Evidence-first research workflow for evidence discovery, source verification, and citation grounding. Use when the task requires searching, verifying, and organizing real papers, datasets, benchmarks, case studies, and project artifacts without fabricating results, citations, or claims.
Sopaper Evidence is an evidence-first research skill. Its job is to build a reliable evidence pack before supporting any downstream paper outline, abstract, related work summary, experiment plan, or draft section.
Version: v1.0.0
Canonical repository: https://github.com/sheepxux/SoPaper-Evidence
This published skill bundle includes the helper scripts it references under scripts/. The GitHub repository remains the public source of truth for releases, examples, and issue tracking.
Use the highest-quality source available for each claim.
Read references/source-priority.md when source quality or conflicts matter. Read references/input-schemas.md when stronger input structure is needed before running the workflow.
Collect or infer:
If the project scope is unclear, produce a short working scope and label assumptions.
Search for:
For each source, capture the title, URL or path, source type, and why it matters.
Use references/prior-work-search-playbook.md for a repeatable search process. For OpenClaw-specific work, use references/openclaw-evidence-playbook.md.
For each evidence item, classify it as:
verified_factproject_evidenceinferenceunverifiedDo not merge these labels. If a statement depends on inference, say so explicitly.
Use the schema in references/evidence-schema.md.
At minimum, extract:
Organize findings into:
related_workdatasets_and_benchmarksbaselinescase_studiesproject_resultsclaim_to_evidenceevidence_gapsUse assets/claim-evidence-map-template.md when the user needs a reusable deliverable.
Use assets/related-work-matrix-template.md when comparing papers, baselines, and benchmark coverage.
Use assets/experiment-gap-report-template.md when the task requires prioritizing missing experiments before drafting.
Use bundled scripts/build_evidence_ledger.py when the user already has markdown notes or source lists and needs a first-pass evidence ledger.
Use bundled scripts/generate_search_plan.py when the user starts only with a topic and needs a first-pass evidence search plan.
Use bundled scripts/generate_topic_claims.py when the user starts only with a topic and needs a cautious structured claims draft.
Use bundled scripts/search_external_sources.py when the user needs a first-pass source list from a topic or search plan.
Use bundled scripts/fetch_external_sources.py when raw URLs should be converted into structured source-note drafts before review.
Use bundled scripts/verify_source_notes.py when fetched notes should be conservatively upgraded into page-level verified facts or reviewed primary-source summaries before entering the ledger.
Use bundled scripts/run_evidence_pipeline.py when the user already has source files, claims, and optional result artifacts and wants one end-to-end draft pack. Result artifacts may be structured markdown, .csv, .tsv, or .json, and multiple result artifacts can be fused into aggregate project evidence.
Use bundled scripts/bootstrap_claim_map.py when the user already has a claims list and a ledger draft and needs a first-pass claim map.
Use bundled scripts/triage_evidence_gaps.py when the user needs a first-pass blocker/major/minor gap report from the current claims and evidence ledger.
Use bundled scripts/review_comparison_fairness.py when the user needs a dedicated fairness check on comparative claims, baseline breadth, metric grounding, and scope alignment.
Use bundled scripts/run_topic_evidence_pipeline.py when the user wants the full topic-driven workflow from theme to search plan, source list, fetched notes, ledger, claim map, and gap report.
Use bundled scripts/validate_input_bundle.py when the user has partially structured inputs and needs a quick schema check before running the pipeline.
Only after the evidence map is complete, support tasks such as:
Before writing, run the checks in references/claim-audit-rules.md. Use assets/paper-outline-from-evidence-template.md when the user needs a draft-safe paper structure.
Unless the user asks for something else, default to this output shape:
Evidence briefKey sourcesClaim-to-evidence mapEvidence gapsSafe writing notesExperiment gap report when blocker gaps existSee the example set in:
When supporting downstream paper writing:
When the user is working on OpenClaw or a similar embodied AI / robotics project, prioritize:
Do not assume OpenClaw has capabilities, datasets, or benchmark wins unless they are present in project artifacts or verified sources. Use references/benchmark-baseline-checklist.md before accepting benchmark-fit or baseline coverage claims. Use references/evidence-gap-triage.md when deciding whether to keep drafting or stop and report blockers.