Fr0zkaandClaude Opus 5.5
dee53c8d2a
Audits collect gates and run seams (landing discs + end runs, LOD0/1), tagged uncapped failures, separate exempt counts, self-overlap census; level end runs; overlap/turn constraints. Partial sweep: end runs pass 3/3 per seed; 1-2 end-turn violations per seed; 50-110k tagged seam mismatches per seed (real vs audit artefact not yet decided). Not verified; resume L174. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
226 lines
11 KiB
Python
226 lines
11 KiB
Python
#!/usr/bin/env python3
|
|
"""Capture composed passage, route, shell, and seam attribution for seeds 0..15."""
|
|
from __future__ import annotations
|
|
|
|
import argparse
|
|
import json
|
|
import re
|
|
import subprocess
|
|
import sys
|
|
import time
|
|
from pathlib import Path
|
|
|
|
|
|
ROOT = Path(__file__).resolve().parents[4]
|
|
DEFAULT_BIN = ROOT / "Standalone/bin/linux/VoxelForgeCoreHeadless-l174-step0-clang"
|
|
DATA = ROOT / "Standalone/data/owner_world.json"
|
|
STAMP_TOOL = ROOT / "Standalone/Tools/verify/source_stamp.py"
|
|
STAMP_RE = re.compile(r"^\[VoxelForgeSourceStamp\] sha256=([0-9a-f]+)$", re.M)
|
|
|
|
GATES = {
|
|
"passages": r"^\[ResolvedPassageAuditSummary\].*status=PASS$",
|
|
"seams": r"^\[PassageTileSeamAudit\].*status=PASS$",
|
|
"shell": r"^\[PassageShellAuditSummary\].*status=PASS$",
|
|
"naturalness": r"^\[PassageNaturalnessSummary\].*status=PASS$",
|
|
"mouth_profile": r"^\[PassageMouthProfileSummary\].*status=PASS$",
|
|
"ripple": r"^\[PassageRippleAuditSummary\].*status=PASS$",
|
|
"passage_interior": r"^\[PassageInteriorCapsuleAuditSummary\].*status=PASS$",
|
|
"self_overlap": r"^\[PassageTubeSelfOverlapAuditSummary\].*status=PASS$",
|
|
}
|
|
|
|
|
|
def evaluate(content: str, expected_stamp: str, layout_path: Path,
|
|
require_end_runs: bool = False) -> tuple[dict, dict, str | None]:
|
|
stamp = STAMP_RE.search(content)
|
|
checks = {name: re.search(pattern, content, re.M) is not None
|
|
for name, pattern in GATES.items()}
|
|
if require_end_runs:
|
|
checks["end_runs"] = re.search(
|
|
r"^\[PassageEndRunAuditSummary\].*status=PASS$", content, re.M) is not None
|
|
checks["source_stamp"] = stamp is not None and stamp.group(1) == expected_stamp
|
|
for kind, strate in (("tunnel_network", 2), ("vertical_shafts", 3)):
|
|
route = re.compile(
|
|
rf"^\[ResolvedStrateRouteAudit\] strate={strate} .*unsupported=0 "
|
|
rf"grade_failures=0 .*status=PASS$", re.M)
|
|
checks[f"{kind}_route"] = route.search(content) is not None
|
|
checks[f"{kind}_blocked_route_control"] = re.search(
|
|
rf"^\[{kind.title().replace('_', '')}BlockedRouteControl\].*"
|
|
r"expected=FAIL status=PASS$", content, re.M) is not None
|
|
checks["passage_count"] = len(re.findall(
|
|
r"^\[ResolvedPassageAudit\] .*walkability=PASS$", content, re.M)) == 3
|
|
checks["negative_controls"] = (
|
|
len(re.findall(r"^\[ResolvedPassageInteriorNegativeControl\].*"
|
|
r"detected=true status=PASS$", content, re.M)) == 3
|
|
and len(re.findall(r"^\[ResolvedPassageBlockedControl\].*detected=true$",
|
|
content, re.M)) == 3
|
|
)
|
|
attribution = {
|
|
"seam_samples_by_tag": {
|
|
tag: len(re.findall(rf"^\[PassageTileSeamSample\] tag={re.escape(tag)} ",
|
|
content, re.M))
|
|
for tag in ("disc", "rim", "end run", "body", "route")
|
|
},
|
|
"landing_support_failures_upper": len(re.findall(
|
|
r"^\[PassageLandingSupportFailure\].*side=upper", content, re.M)),
|
|
"landing_support_failures_lower": len(re.findall(
|
|
r"^\[PassageLandingSupportFailure\].*side=lower", content, re.M)),
|
|
"unsupported_walk_samples": len(re.findall(
|
|
r"^\[ResolvedWalkDispatch\] kind=unsupported", content, re.M)),
|
|
"shell_floor_samples": len(re.findall(
|
|
r"^\[PassageShellAuditSample\]", content, re.M)),
|
|
"shell_wall_samples": len(re.findall(
|
|
r"^\[PassageShellAuditSample\].*kind=wall_missing", content, re.M)),
|
|
"tube_self_overlap_rows": len(re.findall(
|
|
r"^\[PassageTubeSelfOverlapAudit\]", content, re.M)),
|
|
"tube_self_overlap_pairs": sum(int(value) for value in re.findall(
|
|
r"^\[PassageTubeSelfOverlapAudit\].*overlap_pairs=(\d+)",
|
|
content, re.M)),
|
|
"tube_end_turn_violations": sum(int(value) for value in re.findall(
|
|
r"^\[PassageTubeSelfOverlapAudit\].*end_turn_violations=(\d+)",
|
|
content, re.M)),
|
|
"tube_landing_column_violations": sum(int(value) for value in re.findall(
|
|
r"^\[PassageTubeSelfOverlapAudit\].*landing_column_violations=(\d+)",
|
|
content, re.M)),
|
|
"end_run_failures": len(re.findall(
|
|
r"^\[PassageEndRunAudit\].*status=FAIL$", content, re.M)),
|
|
"end_run_measured": len(re.findall(
|
|
r"^\[PassageEndRunAudit\].*measured_start_m=(?!0\.000)"
|
|
r".*measured_end_m=(?!0\.000)", content, re.M)),
|
|
}
|
|
seam_summary = re.search(r"^\[PassageTileSeamAudit\].*$", content, re.M)
|
|
if seam_summary:
|
|
for key in ("candidates", "examined", "pairs", "lod0_pairs", "lod0_samples",
|
|
"lod1_pairs", "lod1_samples", "landing_discs", "end_runs",
|
|
"lod0_disc_samples", "lod1_disc_samples", "lod0_rim_samples",
|
|
"lod1_rim_samples", "lod0_end_run_samples",
|
|
"lod1_end_run_samples", "exempt_pairs", "mismatches"):
|
|
match = re.search(rf"\b{key}=(\d+)", seam_summary.group(0))
|
|
attribution[f"seam_{key}"] = int(match.group(1)) if match else None
|
|
for key in ("landing_discs", "end_runs"):
|
|
match = re.search(rf"\b{key}=(\d+)/(\d+)", seam_summary.group(0))
|
|
if match:
|
|
attribution[f"seam_{key}_covered"] = int(match.group(1))
|
|
attribution[f"seam_{key}_required"] = int(match.group(2))
|
|
attribution["seam_uncovered_disc_regions"] = len(re.findall(
|
|
r"^\[PassageTileSeamCoverage\].*\bdisc_status=FAIL\b", content, re.M))
|
|
attribution["seam_uncovered_end_run_regions"] = len(re.findall(
|
|
r"^\[PassageTileSeamCoverage\].*\bend_run_status=FAIL\b", content, re.M))
|
|
attribution["seam_uncovered_regions"] = (
|
|
attribution["seam_uncovered_disc_regions"]
|
|
+attribution["seam_uncovered_end_run_regions"])
|
|
passage_summary = re.search(r"^\[ResolvedPassageAuditSummary\].*$", content, re.M)
|
|
if passage_summary:
|
|
for key in ("support_probes", "exempt_support_probes", "unsupported"):
|
|
match = re.search(rf"\b{key}=(\d+)", passage_summary.group(0))
|
|
if match:
|
|
attribution[key] = int(match.group(1))
|
|
identity = None
|
|
try:
|
|
if layout_path.is_file():
|
|
identity = json.loads(layout_path.read_text(encoding="utf-8")).get(
|
|
"stable_identity")
|
|
except (OSError, json.JSONDecodeError):
|
|
pass
|
|
checks["layout_output"] = identity is not None
|
|
return checks, attribution, identity
|
|
|
|
|
|
def main() -> int:
|
|
parser = argparse.ArgumentParser(description=__doc__)
|
|
parser.add_argument("--binary", type=Path, default=DEFAULT_BIN)
|
|
parser.add_argument("--out", type=Path,
|
|
default=ROOT / "Saved/Linux/l174/rebaseline-l173/seeds")
|
|
parser.add_argument("--seed-start", type=int, default=0)
|
|
parser.add_argument("--seed-stop", type=int, default=16,
|
|
help="exclusive seed bound")
|
|
parser.add_argument("--expected-stamp", type=str,
|
|
help="source stamp embedded in an earlier binary")
|
|
parser.add_argument("--require-end-runs", action="store_true",
|
|
help="require the L174 end-run audit (use for post-Step-1 sweeps)")
|
|
parser.add_argument("--no-resume", action="store_true",
|
|
help="rerun even when both the log and layout already exist")
|
|
args = parser.parse_args()
|
|
if not args.binary.is_file() or not DATA.is_file():
|
|
raise SystemExit(f"missing binary or owner fixture: {args.binary}")
|
|
args.binary = args.binary.resolve()
|
|
args.out = args.out.resolve()
|
|
args.out.mkdir(parents=True, exist_ok=True)
|
|
stamp = args.expected_stamp or subprocess.check_output(
|
|
[sys.executable, str(STAMP_TOOL), "stamp", "--root", str(ROOT)], text=True).strip()
|
|
results = []
|
|
partial = args.out / "runs-partial.json"
|
|
for seed in range(args.seed_start, args.seed_stop):
|
|
log = args.out / f"seed{seed}.log"
|
|
layout = args.out / f"seed{seed}.layout.json"
|
|
exit_path = args.out / f"seed{seed}.exit"
|
|
command = [
|
|
"taskset", "-c", "0-11", "prlimit", "--as=6442450944", "--",
|
|
"timeout", "1740s", str(args.binary), str(DATA),
|
|
"--world-assembler-prototype", "--stream-benchmark", "--stream-pipeline",
|
|
"--workers=8", "--clip-radius=3", "--max-clip-level=8", "--max-level=0",
|
|
"--force-generate", f"--seed={seed}", "--recipe=surface-world",
|
|
"--density-dispatch-report", "--audit-resolved-layout-passages",
|
|
"--audit-passage-tile-seams", f"--resolved-layout-output={layout}",
|
|
]
|
|
started = time.monotonic()
|
|
resumed = False
|
|
if not args.no_resume and log.is_file() and layout.is_file():
|
|
content = log.read_text(encoding="utf-8", errors="replace")
|
|
if exit_path.is_file():
|
|
code: int | str = int(exit_path.read_text().strip())
|
|
elif ("[PipelineExactQueryHash]" in content
|
|
and "[ResolvedPassageAuditSummary]" in content):
|
|
code = 10 if re.search(r"^\[[^]]*(?:Summary|Audit)\].*status=FAIL$",
|
|
content, re.M) else 0
|
|
else:
|
|
code = "NO_EXIT_RECORD"
|
|
resumed = code != "NO_EXIT_RECORD"
|
|
if not resumed:
|
|
try:
|
|
with log.open("w", encoding="utf-8") as output:
|
|
proc = subprocess.run(command, cwd=ROOT, stdout=output,
|
|
stderr=subprocess.STDOUT, check=False, timeout=1740)
|
|
code = proc.returncode
|
|
except subprocess.TimeoutExpired:
|
|
code = "OUTER_TIMEOUT"
|
|
if isinstance(code, int):
|
|
exit_path.write_text(str(code) + "\n", encoding="utf-8")
|
|
content = log.read_text(encoding="utf-8", errors="replace") if log.exists() else ""
|
|
checks, attribution, identity = evaluate(
|
|
content, stamp, layout, args.require_end_runs)
|
|
passed = code == 0 and all(checks.values())
|
|
failed = [name for name, ok in checks.items() if not ok]
|
|
row = {
|
|
"seed": seed,
|
|
"verdict": "PASS" if passed else "FAIL",
|
|
"exit": code,
|
|
"failed_gates": failed,
|
|
"checks": checks,
|
|
"attribution": attribution,
|
|
"source_stamp": stamp,
|
|
"layout_identity": identity,
|
|
"elapsed_s": round(time.monotonic() - started, 3),
|
|
"log": str(log.relative_to(ROOT)),
|
|
"layout": str(layout.relative_to(ROOT)),
|
|
"resumed": resumed,
|
|
}
|
|
results.append(row)
|
|
partial.write_text(json.dumps(results, indent=2) + "\n", encoding="utf-8")
|
|
print("[L174CaveSeedSweep] " + json.dumps(row, sort_keys=True), flush=True)
|
|
summary = {
|
|
"source_stamp": stamp,
|
|
"seeds": list(range(args.seed_start, args.seed_stop)),
|
|
"passed": sum(row["verdict"] == "PASS" for row in results),
|
|
"failed": sum(row["verdict"] != "PASS" for row in results),
|
|
"results": results,
|
|
}
|
|
(args.out / "summary.json").write_text(
|
|
json.dumps(summary, indent=2) + "\n", encoding="utf-8")
|
|
print(f"[L174CaveSeedSweepSummary] passed={summary['passed']}/{len(results)} "
|
|
f"failed={summary['failed']}", flush=True)
|
|
return 0 if summary["failed"] == 0 else 1
|
|
|
|
|
|
if __name__ == "__main__":
|
|
raise SystemExit(main())
|