"""Recompute fact-ledger metrics for chain experiments from the stored hop messages (runs/raw/chains/*.json) with the
current `_ledger_metrics` implementation, so that metric fixes apply uniformly to all runs (records keep the original values)."""
from __future__ import annotations
import json, glob, re
from pathlib import Path
import pandas as pd
from ..util import ROOT, read_json
from ..chains import _ledger_metrics
from ..data.prepare import load_doc, load_key

def recompute(df: pd.DataFrame, exp: str) -> pd.DataFrame:
    files = glob.glob(str(ROOT / "runs/raw/chains" / f"{exp}_*.json"))
    cache = {}
    for f in files:
        name = Path(f).stem[len(exp) + 1:]
        m = re.match(r"(syn_\d+_\d+)_(.*)", name)
        if not m: continue
        doc_id, cid = m.group(1), m.group(2).replace("_", "|", 2)
        d = json.load(open(f))
        key = load_key(doc_id, "synthetic"); doc = load_doc(doc_id, "synthetic")
        for h, msg in d["hops"].items():
            L = _ledger_metrics(msg, key["facts"], doc["text"])
            cache[(doc_id, cid, int(h))] = L
    def upd(row):
        cid = row["run_group_id"].split(":")[1].replace("-", "_") if isinstance(row["run_group_id"], str) else None
        for (doc_id, c, h), L in cache.items():
            if doc_id == row["document_id"] and h == row["handoff_index"] and c.replace("-", "_") == cid.replace("-", "_"):
                row["retention_rate"] = L["retention_rate"]; row["unsupported"] = L["unsupported_numbers"] + L["unsupported_identifiers"]; row["distorted"] = L["distorted_sentences"]
                break
        return row
    return df.apply(upd, axis=1)