#!/usr/bin/env python3
"""Aggregate the frozen BabyAI population-predictor capacity screen."""

import argparse
import json
from pathlib import Path
import statistics


ROOT = Path(__file__).resolve().parents[1]
DEFAULT_RESULTS = ROOT / "results" / "babyai_shared" / "population_p3"
DEFAULT_OUT = ROOT / "results" / "babyai_shared" / "population_p3_analysis.json"
SEEDS = (4101, 4102, 4103)


def main():
    parser = argparse.ArgumentParser()
    parser.add_argument("--results", type=Path, default=DEFAULT_RESULTS)
    parser.add_argument("--out", type=Path, default=DEFAULT_OUT)
    args = parser.parse_args()
    records = []
    for seed in SEEDS:
        with open(args.results / f"seed{seed}.json", encoding="utf-8") as handle:
            records.append(json.load(handle))
    rows = [
        {
            "seed": record["model_seed"],
            "epoch": audit["epoch"],
            "layer": layer["layer"],
            **layer["holdout"],
        }
        for record in records for audit in record["audits"]
        for layer in audit["layers"]
    ]
    checks = {
        "all_seed_gates_pass": all(record["gate"] == "pass"
                                    for record in records),
        "r2_at_least_0p8_every_seed_epoch_layer": all(
            row["mean_per_cell_r2"] >= 0.8 for row in rows),
        "residual_ratio_at_most_0p25_every_seed_epoch_layer": all(
            row["residual_context_rms_ratio"] <= 0.25 for row in rows),
        "zero_action_or_teaching_observations": all(
            layer["action_observations"] == 0
            and layer["teaching_observations"] == 0
            for record in records for audit in record["audits"]
            for layer in audit["layers"]),
    }
    epoch_layer = []
    for epoch in (0, 1, 5, 10, 20, 40):
        for layer in range(4):
            selected = [row for row in rows
                        if row["epoch"] == epoch and row["layer"] == layer]
            epoch_layer.append({
                "epoch": epoch,
                "layer": layer,
                "mean_holdout_r2": statistics.mean(
                    row["mean_per_cell_r2"] for row in selected),
                "mean_holdout_residual_ratio": statistics.mean(
                    row["residual_context_rms_ratio"] for row in selected),
            })
    worst_r2 = min(rows, key=lambda row: row["mean_per_cell_r2"])
    worst_ratio = max(
        rows, key=lambda row: row["residual_context_rms_ratio"])
    report = {
        "stage": "babyai_population_p3_analysis",
        "gate": "pass" if all(checks.values()) else "fail",
        "checks": checks,
        "minimum_holdout_r2": worst_r2,
        "maximum_holdout_residual_context_rms_ratio": worst_ratio,
        "epoch_layer_means_across_seeds": epoch_layer,
        "raw_sdil_rollout_or_test_outcomes_read": False,
        "decision": (
            "Close the population-linear predictor rescue. It is highly "
            "predictive at initialization but fails the frozen early-layer "
            "threshold after clean task learning; do not run another "
            "PickupLoc downstream endpoint."),
    }
    args.out.parent.mkdir(parents=True, exist_ok=True)
    with open(args.out, "w", encoding="utf-8") as handle:
        json.dump(report, handle, indent=2, sort_keys=True)
        handle.write("\n")
    print(json.dumps(report, indent=2, sort_keys=True))


if __name__ == "__main__":
    main()