conjure-eval 0.1.0__tar.gz

conjure_eval-0.1.0/PKG-INFO

@@ -0,0 +1,83 @@
+Metadata-Version: 2.4
+Name: conjure-eval
+Version: 0.1.0
+Summary: Public-slice harness for the CONJURE transformative-creativity benchmark.
+Author-email: Patrick Cooper <patrick.cooper@colorado.edu>
+License: Apache-2.0
+Keywords: benchmark,lean4,mathlib,llm,creativity
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Science/Research
+Classifier: License :: OSI Approved :: Apache Software License
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+Provides-Extra: verify
+
+# conjure-eval
+
+Public-slice harness for the CONJURE transformative-creativity benchmark.
+Ships the 358-instance public split (70 percent of the 510-instance Phase 4.6
+frozen corpus across 17 Lakatos families, SHA-256
+`33e9daebbfc1382b08c4b518f6bc9b30e62c13cc9d7e178327675929ebd74cc9`) so
+frontier-model developers can self-evaluate locally before submitting to the
+hidden split.
+
+This package contains:
+
+- The frozen public-slice corpus JSON (`conjure_eval.data.public_corpus`).
+- A CLI for inspecting the corpus, driving a model pass, and checking
+  submission files before they are sent to the hidden-split adjudicator.
+- The deterministic split provenance, so any third party can re-derive the
+  public/hidden split byte-for-byte from the source corpus.
+
+## What this package is and isn't
+
+`conjure-eval` is a self-service developer convenience: it lets a model team
+inspect the public contracts, run their model against the public slice, and
+smoke-test their submission format before sending results to the benchmark
+author. It does not ship the hidden split, and it does not run the
+kernel-verified tight-mode adjudicator that produces the headline accept rate.
+Those live in the private `blanc` repository and are operated by the benchmark
+author against frozen model snapshots; the headline number reported in the
+brief is the hidden-split rate.
+
+## Install
+
+```bash
+pip install conjure-eval
+```
+
+## Usage
+
+```bash
+# List all 358 public-slice instance IDs
+conjure-eval list-public
+
+# Inspect a single instance
+conjure-eval show C1-bv-001
+
+# Drive a model pass (OpenAI-compatible endpoint)
+conjure-eval run \
+    --base-url https://your-endpoint/v1 \
+    --api-key-env MY_API_KEY \
+    --model your-model-name \
+    --out submissions.jsonl
+
+# Check submission file well-formedness before sending
+conjure-eval verify-submission submissions.jsonl
+
+# Print corpus provenance fields
+conjure-eval provenance
+```
+
+## Provenance
+
+The public corpus is a deterministic 70/30 axis-stratified slice of the
+510-instance Phase 4.6 frozen corpus maintained in the private `blanc`
+repository. Seed: `4317`. Anyone with the source corpus can reproduce both
+slices via `scripts/build_conjure_split.py`.

conjure_eval-0.1.0/README.md

@@ -0,0 +1,63 @@
+# conjure-eval
+
+Public-slice harness for the CONJURE transformative-creativity benchmark.
+Ships the 358-instance public split (70 percent of the 510-instance Phase 4.6
+frozen corpus across 17 Lakatos families, SHA-256
+`33e9daebbfc1382b08c4b518f6bc9b30e62c13cc9d7e178327675929ebd74cc9`) so
+frontier-model developers can self-evaluate locally before submitting to the
+hidden split.
+
+This package contains:
+
+- The frozen public-slice corpus JSON (`conjure_eval.data.public_corpus`).
+- A CLI for inspecting the corpus, driving a model pass, and checking
+  submission files before they are sent to the hidden-split adjudicator.
+- The deterministic split provenance, so any third party can re-derive the
+  public/hidden split byte-for-byte from the source corpus.
+
+## What this package is and isn't
+
+`conjure-eval` is a self-service developer convenience: it lets a model team
+inspect the public contracts, run their model against the public slice, and
+smoke-test their submission format before sending results to the benchmark
+author. It does not ship the hidden split, and it does not run the
+kernel-verified tight-mode adjudicator that produces the headline accept rate.
+Those live in the private `blanc` repository and are operated by the benchmark
+author against frozen model snapshots; the headline number reported in the
+brief is the hidden-split rate.
+
+## Install
+
+```bash
+pip install conjure-eval
+```
+
+## Usage
+
+```bash
+# List all 358 public-slice instance IDs
+conjure-eval list-public
+
+# Inspect a single instance
+conjure-eval show C1-bv-001
+
+# Drive a model pass (OpenAI-compatible endpoint)
+conjure-eval run \
+    --base-url https://your-endpoint/v1 \
+    --api-key-env MY_API_KEY \
+    --model your-model-name \
+    --out submissions.jsonl
+
+# Check submission file well-formedness before sending
+conjure-eval verify-submission submissions.jsonl
+
+# Print corpus provenance fields
+conjure-eval provenance
+```
+
+## Provenance
+
+The public corpus is a deterministic 70/30 axis-stratified slice of the
+510-instance Phase 4.6 frozen corpus maintained in the private `blanc`
+repository. Seed: `4317`. Anyone with the source corpus can reproduce both
+slices via `scripts/build_conjure_split.py`.

conjure_eval-0.1.0/pyproject.toml

@@ -0,0 +1,43 @@
+[build-system]
+requires = ["setuptools>=68.0", "wheel"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "conjure-eval"
+version = "0.1.0"
+description = "Public-slice harness for the CONJURE transformative-creativity benchmark."
+readme = "README.md"
+requires-python = ">=3.10"
+license = { text = "Apache-2.0" }
+authors = [
+    { name = "Patrick Cooper", email = "patrick.cooper@colorado.edu" },
+]
+keywords = ["benchmark", "lean4", "mathlib", "llm", "creativity"]
+classifiers = [
+    "Development Status :: 4 - Beta",
+    "Intended Audience :: Science/Research",
+    "License :: OSI Approved :: Apache Software License",
+    "Programming Language :: Python :: 3",
+    "Programming Language :: Python :: 3.10",
+    "Programming Language :: Python :: 3.11",
+    "Programming Language :: Python :: 3.12",
+    "Programming Language :: Python :: 3.13",
+    "Topic :: Scientific/Engineering :: Artificial Intelligence",
+]
+dependencies = []
+
+[project.optional-dependencies]
+verify = []
+
+[project.scripts]
+conjure-eval = "conjure_eval.cli:main"
+
+[tool.setuptools]
+package-dir = { "" = "src" }
+
+[tool.setuptools.packages.find]
+where = ["src"]
+include = ["conjure_eval*"]
+
+[tool.setuptools.package-data]
+conjure_eval = ["data/*.json"]

conjure_eval-0.1.0/setup.cfg

@@ -0,0 +1,4 @@
+[egg_info]
+tag_build = 
+tag_date = 0
+

conjure_eval-0.1.0/src/conjure_eval/__init__.py

@@ -0,0 +1,27 @@
+"""Public-slice harness for the CONJURE transformative-creativity benchmark.
+
+This package ships the 358-instance public split (70 percent of the
+510-instance Phase 4.6 frozen corpus across 17 Lakatos families).
+See README.md for the full description and the hidden-split policy.
+"""
+
+from __future__ import annotations
+
+from .corpus import (
+    PUBLIC_CORPUS_PATH,
+    Instance,
+    load_public_corpus,
+    public_instance_ids,
+    public_instance_by_id,
+)
+
+__all__ = [
+    "PUBLIC_CORPUS_PATH",
+    "Instance",
+    "load_public_corpus",
+    "public_instance_ids",
+    "public_instance_by_id",
+    "__version__",
+]
+
+__version__ = "0.1.0"

conjure_eval-0.1.0/src/conjure_eval/cli.py

@@ -0,0 +1,233 @@
+"""`conjure-eval` command-line entry point.
+
+Three subcommands ship in v0.1.0:
+
+  list-public                        print every public-slice instance ID
+  show <instance_id>                 print one instance prompt + axis
+  verify-submission <jsonl>          well-formedness check on a JSONL of
+                                     submissions; reports per-record reasons.
+
+These are *all* developer conveniences; the actual kernel-verified
+adjudication runs against the hidden split inside the private repo.
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+import os
+import sys
+from pathlib import Path
+
+from .corpus import (
+    load_public_corpus,
+    public_instance_by_id,
+    public_instance_ids,
+)
+from .runner import run_against_endpoint
+
+
+def cmd_list_public(_args: argparse.Namespace) -> int:
+    for iid in public_instance_ids():
+        print(iid)
+    return 0
+
+
+def cmd_show(args: argparse.Namespace) -> int:
+    try:
+        inst = public_instance_by_id(args.instance_id)
+    except KeyError as exc:
+        print(str(exc), file=sys.stderr)
+        return 2
+    print(f"instance_id: {inst.instance_id}")
+    print(f"axis:        {inst.axis}")
+    print("-" * 60)
+    print(inst.prompt or "<no prompt body in public-slice JSON>")
+    return 0
+
+
+def _verify_record(rec: dict, public_ids: set[str]) -> list[str]:
+    """Return a list of human-readable errors for a single submission record.
+
+    Empty list means the record is well-formed.
+    """
+    errors: list[str] = []
+    iid = rec.get("instance_id")
+    if iid is None:
+        errors.append("missing field: instance_id")
+    elif iid not in public_ids:
+        errors.append(f"instance_id {iid!r} is not in the public slice")
+    if not isinstance(rec.get("submission"), str):
+        errors.append("missing or non-string field: submission")
+    return errors
+
+
+def cmd_verify_submission(args: argparse.Namespace) -> int:
+    path = Path(args.path)
+    if not path.exists():
+        print(f"no such file: {path}", file=sys.stderr)
+        return 2
+
+    public_ids = set(public_instance_ids())
+    total = 0
+    failed = 0
+    seen_ids: set[str] = set()
+
+    for lineno, line in enumerate(path.read_text(encoding="utf-8").splitlines(), 1):
+        line = line.strip()
+        if not line:
+            continue
+        total += 1
+        try:
+            rec = json.loads(line)
+        except json.JSONDecodeError as exc:
+            failed += 1
+            print(f"line {lineno}: invalid JSON: {exc}")
+            continue
+        errs = _verify_record(rec, public_ids)
+        iid = rec.get("instance_id", "<unknown>")
+        if errs:
+            failed += 1
+            for e in errs:
+                print(f"line {lineno} ({iid}): {e}")
+        else:
+            seen_ids.add(iid)
+
+    missing = sorted(public_ids - seen_ids)
+    print()
+    print(f"records:    {total}")
+    print(f"well-formed: {total - failed}")
+    print(f"failed:     {failed}")
+    print(f"covered ids: {len(seen_ids)} / {len(public_ids)} public-slice instances")
+    if missing:
+        print("missing-id sample:", ", ".join(missing[:5]) + ("..." if len(missing) > 5 else ""))
+
+    return 1 if failed else 0
+
+
+def cmd_run(args: argparse.Namespace) -> int:
+    api_key = os.environ.get(args.api_key_env, "")
+    if not api_key:
+        print(
+            f"error: environment variable {args.api_key_env!r} is not set or empty",
+            file=sys.stderr,
+        )
+        return 2
+
+    corpus = load_public_corpus()
+    instances = corpus["instances"]
+    total = min(args.limit, len(instances)) if args.limit else len(instances)
+
+    print(
+        f"conjure-eval run: {total} instances, model={args.model}, "
+        f"out={args.out}",
+        file=sys.stderr,
+    )
+
+    def _progress(i: int, n: int, rec) -> None:
+        status = "ok" if rec.error is None else f"ERR: {rec.error[:60]}"
+        print(f"  [{i}/{n}] {rec.instance_id} {status}", file=sys.stderr)
+
+    run_against_endpoint(
+        base_url=args.base_url,
+        api_key=api_key,
+        model=args.model,
+        out_path=args.out,
+        instances=instances,
+        limit=args.limit,
+        rate_limit_ms=args.rate_limit_ms,
+        max_retries=args.max_retries,
+        timeout_s=args.timeout_s,
+        progress_fn=_progress,
+    )
+    print(f"done: results written to {args.out}", file=sys.stderr)
+    return 0
+
+
+def build_parser() -> argparse.ArgumentParser:
+    parser = argparse.ArgumentParser(
+        prog="conjure-eval",
+        description=(
+            "Public-slice harness for the CONJURE benchmark "
+            "(Phase 4.6 frozen corpus, 358 public instances across 17 Lakatos families)."
+        ),
+    )
+    sub = parser.add_subparsers(dest="cmd", required=True)
+
+    sub_list = sub.add_parser("list-public", help="print every public-slice instance ID")
+    sub_list.set_defaults(func=cmd_list_public)
+
+    sub_show = sub.add_parser("show", help="print one public-slice instance prompt + axis")
+    sub_show.add_argument("instance_id", help="instance ID (see `list-public`)")
+    sub_show.set_defaults(func=cmd_show)
+
+    sub_verify = sub.add_parser(
+        "verify-submission",
+        help="well-formedness check on a JSONL of {instance_id, submission} records",
+    )
+    sub_verify.add_argument("path", help="path to submissions.jsonl")
+    sub_verify.set_defaults(func=cmd_verify_submission)
+
+    sub_prov = sub.add_parser("provenance", help="print public-slice provenance fields")
+    sub_prov.set_defaults(func=cmd_provenance)
+
+    sub_run = sub.add_parser(
+        "run",
+        help="run a model against the public slice and write submissions.jsonl",
+    )
+    sub_run.add_argument(
+        "--base-url", required=True,
+        help="OpenAI-compatible base URL, e.g. https://api.openai.com/v1",
+    )
+    sub_run.add_argument(
+        "--api-key-env", required=True,
+        help="name of the environment variable holding the bearer token",
+    )
+    sub_run.add_argument("--model", required=True, help="model name string")
+    sub_run.add_argument(
+        "--out", required=True,
+        help="output JSONL path (appended, not overwritten)",
+    )
+    sub_run.add_argument(
+        "--limit", type=int, default=None,
+        help="stop after N instances (for smoke tests)",
+    )
+    sub_run.add_argument(
+        "--rate-limit-ms", type=int, default=0,
+        help="milliseconds to sleep between requests (default 0)",
+    )
+    sub_run.add_argument(
+        "--max-retries", type=int, default=3,
+        help="retries on 429/5xx with exponential back-off (default 3)",
+    )
+    sub_run.add_argument(
+        "--timeout-s", type=float, default=120.0,
+        help="per-request HTTP timeout in seconds (default 120)",
+    )
+    sub_run.set_defaults(func=cmd_run)
+
+    return parser
+
+
+def cmd_provenance(_args: argparse.Namespace) -> int:
+    corpus = load_public_corpus()
+    prov = corpus.get("public_split_provenance", {})
+    print(f"public_corpus.name:    {corpus.get('name', '<unnamed>')}")
+    print(f"split_schema_version:  {prov.get('split_schema_version', '<unknown>')}")
+    print(f"split_seed:            {prov.get('split_seed', '<unknown>')}")
+    print(f"split_public_ratio:    {prov.get('split_public_ratio', '<unknown>')}")
+    print(f"source_corpus_sha256:  {prov.get('corpus_sha256', '<unknown>')}")
+    print("public_axis_counts:")
+    for axis, n in sorted((prov.get("public_axis_counts") or {}).items()):
+        print(f"  {axis}: {n}")
+    return 0
+
+
+def main(argv: list[str] | None = None) -> int:
+    parser = build_parser()
+    args = parser.parse_args(argv)
+    return args.func(args)
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

conjure_eval-0.1.0/src/conjure_eval/corpus.py

@@ -0,0 +1,79 @@
+"""Public-corpus loader for `conjure-eval`.
+
+The frozen public-slice corpus JSON ships as package data under
+`conjure_eval/data/public_corpus.json`. This module loads it lazily and
+exposes a thin typed view over the instance records.
+"""
+
+from __future__ import annotations
+
+import json
+from dataclasses import dataclass
+from importlib import resources
+from typing import Any
+
+
+PUBLIC_CORPUS_RESOURCE = ("conjure_eval.data", "public_corpus.json")
+
+# Filesystem path is exposed for tests and the CLI; package-data lookup is
+# the source of truth at runtime.
+PUBLIC_CORPUS_PATH = resources.files("conjure_eval.data").joinpath(
+    "public_corpus.json"
+)
+
+
+@dataclass(frozen=True)
+class Instance:
+    """A single public-slice CONJURE instance.
+
+    The `raw` field preserves the full source JSON record so downstream
+    tooling can read schema fields this package doesn't model yet.
+    """
+
+    instance_id: str
+    axis: str
+    prompt: str
+    raw: dict[str, Any]
+
+
+def load_public_corpus() -> dict[str, Any]:
+    """Return the parsed public-corpus JSON.
+
+    Raises
+    ------
+    FileNotFoundError
+        If the package was installed without the data file present (i.e.,
+        somebody mis-built the wheel).
+    """
+    pkg, name = PUBLIC_CORPUS_RESOURCE
+    text = resources.files(pkg).joinpath(name).read_text(encoding="utf-8")
+    return json.loads(text)
+
+
+def public_instance_ids() -> list[str]:
+    """Return the sorted list of public instance IDs."""
+    corpus = load_public_corpus()
+    return sorted(inst["instance_id"] for inst in corpus["instances"])
+
+
+def public_instance_by_id(instance_id: str) -> Instance:
+    """Look up a single public-slice instance by ID.
+
+    Raises
+    ------
+    KeyError
+        If the ID is not part of the public slice.
+    """
+    corpus = load_public_corpus()
+    for inst in corpus["instances"]:
+        if inst["instance_id"] == instance_id:
+            return Instance(
+                instance_id=inst["instance_id"],
+                axis=inst["axis"],
+                prompt=inst.get("prompt", ""),
+                raw=inst,
+            )
+    raise KeyError(
+        f"{instance_id!r} is not in the public slice (or not present in "
+        f"this conjure-eval release)"
+    )

conjure_eval-0.1.0/src/conjure_eval/data/__init__.py

@@ -0,0 +1 @@
+"""Frozen public-corpus data for `conjure-eval`."""