Storage-agnostic core for FAIRSCAPE RO-Crate graph tools — provenance-graph condensation, evidence graph construction, and AI-driven interpretation.
Both mds_python (FastAPI server, MongoDB-backed) and fairscape-cli (local
RO-Crate tool) consume this package and supply their own adapters for
entity lookup, result persistence, progress tracking, and software fetching.
Contents of this package (by module):
fairscape_graph_tools.models— Pydantic models forAnnotatedComputation,AnnotatedEvidenceGraph,EvidenceGraph(plus assumption / error / evidence types).fairscape_graph_tools.pipeline.graph_utils—@graphtraversal helpers (type detection, DAG ordering, ARK ID matching).fairscape_graph_tools.pipeline.condense— provenance-graph condensation (condense_graphfor RO-Crate-level,condense_evidence_graph_cachefor evidence-graph-level).fairscape_graph_tools.pipeline.stats— dataset statistics prompt formatting.fairscape_graph_tools.pipeline.github— GitHub source-code fetching.fairscape_graph_tools.prompts— system + synthesis prompts per audience.fairscape_graph_tools.runtime— async rate limiter, retry helper, Celery-safe event loop.fairscape_graph_tools.ports—GraphSource/ResultSink/TaskTracker/SoftwareFetcherProtocol definitions.fairscape_graph_tools.condenser—Condenserorchestrator.fairscape_graph_tools.interpreter—Interpreterorchestrator.fairscape_graph_tools.evidence_graph_builder—EvidenceGraphBuilderorchestrator.
This package was renamed from fairscape_interpret on 2026-04-21 to reflect
the broader scope beyond AI interpretation. See MIGRATION.md and
EVIDENCE_GRAPH_MIGRATION.md for migration history and pending work.