Jyotisha/scripts/research/midnight_date_anchor_regression.py

#!/usr/bin/env python3
"""Independent-process native A/B and dated-window goldens; not an accuracy benchmark."""
from __future__ import annotations

import argparse
import hashlib
import importlib
import json
import sys
import subprocess
from pathlib import Path

ROOT = Path(__file__).resolve().parents[2]


def load(root: Path):
    for name in list(sys.modules):
        if name == "scripts" or name.startswith("scripts."):
            del sys.modules[name]
    sys.path.insert(0, str(root))
    importlib.invalidate_caches()
    from scripts.rectification.api_service import score_candidates
    from scripts.research.probe_supply_after_six import request_from_case
    from scripts.rectification.decision_policy import indistinguishable_width_minutes
    return score_candidates, request_from_case, indistinguishable_width_minutes


def canonical(value):
    return json.dumps(value, sort_keys=True, ensure_ascii=True, separators=(",", ":")).encode()


def projection(result, width):
    new_fields = {"candidate_id", "candidate_date", "window_index", "window_offset_minutes", "segment_index", "cluster_intervals"}
    return {
        "candidate_scores": result["candidate_scores"],
        "matrix": result["event_contribution_matrix"],
        "decisions": [{key: value for key, value in row.items() if key not in new_fields} for row in result["candidate_decisions"]],
        "width": width(result["candidate_decisions"]),
        "confirmation_allowed": result["confirmation_allowed"],
        "selection_allowed": result["selection_allowed"],
        "representative_time": result["decision_receipt"]["representative_time"],
    }


def worker(root: Path, dated: bool):
    score, make_request, width = load(root)
    dataset = ROOT / "references/real_case_calibration/minute_rectification_holdout_v3.json"
    result = []
    for case in json.loads(dataset.read_text(encoding="utf-8"))["cases"][:3]:
        request = make_request(case)
        if dated:
            request["candidate_intervals"] = [{"start_at": f'{request["birth_date"]}T{request["start_time"]}', "end_at": f'{request["birth_date"]}T{request["end_time"]}'}]
        result.append(projection(score(request, include_internal=True), width))
    print(json.dumps(result))


def independent(root: Path, dated: bool):
    completed = subprocess.run([sys.executable, str(Path(__file__).resolve()), "--worker", str(root), *( ["--dated"] if dated else [])],
                               cwd=root, check=True, capture_output=True, text=True, encoding="utf-8")
    return json.loads(completed.stdout)


def main():
    if "--worker" in sys.argv:
        worker(Path(sys.argv[sys.argv.index("--worker") + 1]), "--dated" in sys.argv)
        return
    parser = argparse.ArgumentParser()
    parser.add_argument("--baseline", type=Path, required=True)
    parser.add_argument("--output", type=Path, default=ROOT / "artifacts/midnight-date-anchor")
    parser.add_argument("--golden", action="store_true")
    parser.add_argument("--golden-path", type=Path, default=ROOT / "frontend/tests/fixtures/rectification-midnight-date-anchor.native.json")
    args = parser.parse_args()
    args.output.mkdir(parents=True, exist_ok=True)
    dataset = ROOT / "references/real_case_calibration/minute_rectification_holdout_v3.json"
    cases = json.loads(dataset.read_text(encoding="utf-8"))["cases"][:3]
    assert all(case["birth"]["source"]["rodden_rating"] == "AA" for case in cases)
    old = independent(args.baseline, False)
    current = independent(ROOT, True)
    comparisons = []
    for case, before, after in zip(cases, old, current):
        fields = {key: canonical(before[key]) == canonical(after[key]) for key in before}
        comparisons.append({"case_id": case["case_id"], "source": case["birth"]["source"], "fields": fields,
            "before_sha256": hashlib.sha256(canonical(before)).hexdigest(), "after_sha256": hashlib.sha256(canonical(after)).hexdigest(),
            "candidate_minutes": len(after["candidate_scores"]), "public_clusters": len(after["decisions"]), "width": after["width"]})
        print("current", case["case_id"], fields, flush=True)
    report = {"baseline": str(args.baseline), "dataset_sha256": hashlib.sha256(dataset.read_bytes()).hexdigest(),
              "same_machine_independent_processes": True, "python_executable": sys.executable, "numeric_tolerance": 0, "comparisons": comparisons,
              "excluded_fields": "new date/ordinal metadata; result/candidate IDs intentionally change with algorithm identity"}
    with (args.output / "independent-process-aa-ab.json").open("x", encoding="utf-8") as stream:
        json.dump(report, stream, indent=2)
    assert all(all(row["fields"].values()) for row in comparisons), "same-day native A/B changed"
    if args.golden:
        score, _, _ = load(ROOT)
        # Explicitly fictional birth/event facts; response is produced only by the unmocked native engine.
        request = {"birth_date": "2000-03-01", "start_time": "23:58", "end_time": "00:02", "lat": 0.0, "lon": 0.0, "tz": 0.0,
            "candidate_intervals": [{"start_at": "2000-02-29T23:58", "end_at": "2000-03-01T00:02"}],
            "events": [{"id": "00000000-0000-4000-8000-000000000001", "domain": "career", "event_kind": "career_entry",
                        "date_start": "2020-01-01", "date_end": "2020-01-01", "precision": "day", "summary": "Fictional career entry for date-contract testing"}]}
        from scripts.rectification.contracts import normalize_rectification_request
        response = score(normalize_rectification_request(request))
        target = args.golden_path
        with target.open("x", encoding="utf-8") as stream:
            json.dump({"provenance": {"kind": "unmocked_native_engine", "facts": "explicitly_fictional", "generator": "scripts/research/midnight_date_anchor_regression.py --baseline <baseline-worktree> --golden", "algorithm": response["algorithm_version"]}, "request": request, "response": response}, stream, ensure_ascii=False, indent=2)
        print("golden", target, flush=True)


if __name__ == "__main__":
    main()