Files
openprose__prose/packages/std/ops/prose-author.parallel-research-approval.test.prose.md
2026-05-18 13:00:54 -04:00

2.1 KiB

name, kind, subject
name kind subject
test-prose-author-parallel-research-approval test prose-author

Test Prose Author Parallel Research Approval

Fixtures

  • request: | workflow: weekly competitor intelligence brief input: company names, market segment, cutoff date

    for each company in parallel: research recent product launches, pricing changes, partnerships, hiring signals verify every major claim against at least two sources when possible extract notable moves, source links, confidence, and why it matters

    merge findings into a ranked brief: lead with the top 5 strategic changes include watchlist items separately from confirmed facts call out unknowns and weak evidence

    editor reviews the brief loop up to 3 times until editor approves: editor gives notes on unsupported claims, unclear significance, missing competitors analyst revises the brief and evidence table

    return final brief, source table, unresolved caveats, and editor approval record

Expects

  • source_package: prefers folder output because the workflow includes parallel research, claim verification, synthesis, revision, and approval
  • source_package: maps the input line into ### Requires items for companies, market_segment, and cutoff_date
  • source_package: preserves all domain outputs from the return line: final_brief, source_table, unresolved_caveats, and editor_approval_record
  • source_package: includes explicit fan-out/fan-in choreography for per-company parallel research rather than leaving it only as strategy text
  • source_package: makes claim provenance first class when verification, source links, and confidence are requested
  • source_package: includes a bounded approval loop with max 3 review rounds
  • source_package: declares the exhausted-loop outcome when editor approval is not reached
  • lint_report: has status pass and no blocking findings

Expects Not

  • source_package: collapses the return line into one vague report output
  • source_package: drops source_table or confidence/provenance fields
  • source_package: hides the non-approved final state