Products prove what a capability can do online. Research FDE enters the field to decide which task matters here, how success will be measured, and how it enters procurement.

Definition

Research FDE is not a fourth software product. It is Zhiway's field adaptation and delivery method: reconstruct the real workflow, define data and knowledge boundaries, configure a domain Loop / Harness, and use concrete tasks to form acceptance criteria, a procurement plan and a budget.

Why a general product cannot finish the job alone

Research institutions do not procure “more intelligence” in the abstract. They need to know whether a specific task can run within existing data, permissions, equipment and accountability. An NLP laboratory cares about corpora, evaluation sets and reproducibility. A materials laboratory cares about instrument data, sample conditions, SOPs and success criteria.

These boundaries are distributed across files, databases, code, instruments, approvals and tacit expert knowledge. A general chat interface cannot turn them into an executable workflow by itself.

The five-step Research FDE loop

  1. Find a high-value task. Start from work that is frequent, costly and verifiable.
  2. Reconstruct the workflow. Record inputs, decisions, tool calls, human approvals, failure branches and deliverables.
  3. Define data and knowledge boundaries. Clarify what can be used, what may leave the domain, which rules enter the system and which judgments remain with experts.
  4. Configure the domain Loop / Harness. Connect Idea Network, private data, models, Research Claw, specialist tools and human review.
  5. Validate with real tasks. Turn accuracy, traceability, time cost and reuse into acceptance criteria, then procurement and budget.

How FDE builds a durable advantage

One delivery does more than complete one task. Research materials, domain ontology, tool adapters, expert corrections and task trajectories become reusable organizational knowledge. The next similar task does not start from zero.

This closes the loop with Idea Network: the field contributes tacit knowledge and real tasks to the infrastructure; the infrastructure lowers the cost of the next field delivery.

Current public progress

As of August 2026, Research FDE has entered an NLP laboratory at a provincial university in North China, a materials-physics laboratory at a science and engineering university in East China, and a materials-engineering laboratory at an applied university in East China. A technical innovation organization and a science and engineering university in North China provide entry points for pilots and collaboration.

These engagements span natural language processing, materials physics and materials engineering, while opening the next round of pilots and joint validation.

Research Claw is the local and privately deployable product base. Research FDE is the method for task selection, domain adaptation and acceptance. They connect, but they are not the same thing.