DecisionGraph Corev0.3

Deterministic Decision Kernel for Replayable Human Judgment

Infrastructure for replayable, immutable, auditable decision graphs — without inference.

Determinism

Same input → same output. Stable ordering and replay across time.

Immutability

After commit, the graph is append-only. Changes are expressed via supersession.

Replay

Reconstruct graph state as-of a boundary (commitId / timestamp) deterministically.

Deterministic validation in CI

DecisionGraph Core validates the structure of decisions — not just code. If a decision depends on a superseded assumption, CI fails deterministically.

$ decisiongraph traverse ./decisions

  ✔ [01-security-policy.decisionlog.json]
  ✔ [02-adr-auth.decisionlog.json]
  ✖ [03-adr-depends-on-old.decisionlog.json]

✖ ERROR: depends_on target 'N:auth-v1' is Superseded
  at graphs.G:adr.edges.E:adr-001.to

  N:adr-001  Use new auth system
        └─ depends_on → N:auth-v1  Old auth decision (superseded)

Result: FAILED

The code did not change. The assumption did.

Latest release

v0.4.1 adds DEPENDENCY_ON_DEPRECATED detection (WARN) and the --strict flag to treat warnings as errors. All four packages are now published on npm.