@buildtovalue/sfeel

Minimal S-FEEL subset parser + evaluator for decision tables (Handoff 9 §5). Serves the token simulator (Handoff 7 — businessRuleTask routes through real decision tables) and the DMN editor (Handoff 5 — ⚠ "não-simulável" cell markers) without importing either: the package operates on plain string cells plus a {variable: value} context and has zero dependencies, zero imports from the ecosystem.

The subset (supported)

| Construct | Example | |---|---| | Comparisons | < 5, <= 5, > 5, >= 5, = "gold" | | Ranges (incl./excl./mixed) | [1..10], ]1..10[, [1..10[, ]1..10] | | Value lists (implicit OR) | "a", "b" · 1, 2, 3 · < 3, > 10 | | List negation | not("a", "b") | | Irrelevant | - (or an empty cell) | | Types | number, string, boolean | | Hit policies | U (Unique), F (First) |

Outputs are literals only (number, "string", boolean).

Explicitly excluded (cerca §1.6)

Function invocation (date(), duration(), …), arithmetic in cells, for/some/every, nested contexts, date/time/duration types and literals, non-literal output expressions, identifier references in cells, and evaluation of the A/P/R/O/C hit policies (they remain editor metadata — registered in pendencias.md).

Honesty contract

evaluate(table, context) returns exactly one of:

• { result } — the matched rule's outputs (or null when no rule matches);
• { nonSimulable: { cell, reason } } — a declared failure naming the offending cell and why: a construct outside the subset, a malformed cell, a missing input variable, a type mismatch, or a Unique-hit-policy violation (two rules matching).

There is no third, silent state — a cell the package cannot evaluate correctly is never evaluated at all. checkTable / checkUnaryCell / checkOutputCell expose the same verdicts statically for editor feedback before simulation.

Input-column expressions are used verbatim as context keys — they are not evaluated (an input expression that needs evaluation is outside the subset).