Outputs¶
Read the CLI result for current status, counts, next actions and available paths. An absent optional path is JSON null. Keep machine state for recovery; normally read only the report and nonempty action queues.
| Workflow / stage | State to retain | Human result / optional export |
|---|---|---|
| Single paper | Verified PDF, raw publisher BibTeX and provenance | The verified pair; no report tree |
| Corpus discovery | Request, coverage.jsonl, candidates, evidence packets, discovery errors and manifest |
pending-metadata.csv only when blocked metadata exists |
| Corpus review | Reviewed candidates, selected-papers.jsonl, selection and review manifests |
pending-review.csv stays hash-bound, including an empty completed queue |
| Corpus acquisition | Acquisition ledger, selection binding, delivery manifest and official pairs | paper-manifest.csv; manual/retry queues only when nonempty |
| Research discovery | research-plan.json, research-candidates.jsonl, citation-graph.json, search-saturation.json, research-run.json |
No evidence/gap reports at this stage |
| Research review / full text | Reviewed decisions, frozen selection, phase manifests, full-text-manifest.jsonl and internal parse cache |
Nonempty review/download queues; status separates acquisition from parsing |
| Research evidence | validated-evidence-reviews.jsonl and research-evidence.json |
One reports/report.md; --export-csv adds papers.csv and claim-coverage.csv |
| Author bibliography | author-discovery.json with source/identity/date evidence |
report.md; --export-csv adds author-discovery.csv |
| Library audit | audit.json |
report.md; nonempty manual queue; --export-csv adds audit.csv |
| Library update / import / organize | Reviewed plan, staged pairs, provenance, backups and apply journal | Review and apply the plan; keep backups for rollback |
New research runs record output_version: 2. Runs without it resume using their legacy report layout; frozen bytes and hashes are retained. In the new layout, the old research manifest and nearest-work matrix are combined into optional papers.csv; evidence map, claim status and the applicable bounded gap conclusion share report.md. Search provenance stays in research-plan.json. Re-running without CSV export clears only the named generated CSV views; resolved action queues disappear. Do not store hand-edited files under these generated names.
Optional Markdown policy¶
acquire research parses selected PDFs into a seven-day SHA-256 MinerU cache for internal evidence review. The cached Markdown is analysis state, not a delivery artifact. Ordinary fetch, manual fetch, and corpus workflows do not produce Markdown. Export only after the user explicitly requests a destination: