Paper-Pro Sample Run

End-to-end tested example of a paper-pro workflow, including prompt, artifacts, measured telemetry, and quality checks.

This page documents a real tested paper-pro run from this repository.

Goal

Generate a full paper-shaped Markdown manuscript from an existing local evidence base about citation verification in research, then validate that the workflow produced all required audit artifacts.

Prompt used

/paper-pro notes/citation-verification-research-notes.md.
Use slug citation-verification-proof.
Primary local evidence files are notes/citation-verification-research-notes.md,
notes/citation-verification-research-verified-excerpts.md, and
notes/citation-verification-research-verification.md.
Do not reuse draft, review, citation, compliance, or provenance artifacts from any other slug.
Create fresh architecture, draft, citation-integrity, review, and compliance artifacts for this slug before finalizing.

CLI equivalent:

bohr paper-pro "notes/citation-verification-research-notes.md"

In the proof run, the CLI was forced to openai/gpt-5.4 so the manuscript subagents inherited the same active model.

End-to-end artifact set

The tested run created these slug-specific files:

outputs/.plans/citation-verification-proof-paper-pro.md
outputs/.plans/citation-verification-proof-architecture.md
papers/citation-verification-proof-draft.md
papers/citation-verification-proof-citation-integrity.md
papers/citation-verification-proof-review.md
notes/citation-verification-proof-verification.md
papers/citation-verification-proof-compliance.md
papers/citation-verification-proof.md
papers/citation-verification-proof.provenance.md

Workflow shape

  1. Lead agent inventories the evidence files and writes the run plan.
  2. paper-architect builds the manuscript structure and claim map.
  3. writer creates the draft.
  4. citation-integrity audits claim-to-source support and URL hygiene.
  5. reviewer performs a skeptical manuscript critique.
  6. verifier checks the revised draft against the evidence.
  7. paper-compliance validates required sections and artifact contract.
  8. Lead agent promotes the final paper and writes provenance.

Measured telemetry

Fresh subagent telemetry from the proof run recorded:

MetricValue
Active model for manuscript subagentsopenai/gpt-5.4
Input tokens133,875
Output tokens37,594
Recorded subagent cost$0.99469

Important scope note:

  • This dollar figure comes from the saved subagent metadata only.
  • The top-level orchestration turn is not included in those subagent totals, so the real end-to-end billable cost is slightly higher.

Quality checks applied

The automated and manual proof checks required all of the following:

  • final paper exists at papers/citation-verification-proof.md
  • provenance exists at papers/citation-verification-proof.provenance.md
  • fresh architecture, draft, review, verification, citation-integrity, and compliance artifacts exist and are non-blank
  • the final manuscript contains:
    • Abstract
    • Introduction
    • Related Work
    • Method
    • Evidence
    • Discussion
    • Limitations
    • Conclusion
  • manuscript subagents report successful exit codes with no hidden error payloads
  • manuscript subagents inherit the active OpenAI model instead of falling back to stale Gemini overrides
  • provenance states what was and was not verified, rather than smoothing over gaps

Output quality summary

This sample paper is a narrow, professional synthesis rather than a broad systematic review. The strongest parts of the output are:

  • claims are explicitly bounded to the biomedical-heavy evidence base
  • de Lacey (1985) is kept as historical context only
  • ICMJE is kept as high-level normative context only
  • practical guidance is labeled as an author-proposed heuristic, not a validated intervention
  • provenance records the scripted URL-liveness caveat honestly

Why this sample matters

This run is useful as a release/demo reference because it proves more than “the command exits 0.” It proves that:

  • the workflow can generate a full paper-shaped Markdown manuscript
  • the new paper-architect agent is active in the pipeline
  • the manuscript-quality gates actually run
  • the old blank-output / wrong-provider caveat was removed by forcing manuscript subagents to inherit the active session model

Proof transcript saved during the automated smoke test:

notes/citation-verification-proof-paper-pro-session.txt