Skip to content

Latest commit

 

History

History
60 lines (44 loc) · 2.24 KB

File metadata and controls

60 lines (44 loc) · 2.24 KB

Research Workflow Guide (Extended)

Phase 1: Idea and Framing

  • Run /start and /research-idea
  • Input: topic hints, constraints, and target venue direction
  • Define topic, hypothesis, target audience, and success metrics
  • Output: research/ideas/ and research/hypotheses/ artifacts

Phase 2: Setup and Backend

  • Run /setup-researchclaw
  • Input: AutoResearchClaw install path and execution environment
  • Ensure researchclaw is available with validated config
  • Standard minimal test command order:
    • If config.arc.yaml is missing: researchclaw init
    • Export or confirm the API key environment variable required by llm.api_key_env
    • Run researchclaw doctor
    • Run researchclaw validate
    • Proceed to the next workflow step only after both checks pass
  • Output: reproducible runtime config snapshot

Phase 3: Literature and Data

  • Run /citation-audit and /data-intake
  • Run /data-clean after source freeze
  • Input: source list, acquisition constraints, and raw datasets
  • Output: curated data, lineage manifest, and citation-quality report

Phase 4: Experiments and Analysis

  • Run /research-run and /analysis-run
  • Input: curated datasets, run config, and metric definitions
  • Keep run logs, metrics definitions, uncertainty reports, and robustness notes
  • Output: artifacts/ runs and analysis summaries under research/results/

Phase 5: Figures and Writing

  • Run /figure-build, /figure-qc, and drafting workflows
  • Run /paper-style-sanity (warn-only) and /paper-gate-check
  • Input: analysis outputs, figure specs, and manuscript draft
  • Output: traceable figures, coherent draft, gate report (PASS/CONCERNS/FAIL)

Phase 6: Reproducibility and Submission

  • Run /repro-audit and /review-report
  • Run /latex-venue-pack and /latex-build-check
  • Input: gated draft, figure bundle, bibliography, and venue requirements
  • Output: venue-ready submission package in deliverables/submission/

Gate Semantics

  • PASS: all required artifacts and checks are complete
  • CONCERNS: submission possible with documented risks
  • FAIL: critical integrity/reproducibility blockers remain unresolved

Companion Docs

  • Conversation templates: docs/examples/
  • Stable checklists and conventions: docs/research-reference/