Supposition

Freeze generated cases before timing, then replay identical inputs across releases. This small fixture checks integer increment/reversal, rather than package performance.

From examples/kitchen-sink in the source checkout, run julia setup.jl extras. Then start Julia with julia --project=.controller/extras for the code below.

using PerfChecker, Supposition, Random
mktempdir() do directory
    path = joinpath(directory, "supposition.json")
    Random.seed!(42)
    freeze_supposition_corpus(path, Supposition.Data.Integers(0,10); count=4,
        encode=x -> Dict("value"=>x), metadata=Dict("task_rng_seed"=>42))
    cases = read_property_corpus(path)["cases"]
    replay = [x["value"] + 1 for x in cases]
    @assert all(replay[i] - 1 == cases[i]["value"] for i in eachindex(cases))
    println(replay)
end

The actual frozen corpus and generation/replay receipt retain 7, 7, 1, 6, replayed as 8, 8, 2, 7, with 4 checks passed. This does not claim discovery of a failing property.

Supposition uses the explicitly seeded task RNG and requires 64-bit Julia. Use PropCheck or a compatible saved corpus on 32-bit Julia. Preserve the file rather than assuming generator versions reproduce the same values.

Real heap workload

From examples/kitchen-sink, prepare julia setup.jl extras. The supplied corpus.jl freeze / corpus.jl replay applies this pattern to 20 heap inputs with the real heap oracle. No separate 20-case package receipt is claimed here. Generation remains outside the timed operation.