Model Replication Diff

Independently recomputes a model's reported outputs from a declared model specification (version, as-of date, transform, intercept, coefficients) and a stated input record set, then diffs the recomputation against the model's own reported outputs within a caller-supplied tolerance. Returns a per-record diff (reported, recomputed, absolute and relative diff, within-tolerance), aggregate tolerance stats, an overall replicated / not-replicated verdict, and the failing segment(s) named when not replicated. If the declared specification cannot be recomputed as stated -- unrecognized transform, no coefficients, no tolerance -- the honest result is not_replicable_as_specified with a reason, not a forced pass or fail. Deterministic recompute-and-diff only: this node never opines on conceptual soundness, assumption reasonableness, or fitness for use -- that judgment belongs in a validator's approval record, not here. Model inputs and specifications are expected to arrive as a workbook export via the shipped WB-BRIDGE-1 workbook-to-OCG artifact bridge (tool 554); this node itself accepts plain JSON records so any upstream source can feed it. Distinct from the shipped model-passport nodes (art-450/451/453), which track a model's inventory tier and revalidation cadence rather than recomputing its numeric output. Inline deterministic transcendental math (no engine Math.exp/log) so the same input always produces the same execution_hash on every surface. NaN-safe. Zero network, zero PII.

Tool URL: https://ainumbers.co/chaingraph/art-488-model-replication-diff.html