dsh-sr-ledger

A systematic-review ledger that survives review: append-only, dual-screened, PRISMA 2020, audit-verifiable.

Records the protocol, every screening decision with reviewer and stage, conflict adjudications and RoB/GRADE assessments — then derives the PRISMA flow and exports the evidence table. Nothing is silently resolved.

English · 简体中文

---

What it does

• sr_init — create/revise the protocol (PICO, include/exclude criteria, per-source identification counts, manual PRISMA boxes) and import records (deduplicated by DOI/PMID/arXiv/title). Every call is a new revision plus an audit event; the full revision is also appended to review-history.jsonl, so the criteria a decision was made under remain reconstructable.
• sr_screen — record include/exclude at title-abstract or full-text. Two reviewers disagreeing = conflict; only an adjudicator can settle it, and only when a conflict exists. Exclusions require a reason. RoB 2 / ROBINS-I / GRADE assessments attach to the same ledger.
• sr_prisma — derive the PRISMA 2020 counts, verify the hash-chained audit log, list integrity issues, and write prisma.json, prisma.md, evidence-table.md, ledger.json under <root>/exports.

What it does not do

• No searching or retrieval. Import records from paper-search / Zotero / PubMed exports.
• No full-text reading. Screening decisions are human judgements; the ledger records them.
• No meta-analysis. Effect sizes and pooling belong to dsh-meta-analysis.
• No automatic conflict resolution. A conflict stays open until an adjudicator acts.
• No risk-of-bias automation. It stores and summarises assessments; it does not infer them from a paper.

PRISMA boxes

Box	Source
Records identified	identification per source (manual; fallback = imported records, labelled in the report)
Duplicates removed	manual.duplicatesRemoved
Records screened	identified − duplicates
Excluded at title/abstract	derived from final decisions
Reports sought for retrieval	final TA includes
Reports not retrieved	manual.recordsNotRetrieved
Assessed for eligibility	sought − not retrieved
Excluded at full text (+ reasons)	derived; reasons required
Studies included	final full-text includes
Reports of included studies	manual.reportsOfIncludedStudies (default = included)

Integrity issues (never hidden)

unknown_record, exclude_without_reason, adjudication_without_conflict, duplicate_decision, full_text_without_ta_include, open_conflicts, identification_counts_missing, no_records, included_without_assessment, assessment_of_excluded — plus audit-chain failures from sr_prisma.integrity. Errors make the export honest; warnings are surfaced in the Markdown.

Tools

sr_init

{ title?, pico?, criteria?, identification?, manual?, records? } → { ok, review, imported: {added, skipped}, recordCount }.

sr_screen

{ recordId, reviewer, stage, decision, reason?, role?, assessment? } → { ok, state, openConflicts, assessment? }.

sr_prisma

{} → { ok, counts, issues, robSummary, integrity, exports }.

Error codes

INVALID_INPUT · NOT_INITIALIZED · RECORD_UNKNOWN · NO_CONFLICT · IO

Compatibility & install

DeepSeek Harness	dsh-v0.2.1-rc.5 (verified 2026-11-08); peer >=0.2.1-0 <0.3.0 || >=0.2.2-0 <0.3.0
Node	^22.19.0 || >=24.9.0 (tested 24.22.1), ESM only
Peers	@perrylink/dsh-research-core ^0.1.0, @deepseek-ai/cordis ^4.0.5-alpha.1, @deepseek-ai/dsh-tools, @deepseek-ai/schemastery ^3.18.5-alpha.1
Runtime dependencies	zero

dsh plugin --profile <name> add @perrylink/dsh-research-core dsh-sr-ledger
dsh --profile <name> --dump-config | grep dsh-sr-ledger

Configuration

Key	Default	Meaning
enabled	true	false mounts nothing
root	.sr-ledger	ledger root (review.json, review-history.jsonl, journals, exports/)

Security boundaries

• Local only: no network calls, no telemetry.
• Append-only journals; every protocol revision is preserved in review-history.jsonl before review.json is updated.
• Audit events are hash-chained with the REP core helpers; sr_prisma verifies the chain.
• No PII handling beyond reviewer names the user supplies (use identifiers, not personal data).

Evaluation

16 tests: protocol revision, full revision history and dedupe, invalid-record rejection, dual-screening conflict creation, adjudication-only resolution, exclusion-reason enforcement, adjudication-without-conflict rejection, unknown-record rejection, atomic rejection of an invalid assessment/role, RoB/GRADE recording, PRISMA count derivation, export bundle contents, audit-chain tamper detection, IO mapping, tool schemas and end-to-end execution, package contract.

No benchmark is claimed: the ledger is bookkeeping with explicit invariants, not a detector. The invariants are the tests.

Known limitations (v0.1)

• One review per root; multiple reviews need multiple profile configs or roots.
• identification counts are manual; the plugin does not search.
• No PRISMA-for-individual-studies variant, no forest plots, no GRADE evidence-profile tables.
• Reviewer identity is whatever the caller supplies; there is no authentication.

License

Apache-2.0. No third-party runtime code.