File size: 4,006 Bytes
3b92b73 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 | #!/usr/bin/env python3
"""Derive the graph-free direct GPT-OSS baseline from saved model choices.
This script performs no inference and loads no benchmark labels. It releases a
case iff the frozen primary GPT-OSS choice is NONE; otherwise it reviews only
the selected component. Governance actions use the benchmark's public,
deterministic authority/event router shared by the open-weight reference arms.
"""
from __future__ import annotations
import argparse
import json
from pathlib import Path
ROOT = Path(__file__).resolve().parents[1]
def load_jsonl(path: Path) -> list[dict]:
return [json.loads(line) for line in path.read_text().splitlines() if line.strip()]
def default_measurements() -> Path:
released = ROOT / "results/open_weight/gptoss20b_direct_measurements.jsonl"
if released.exists():
return released
return ROOT / "results/open_weight_full_archive/gptoss20b_open_weight_token_sseg_measurements.jsonl"
def input_rows() -> list[dict]:
base = ROOT / "closed_loop_inputs" if (ROOT / "closed_loop_inputs").exists() else ROOT / "hf/closed_loop_inputs"
return sorted(
(row for split in ("development", "validation", "test") for row in load_jsonl(base / f"{split}.jsonl")),
key=lambda row: row["case_id"],
)
def action_for(row: dict, release: bool) -> str:
if release:
return "release"
trigger = row["event"]["trigger_category"]
tier = row["workflow_context"]["authority_tier"]
if any(c.get("status") == "failed" and not c.get("compensable", True) for c in row.get("red_line_controls", [])):
return "suspend"
if trigger == "tool_permission":
return "restrict" if tier.startswith("C") else "scoped_revalidation"
if trigger == "incident_near_miss":
return "withhold"
if trigger in {"performance_behaviour", "model_vendor"}:
return "escalate"
if trigger == "prompt_policy_orchestration":
return "withhold"
if trigger == "purpose_population":
return "scoped_revalidation"
if trigger == "memory_data":
return "rollback" if tier.startswith("A") else ("withhold" if row["domain"] == "market_data" else "scoped_revalidation")
return "escalate"
def main() -> None:
parser = argparse.ArgumentParser()
parser.add_argument("--measurements", type=Path, default=default_measurements())
parser.add_argument("--output", type=Path, default=ROOT / "results/open_weight/direct_llm.jsonl")
args = parser.parse_args()
measurements = {row["case_id"]: row for row in load_jsonl(args.measurements)}
predictions = []
for row in input_rows():
measurement = measurements[row["case_id"]]
primary = measurement["observed_primary"]
selected = primary.get("selected_node") or "NONE"
release = selected == "NONE"
scope = [] if release else [selected]
predictions.append({
"case_id": row["case_id"],
"method": "direct_llm",
"model": measurement.get("model_requested", "openai/gpt-oss-20b"),
"measurement_channel": "frozen primary model choice; no probability gate or graph propagation",
"iceberg_detected": not release,
"governance_action": action_for(row, release),
"affected_components": scope,
"revalidation_components": scope,
"reviewed_components": scope,
"release_permitted": release,
"diagnostic": {"selected_code": primary.get("selected_code"), "selected_node": selected},
})
args.output.parent.mkdir(parents=True, exist_ok=True)
args.output.write_text("".join(json.dumps(row, sort_keys=True) + "\n" for row in predictions))
print(json.dumps({
"method": "direct_llm",
"predictions": len(predictions),
"model_calls": 0,
"uses_probability_gate": False,
"uses_graph_propagation": False,
"output": str(args.output),
}, indent=2))
if __name__ == "__main__":
main()
|