Evidence-first research workflow for paper writing. Use when the task requires searching, verifying, and organizing real papers, datasets, benchmarks, case s...
Sopaper Evidence is an evidence-first research skill. Its job is to build a reliable evidence pack before supporting any paper outline, abstract, related work summary, experiment plan, or draft section.
Version: v0.3
Use the highest-quality source available for each claim.
Read references/source-priority.md when source quality or conflicts matter.
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 scripts/build_evidence_ledger.py when the user already has markdown notes or source lists and needs a first-pass evidence ledger.
Use 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 scripts/triage_evidence_gaps.py when the user needs a first-pass blocker/major/minor gap report from the current claims and evidence ledger.
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 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.
ZIP package — ready to use