fix(benchmarks): make rlpr/ifeval/bfcl_simple load() callable instead of raising TypeError (#152)

[joaovictor91123]

Fixes #152

Summary

The three OOD benchmark facades added in #104/#103/#102 — benchmarks/rlpr.py, benchmarks/ifeval.py, benchmarks/bfcl_simple.py — each raised TypeError on every call to their public load(). This is the identical defect that benchmarks/livecodebench.py had and that was fixed + regression-tested in #90; the three newer facades were copied from the pre-fix template and regressed the same way.

def load(split, *, max_items=None, seed=0, allow_toy_fallback=False):
    return load_tasks(          # <-- the SIBLING load_tasks, not _load_tasks
        "rlpr", split, max_items=max_items, seed=seed,
    )

def load_tasks(split, *, max_items=None, seed=0, allow_toy_fallback=False):
    return _load_tasks("rlpr", split, max_items=max_items, seed=seed)

load() passes two positionals, but the sibling load_tasks(split, *, ...) takes one. Offline, before any dataset loading:

>>> import benchmarks.rlpr, benchmarks.ifeval, benchmarks.bfcl_simple
>>> benchmarks.rlpr.load("test", max_items=1)
TypeError: load_tasks() takes 1 positional argument but 2 positional arguments (and 2 keyword-only arguments) were given

Same for ifeval and bfcl_simple. livecodebench.load(...) (fixed in #90) does not raise.

Why This Matters

load() is the documented config-facing entrypoint. benchmarks/__init__.py states "Each loader exposes load(split, **kw) -> list[Task]", and configs/benchmarks.yaml names all three modules as loader: for the OOD suite (ifeval, rlpr, bfcl_simple). Any caller going through the documented facade — rather than reaching past it to dataset.load_tasks directly — cannot load these three benchmarks at all. This is the same public-API break #90 fixed for LiveCodeBench, now closed for its three siblings.

Why CI stayed green

Each facade test masks the bug exactly as the pre-#90 livecodebench test did — it monkeypatches the sibling load_tasks (the very function whose signature is the bug) with a four-positional stub, so the two-positional miscall type-checks against the fake:

def fake_load_tasks(benchmark, split, max_items, seed):   # four positionals
    ...
monkeypatch.setattr(RLPR, "load_tasks", fake_load_tasks)

A stub whose signature does not match the function it replaces cannot catch a signature bug. #90 documented this and fixed livecodebench's test, but the fix was never propagated to the three siblings that share the template.

What Changed

• benchmarks/rlpr.py, benchmarks/ifeval.py, benchmarks/bfcl_simple.py: load() now delegates to load_tasks(split, max_items=..., seed=...), naming the benchmark in one place (matching the merged fix(benchmarks): make livecodebench.load() callable instead of raising TypeError (#86) #90 fix).
• tests/test_rlpr.py, tests/test_ifeval.py, tests/test_bfcl.py: each *_facade_delegates test now patches the real _load_tasks boundary, and each gains a *_load_is_callable test that calls load() without patching load_tasks — the regression test fix(benchmarks): make livecodebench.load() callable instead of raising TypeError (#86) #90 established as the standard.
• docs/JOURNAL.md: replication-log entry per repository protocol (AGENTS.md §6).

Notes

• Verified locally: all 6 touched/new facade tests fail on pre-fix main with the reported TypeError (including the corrected *_facade_delegates, which passed before only because of the mismatched stub) and pass after. pytest tests -q → 236 passed.
• The HF-row parsing and scoring tests in the same three files are untouched and still pass, so the loading/grading paths are unchanged — this only repairs the facade call.
• Same one-line delegation change in all three files; no behavior change beyond making load() callable.

[joaovictor91123]

Temporarily closing to keep two PRs in flight at a time; the branch sn74-joaovictor91123-benchmark-facades-typeerror is preserved and this will be re-opened once #175 or #176 lands. The fix (and issue #152) stand — benchmarks.{rlpr,ifeval,bfcl_simple}.load() still raise TypeError on main.