bench/bench.py

\
#!/usr/bin/env python3
from __future__ import annotations

import argparse
import json
from pathlib import Path
from typing import Any, Dict, Iterable, Tuple

REQUIRED_FIELDS = {"id","created","topic","task_type","difficulty","instruction","input","output","metadata","hash"}

def iter_jsonl(path: Path, max_rows: int | None) -> Iterable[Dict[str, Any]]:
    n = 0
    with path.open("r", encoding="utf-8") as f:
        for line in f:
            line = line.strip()
            if not line:
                continue
            try:
                obj = json.loads(line)
            except Exception:
                yield {"__parse_error__": True}
                n += 1
                if max_rows and n >= max_rows:
                    return
                continue
            yield obj
            n += 1
            if max_rows and n >= max_rows:
                return

def is_tool_trace_valid(out: Any) -> bool:
    # Accept either a parsed list OR a JSON string representing a list
    if isinstance(out, str):
        s = out.strip()
        if not s:
            return False
        try:
            out = json.loads(s)
        except Exception:
            return False
    if not isinstance(out, list) or len(out) == 0:
        return False
    for step in out:
        if not isinstance(step, dict):
            return False
        if "tool" not in step or "args" not in step:
            return False
        if not isinstance(step["tool"], str) or not isinstance(step["args"], dict):
            return False
    return True

def is_patch_diff_valid(out: Any) -> bool:
    if not isinstance(out, str):
        return False
    s = out.strip()
    if "```diff" in s:
        # fenced diff
        return ("-" in s and "+" in s)
    # also allow plain diff-like strings
    return s.startswith("-") or s.startswith("diff") or ("- " in s and "+ " in s)

def is_self_grade_valid(out: Any) -> bool:
    if isinstance(out, str):
        try:
            out = json.loads(out)
        except Exception:
            return False
    if not isinstance(out, dict):
        return False
    score = out.get("score")
    conf = out.get("confidence")
    notes = out.get("notes")
    if not (isinstance(score, int) and 0 <= score <= 10):
        return False
    if not (isinstance(conf, (int, float)) and 0.0 <= float(conf) <= 1.0):
        return False
    if not (isinstance(notes, str) and len(notes) >= 3):
        return False
    return True

def main() -> int:
    ap = argparse.ArgumentParser(description="Genesis AI Code Bench (Within Us AI)")
    ap.add_argument("--jsonl", required=True, help="Path to JSONL file")
    ap.add_argument("--max_rows", type=int, default=5000)
    args = ap.parse_args()

    path = Path(args.jsonl)
    total = 0
    format_valid = 0
    required_ok = 0
    tool_ok = tool_total = 0
    diff_ok = diff_total = 0
    grade_ok = grade_total = 0
    gov_ok = 0
    econ_ok = 0
    hashes = set()
    hash_total = 0

    for obj in iter_jsonl(path, args.max_rows):
        total += 1
        if obj.get("__parse_error__"):
            continue
        format_valid += 1

        keys = set(obj.keys())
        if REQUIRED_FIELDS.issubset(keys):
            required_ok += 1

        # hash / uniqueness
        h = obj.get("hash")
        if isinstance(h, str) and h:
            hash_total += 1
            hashes.add(h)

        # governance + economics (metadata)
        md = obj.get("metadata", {})
        if isinstance(md, dict):
            if md.get("audit_required") is True:
                gov_ok += 1
            if isinstance(md.get("cost_budget"), int) and isinstance(md.get("latency_ms_target"), int):
                econ_ok += 1

        # task-specific validity
        task = obj.get("task_type")
        out = obj.get("output")
        if task == "tool_trace":
            tool_total += 1
            if is_tool_trace_valid(out):
                tool_ok += 1
        elif task == "patch_diff":
            diff_total += 1
            if is_patch_diff_valid(out):
                diff_ok += 1
        elif task == "self_grade":
            grade_total += 1
            if is_self_grade_valid(out):
                grade_ok += 1

    def rate(num: int, den: int) -> float:
        return 0.0 if den == 0 else round(num / den, 4)

    report = {
        "rows_scanned": total,
        "format_valid_rate": rate(format_valid, total),
        "required_fields_rate": rate(required_ok, total),
        "tool_trace_valid_rate": rate(tool_ok, tool_total),
        "patch_diff_valid_rate": rate(diff_ok, diff_total),
        "self_grade_valid_rate": rate(grade_ok, grade_total),
        "governance_present_rate": rate(gov_ok, total),
        "economics_present_rate": rate(econ_ok, total),
        "uniqueness_rate": rate(len(hashes), hash_total),
    }

    print(json.dumps(report, indent=2))
    return 0

if __name__ == "__main__":
    raise SystemExit(main())