Question To DAG Demo

Turn an ambiguous problem into a buildable graph plan.

Jaysearch accepts a question, emits bounded evidence, creates candidate DAGs, selects the strongest graph, and materializes execution units. Development automation starts after this boundary.

GitHub Repository Project Overview Design Note

Run It

One command produces the demo artifacts

bin/run-jaysearch-question-dag-demo \
  --root . \
  --question "How should we build candidate generation?"

Each run writes packet artifacts and a local demo.html view under artifacts/demo/question-dag-runs/<run_id>/.

Flow

The lineage stays inspectable

flowchart LR
  Q["User question"]
  RQ["research_question_packet"]
  EV["evidence_packet
bounded fixture mode"]
  RP["research_problem_packet"]
  HY["research_hypothesis_packet[]"]
  RC["research_candidate_packet[]"]
  RR["research_recommendation_packet"]
  CDM["candidate DAG manifest"]
  SEL["candidate DAG selection"]
  DAG["selected DAG"]
  EUM["execution-unit manifest"]
  EU["execution_unit[]"]
  STOP["implementation boundary"]

  Q --> RQ --> EV --> RP --> HY --> RC --> RR --> CDM --> SEL --> DAG --> EUM --> EU --> STOP

Question

The user goal becomes a bounded research question packet with decision target, constraints, evaluation targets, and artifact targets.

Evidence

V1 emits a contract-compatible evidence packet in bounded fixture mode so the demo is deterministic. The architecture keeps source lineage visible for later live research integration.

DAG

Candidate graph plans are adapted from ranked research candidates, rejected alternatives stay visible, and the selected DAG is recorded with selection evidence.

Units

The selected DAG becomes execution-unit packets. That is the handoff point for implementation planning and future autonomous development.

Honest Boundary

What this does not claim yet

This demo does not perform live autonomous web research, code synthesis, patch application, or production execution. Those are explicit next slices. The current claim is narrower and useful: Jaysearch can turn a problem into a traceable, evidence-backed graph plan and execution-unit handoff.