evgraph-core

Foundational types for the Evgraph evidence-graph stack.

What is it

evgraph-core defines the small, stable vocabulary that every other Evgraph package is built on: what an Evidence Graph is, how certain a piece of evidence is allowed to claim to be, and the contracts an adapter or a rule must honor to participate in that graph. It has no opinion about how evidence gets collected or what should be checked — those are the concerns of evgraph-rules (built-in checks) and evgraph (adapters, reporters, the public scan() API). evgraph-core only defines the shared data model those higher layers agree on, which is what lets adapters, rule packs, and reporters be written independently and still interoperate.

The types here implement the Evidence Graph Specification (EGS) and the Rule Evaluation Specification (RES) — normative documents in the main repository (docs/specs/) that this package is the reference implementation of. If a detail here and a spec ever disagree, the spec wins and this package has a bug.

Main features

• EvidenceGraph / EvidenceNode / EvidenceEdge — an immutable, frozen-dataclass model of a directed graph of typed facts. Node and edge identity is enforced at construction: duplicate node ids and edges that reference a missing node both raise ValueError immediately, so a graph that exists is always internally consistent.
• EvidenceLevel — an ordered IntEnum (STRUCTURAL < CONSISTENCY < HEURISTIC < INTERPRETIVE) expressing how certain a fact or a finding is. Reasoning can only ever get less certain as it propagates through a rule (least_certain() enforces this), never more — so a finding can never claim stronger evidence than the facts it cites.
• Rule (Protocol) / Finding / Outcome — the contract a rule implementation satisfies (purity, no shared-state side effects, every citation must point at a real node in the input graph) and the immutable, explainable result a rule produces: an outcome (EXPECTATION_MET / EXPECTATION_NOT_MET / INCONCLUSIVE), a level, and the node ids that justify it. compute_finding_level() is the only sanctioned way to derive Finding.level — rule authors never assign it directly, which is what keeps findings honest about their own certainty.
• Adapter (Protocol) / AdapterError / Assumption — the minimal surface an extractor declares (name, version, evidence level, extraction method, its assumptions) without standardizing how extraction itself is invoked, so adapters stay free to have artifact-appropriate signatures. AdapterError is reserved for failures of interpretation — malformed or unreadable input — never for governance deficiencies in otherwise-valid data.
• Graph (de)serialization — round-trippable JSON encoding for EvidenceGraph, so a graph built once can be persisted, diffed, or handed to a different process without re-running adapters.

This package has no runtime dependency on any other Evgraph package — evgraph-rules, evgraph, and evgraph-cli all depend on it, never the other way around, which keeps the core vocabulary stable independent of how many rule packs or adapters get built on top of it.

Where to get it

The source is hosted on GitHub at: https://github.com/SVamseekar/evgraph

Binary installers for the latest released version are available at the Python Package Index (PyPI):

pip install evgraph-core

From source, for contributing:

git clone https://github.com/SVamseekar/evgraph.git
cd evgraph
pip install -e reference/python/evgraph-core

Dependencies

None at runtime. pytest>=7 is required for the test suite (pip install "evgraph-core[test]").

Documentation

Getting help

Ask questions and report bugs via GitHub Issues on the main repository — this package doesn't track issues separately from the monorepo.

Contributing

All contributions, bug reports, and feature requests are welcome on the main repository. See CONTRIBUTING.md and the Code of Conduct. Security reports: SECURITY.md.

License

BSD 3-Clause.