ILS Risk Conformance Suite

A language-neutral specification and test corpus for insurance-linked securities risk calculations, with independent implementations in Python, Scala, Julia and Excel.

The specification is the product. The implementations are evidence that it is implementable in each language and should provide some sense of the relative readability of each..

Why

Two desks can compute the same expected loss from the same loss table and get different numbers. Usually the cause is not a modelling disagreement but an unstated convention — summation order, quantile interpolation, tie-breaking, whether a running total is carried down a column. These are invisible in code review and rarely written down.

This repository writes them down, and then tests four independent implementations against them.

Each implementation uses its own language's standard arithmetic, so the four do not produce byte-identical answers. Each case declares a relative tolerance wide enough to cover that and narrow enough to catch a real error, and every implementation is measured against an independently derived golden value rather than against the others.

See Results for the current scorecard.

Scope

In scope: the deterministic contract mechanics — layer attachment and exhaustion, franchise and aggregate deductibles, reinstatements and reinstatement premium, indexation, hours clauses, and EP curve construction from a fixed ELT or YLT. All pure functions of tabular input.

Out of scope: catastrophe model construction, hazard simulation, vendor model calibration.

At v1.0 only expected loss is specified. The scope above is the roadmap, not the current contents.

Structure

spec/spec.md          normative specification -- section numbers are a public API
conformance/cases/    the corpus: one JSON file per case
conformance/data/     input tables, digest-pinned
conformance/manifest.toml   tiers, capabilities, xfails
impl/python/          reference implementation (python / uv)
impl/scala/           independent implementation (scala / Mill)
impl/julia/           independent implementation (Julia / Pkg)
impl/excel/           template workbooks + harness
tools/                corpus checks, golden-value derivation, scorecard