Declarative Decision-Tree Evaluator

Evaluates a caller-supplied, hash-pinned decision tree against caller-supplied facts entirely offline, client-side. The tree is inert data -- one audited interpreter walks it over a closed operator set (eq, in, lt, lte, gt, gte, between, all_of, any_of, none_of); no eval, no expression strings, no function-valued criteria. Every node, internal criteria included and not leaves only, must carry a citation {clause, source, source_digest, snapshot_location}; a tree with any uncited node is rejected by the loader before evaluation. The tree's own declared tree_digest is never trusted as an oracle for itself: it is recomputed from the tree's canonicalized bytes and evaluation refuses to proceed on a mismatch. Bounds (max_depth=12, max_nodes=256, max_tree_bytes=64KiB) are declared and enforced, cycle-acyclicity is checked at load, and evaluation walks an explicit iterative pointer, never recursion. Ships with one demonstrator tree, the Reg D Rule 501(a) entity-type accredited-investor category test (17 CFR 230.501(a)(1)/(2)/(3)/(8)/(9)), enumerated exhaustively over its 5 declared boolean fields -- 32 combinations, 0 unexplained results -- because every criterion in that tree is a bounded enum. This is never a full accredited-investor determination: the natural-person prongs ((a)(4)-(a)(7), (a)(10)-(a)(13)) and every numeric-threshold test are explicitly out of the demonstrator's declared scope. The kernel itself carries no classification rules of its own; a caller must supply its own tree to get a different classification.

Tool URL: https://ainumbers.co/chaingraph/art-628-declarative-decision-tree-evaluator.html