[davidanthoff]

Part of an effort to get every package holding MimiGIVE components onto the test-item framework.

This package is one of two in that set that shipped no tests at all — no test/ directory, no Test entry in Project.toml, and a jlpkgbutler CI that therefore had nothing to run. So this is less a migration than writing the suite that was missing, directly in test-item form.

Tests

Four test items:

Item	What it checks
API	get_model across all six AR6 SSP scenarios: time dimension is 1750:2300, temperature is finite and the right length, 2100 warming relative to the 1850–1900 mean is ordered by scenario forcing (ssp119 < ssp126 < ssp245 < ssp460 < ssp370 < ssp585), and every scenario lands in 1.0–6.0 °C
Truncated time horizon	Running to 2100 reproduces the first 351 years of the run to 2300 exactly — nothing about the model depends on how far ahead it is asked to go
Original	get_model_original across the four RCP scenarios, at its default horizon and at a common 1765–2300 one, with the same shape/finiteness checks and RCP26 < RCP45 < RCP60 < RCP85 ordering
Monte Carlo Simulation	run_mcs with three trials into a tempdir writes trials.csv and both result files, the saved temperatures are finite and cover trials 1–3, and a single-trial run is rejected

Every threshold and ordering was checked against an actual run of the model first rather than guessed. The two entry points are deliberately not compared against each other: get_model is AR6/SSP-driven and get_model_original is RCP-driven, so they are not expected to agree numerically.

CI

Replaces the six jlpkgbutler-* workflows and .jlpkgbutler.toml with .github/workflows/juliaci.yml calling julia-testitems/testitem-workflow@v2, plus a dependabot config for julia and github-actions.

Compat

julia lower bound goes from 1.6 to 1.10, matching MimiSSPs and MimiGIVE. Without this the workflow's include-smallest-compatible-minor-versions default spends a matrix leg on Julia 1.6. This is release relevant — a new release will be needed before MimiGIVE picks it up. Because this drops support for a released Julia version, the -DEV version in Project.toml is moved to 1.1.0-DEV so the next release is a minor bump rather than a patch.

Verification

All four test items pass locally via julia_run_testitems (4/4 passed, ~78 s).

---

🤖 This PR was opened by Claude (Claude Code).

A bug the new tests found

The first CI run was red on both 32-bit Windows legs, and it turned out to be a real defect rather than a bad test.

run_mcs was declared run_mcs(;trials::Int64 = 2237, ...). On a 32-bit platform an integer literal is an Int32, so run_mcs(trials = 3) never reached the function body at all:

TypeError: in keyword argument trials, expected Int64, got a value of type Int32

The Monte Carlo API was simply unusable on 32-bit Julia. Widening the annotation to Integer is not sufficient on its own, because Mimi's run method for a SimulationDef also takes an Int64 — so the conversion now happens at that call instead. Second commit in this PR.

The test item exercises Int32 and Int64 explicitly, so the 64-bit legs cover the 32-bit path too rather than relying on the x86 runners to catch a regression.