Everything the engine can do

Eligibility & rating logic

What a single rule can express.

Eligibility conditions & Decision Graphs

• Chained Decision Graphs (DAG). model complex multi-step pipelines visually with 6 node types (Input, Eligibility, Table, Rating, Transform, Output), context mutation, and conditional on_pass / on_fail branching.
• Nested AND / OR / NOT. arbitrarily deep condition trees, not a flat list of checks.
• 10 Native comparison operators. numeric/equality (==, !=, >=, >, <=, <, in), rich strings (starts_with, ends_with, contains, not_contains), linear-time regex (matches), and network CIDR range checks (cidr_match).
• Null & empty safety. built-in is_null, is_not_null, is_empty, and is_not_empty checks that handle missing or blank fields gracefully.
• Nested array quantifiers. array_any, array_all, array_none, and array_count with sub-conditions and $item sentinels to evaluate multi-item carts, covered drivers, or claims without code.
• Raw expressions. sandboxed Rhai scripts for arbitrary formulas and arithmetic when needed.
• Composable rules. a condition can reference another one of your eligibility rules by id ($ref) instead of duplicating its logic — expands inline in the trace.

Rating factor kinds & Decision Tables

• DMN 1.5 decision tables & hit policies. exact-match and numeric-range lookup tables with First, Unique, CollectSum, CollectCount, and CollectMin/Max hit policies.
• Spreadsheet CSV import / export. bidirectional CSV round-tripping with row-by-row pre-validation to bulk-load pricing matrices from Excel or Google Sheets.
• Keyboard-first entry grid. arrow keys, Tab, and Enter-adds-row navigation for editing table rows directly, plus pasting a block of cells copied straight from Excel or Google Sheets in one go.
• Flat. a fixed multiplier or addend.
• Condition-based. applies only when a condition passes (e.g. a bundling discount).
• Tiered / bracket. a progressive charge across brackets of a basis value (e.g. layered limits).
• Deductible curve. a factor read off a curve of deductible-to-factor points, step or interpolated.
• Raw expression. a factor's value computed from a formula, for anything the structured kinds can't express.
• Rule composition (rating_ref). compose another rating rule's own computed amount into this one.
• Accumulators. a persistent, cross-call running total — e.g. how much of an annual deductible has been paid so far.
• Reinsurance layers. excess-of-loss attachment point + limit, composable into a full layered tower.

Calculation correctness

• Effective-dating, region, and product-type scoping. one rule id can have multiple versions; the most specific match is selected automatically.
• Multi-line-item and multi-entity rating. rate a bundle of items, or the same rule across several members, in one call.
• Mid-term pro-ration. built in, not something you compute separately.
• Configurable rounding. half-up, half-even, or truncate, with per-region overrides.
• Exact decimal arithmetic. never floating-point rounding errors on money math.

Safety, AI explainability & governance

• Natural-language rule generation. describe what you want in plain English and get a draft condition tree back to review — nothing is ever saved or activated automatically.
• 4-Eyes Maker-Checker governance. strict multi-user approval queues (Draft -> Pending Review -> Active) with independent reviewer sign-off to satisfy banking and insurance compliance.
• Visual semantic diffing & 1-click rollback. AST-level diff viewer highlighting added, removed, and modified clauses between any two versions, with instant rollback.
• Deterministic plain-English explainer. zero-hallucination recursive AST translator rendering complex logic into plain English for audit and compliance teams.
• Auto-generated test scenarios. sample inputs derived from a rule's own structure to exercise its different branches, as a starting point before you write your own.
• Shadow & conflict detector. real interval-overlap math flagging redundant or conflicting rules across your tenant.
• Full evaluation trace on every call. not just a pass/fail or a number — a complete, nested explanation of exactly how each condition and factor contributed, suitable as an audit trail for a regulated industry.
• Loud failure. a calculation error surfaces immediately, never a silent default.
• Load-time and activation-time validation. overlapping effective-date/region/product versions of a rule are rejected outright.
• Semantic lint checks. dangling table references, unused tables, no-op factors, out-of-order brackets, and likely-negative amounts are flagged before deployment.