seine-rs 0.4.33

Differentially certified Drools-subset rule engine over Arrow dataframes

seine-rs

A rules/decisioning engine over Arrow dataframes — a Rust port of a bounded Drools (DRL) subset, differentially certified against real Drools 9.44.0.Final: every supported behavior is pinned by running the same scenarios through both engines, plus multi-seed fuzz campaigns (zero divergences to completion).

import polars as pl
import seine_rs
from seine_rs import Rule, fact

@fact
class Person:
    name: str
    age: int
    score: float

@fact
class Flagged:
    name: str
    score: float

adults = Rule("Adults")
p = adults.when(Person, Person.age >= 18)
adults.then_insert(Flagged, name=p.name, score=p.score)

people = pl.DataFrame({"name": ["ada", "kurt"], "age": [36, 17], "score": [91.5, 99.0]})
res = seine_rs.run([adults], {Person: people, Flagged: []})

pl.DataFrame(res.derived["Flagged"])   # facts the rules created
pl.DataFrame(res.firings)              # full audit trail

• Bulk, columnar boundary: Arrow tables in (polars / pyarrow / pandas ≥ 2.2, zero-copy via the PyCapsule interface), Arrow tables out. Row-object lists (@fact instances, dicts, dataclasses, Pydantic models) work too.
• Rules in Python or DRL: the Python builder compiles to DRL text, so both paths run the identical certified engine. Anything outside the certified grammar is a definition-time CompileError.
• Full working-memory lifecycle: insert → fire → update/delete by handle → fire, each certified differentially.
• No Python in the hot path: conditions, aggregates and salience are native; callbacks are observers over immutable results.

Install: pip install seine-rs — the import is import seine_rs.

Source, scenario corpus, and the full decision log (D-001…D-048): https://github.com/sl-agentics/seine