability.
Risks and responses
Three main risks deserve an honest statement: supply quality fluctuation (addressed with multi-layer quality checks and node tiering), compliance divergence (labor and data rules differ by region, so we operate through local entities), and demand-side cyclicality (AI company budgets fluctuate, so we diversify across industries).
FAQ
Are the four phases strictly sequential?
No. Adjacent phases overlap. For example, the Phase 02 pilot of judgment capability starts in parallel with Phase 01 scaling, in order to shorten the overall cycle.
Can enterprises integrate right now?
Yes. Early access for partner enterprises is already open during Phase 02, which suits teams that want to help shape the evaluation protocol.
Why is global expansion left for the last phase?
Because globalization is an amplifier applied after capability matures, not a starting point. Expanding into new regions before unit economics and the quality system are validated magnifies problems rather than solving them.
Will the roadmap change?
Yes. Both technological progress and the regulatory environment can affect timing. This page will be updated accordingly, and we will explain the reasons behind any adjustment.