Define Tensor role in the Culture research flow

[OriNachum]

Goal

Define tensor-cli as the multidimensional thinker in the split-agent Culture research workflow.

Position in the flow

arxivist -> tensor-cli -> reduce-cli -> prove-cli -> eidetic-cli

Culture chat coordinates specialized agents in a room/thread. Tensor should not become a mega-agent; it should own deep idea-space mapping.

Tensor responsibility

tensor-cli connects dots and expands the conceptual space:

• deep-read a paper brief or source excerpt
• map the idea-space of a paper/system/design
• identify core concepts and relationships
• surface novelty, implications, assumptions, and risks
• generate candidate conjectures and candidate lemmas
• hand complex claims to reduce-cli for decomposition

Non-goals

• Tensor does not manage the paper queue; arxivist does.
• Tensor does not break every complex claim into atomic units; reduce-cli does.
• Tensor does not prove/refute claims; prove-cli does.
• Tensor does not own durable memory; eidetic-cli does.

Proposed artifacts

paper.tensor.md
idea-space.md
conjectures.yaml
candidate-lemmas.yaml
implications.md

Culture chat protocol sketch

arxivist -> tensor:
  Map the idea-space for paper P.

tensor -> reduce:
  Here are complex claims and candidate lemmas that need reduction.

Acceptance criteria

• Define Tensor input contract from arxivist briefs
• Define Tensor output artifact schema
• Define handoff format from Tensor to Reduce
• Decide whether tensor should emit conjectures, candidate lemmas, or both
• Document Tensor as the specialist that expands and connects idea-space

Core sentence

Tensor connects dimensions and turns paper briefs into idea-space maps, conjectures, and candidate lemmas.