Concepts

CoAutoResearch is a human-centered research loop. The agent can do substantial work, but the human remains responsible for direction, claims, evidence, and revision.

Research Loop

        flowchart TD
  H["Human brief or intervention"] --> P["PROJECT.md<br/>research direction"]
  P --> S["STATE.md<br/>current objective"]
  S --> CP["Critical Path<br/>data, method, evaluation, figures, pack"]
  S --> T["Trial<br/>one coherent work package"]
  T --> PL["PLAN.md"]
  PL --> PR["reviews/PLAN_REVIEW.md"]
  PR --> RS["Resource Scout decision/report<br/>when required"]
  RS --> W["Execute in workspace/"]
  W --> RP["REPORT.md"]
  RP --> RA["Reviewer Scope Analyst decision"]
  RA --> RV["reviews/*_REVIEW.md<br/>eight core reviewer files"]
  RP --> F["CURRENT_FINDINGS.md<br/>claims and evidence"]
  F --> PP["PAPER_PLAN.md<br/>obligations and blockers"]
  PP --> BP["BLUEPRINT.md<br/>real full-results handoff"]
  BP --> M["Export gate<br/>Paper Pack or Status Pack"]
  M --> S

  H -. "redirect scope, method, claim, venue, or priority" .-> S
  H -. "formal intervention recorded" .-> T
  CP -. "bottleneck target" .-> T

  classDef human fill:#111411,color:#fffdf6,stroke:#111411;
  classDef truth fill:#f4f7ef,stroke:#9aac9b,color:#20231f;
  classDef trial fill:#fffdf7,stroke:#d8cfbd,color:#20231f;
  classDef output fill:#edf3f6,stroke:#9bb4c2,color:#20231f;

  class H human;
  class P,S,CP,F truth;
  class T,PL,PR,RS,RA,RV,W,RP trial;
  class PP,BP,M output;
    

What the Diagram Means

• The loop is linear enough for a human to follow.

• Every substantive work package gets a plan, report, Resource Scout brief, Reviewer Scope Analyst decision, and eight core review files.

• STATE.md contains a Critical Path table. Trials target the current bottleneck instead of drifting into bookkeeping.

• Claims become current only when reflected in findings and evidence records.

• Human intervention is not an afterthought; it is part of the control system.

• The output should be something the human can understand, defend, and revise.

Critical Path Control

Critical Path items are the dependencies required for a real deliverable: data, methods, evaluation, analysis results, figures, tables, manuscript assembly, and final export readiness. Bookkeeping can support those dependencies, but it is not itself a bottleneck.

If a research-critical resource is missing, the Resource Scout must work through the acquisition/substitution ladder: user-provided ongoing work, local caches, official download, alternate official year/version, alternate public dataset, scope-downgrade proposal, then a concrete human request. Once that ladder is exhausted for the active bottleneck, remaining audits or cleanup do not justify continuing the loop.

Manuscript Deliverables

The manuscript handoff has two artifacts:

Artifact	Purpose
manuscript/PAPER_PLAN.md	Always-current venue-format plan, evidence obligations, planned displays, and blockers.
manuscript/BLUEPRINT.md	Exportable full-results manuscript blueprint containing only real, source-backed sections, results, figures, tables, references, and provenance.

Before results exist, BLUEPRINT.md is only a stub that points to the plan and Critical Path. The UI labels that state as a paper plan or research status, not a paper-writing handoff. The Paper-Writing Pack is available only when the blueprint is non-stub, complete, and free of blocking missing evidence.

Control Surfaces

Surface	What it controls
PROJECT.md	Overall direction, scope, audience, and constraints.
STATE.md	Current objective, active plan, blockers, and reviewer gate state.
Human interventions	Explicit changes to scope, claim, venue, method, or priority.
Reviewer gates	Plan, process, evidence, venue fit, manuscript, figure/table, reference, and final-gate checks.

Autoresearch stops only after strict reviewer pass. A completed trial, approved plan, plausible venue fit, coherent manuscript architecture, or targeted revision-ready draft is progress, not final pass.

Why This Matters

The system is not optimized for autonomous output at all costs. It is optimized for research ownership: the human should be able to explain why a claim exists, where the evidence came from, which alternatives were rejected, and what work remains open.