Evidence-aware
“No,” “unknown,” and missing documentation remain hard gaps. Partial evidence stays visible instead of being converted into a confident verdict.
Research software · local-first · v0.2.1
ReproWeave connects a paper's claims to exact evidence, experiments, required resources, and dependency-aware work. It turns that record into an explicit replication queue—without an API, account, database, or hidden score.
Methods
Every transition is stored as plain JSON. Another researcher can review the evidence locator, resource status, task dependency, and decision rule rather than trusting a generated summary.
Record the question, criteria, paper identity, and screening reason.
Join bounded claims to figures, tables, appendices, experiments, and resources.
Evaluate eight reconstructability dimensions while preserving unknown values.
Combine hard blockers, task readiness, and effort; test resource-access scenarios.
Audit references and cycles, then seal the exact source snapshot with SHA-256.
What v0.2 adds
“No,” “unknown,” and missing documentation remain hard gaps. Partial evidence stays visible instead of being converted into a confident verdict.
Experiments contribute their code, data, model, environment, and hardware requirements. A what-if override changes only the generated scenario.
Task states, prerequisites, and remaining human-entered hours determine whether work can start. The output explains the next action in plain language.
reproweave triage --workspace review \
--resource rf-spectra-v2=available \
--format markdown --output triage.md
Portable evidence report
The demonstration is entirely synthetic. No title, author, URL, score, or result represents a real publication.
Interpretation boundary
Python 3.11+
The tagged release workflow publishes attested distributions to PyPI and the same files to GitHub with SHA-256 checksums.
python -m pip install reproweave
reproweave demo my-review
reproweave audit --workspace my-review
reproweave report --workspace my-review