#!/usr/bin/env python3
"""
log_summary.py: a first-pass datalog summary you (or your AI agent) can run
on any ECU CSV export. From https://vibetuners.com (free to use and modify).

    pip install pandas
    python3 log_summary.py my-log.csv

Edit COLUMNS to match your ECU's CSV headers (see your channel dictionary).
This is a starting point for an agent to extend, NOT a tuning tool. Check
every finding against the log yourself.
"""
import sys
import pandas as pd

COLUMNS = {
    "time":   "Time (s)",
    "rpm":    "Engine Speed (rpm)",
    "tps":    "Throttle Position (%)",
    "map":    "Manifold Pressure (kPa)",
    "lam":    "Lambda 1",
    "target": "Lambda Target",
    "ign":    "Ignition Angle (deg)",
    "knock":  "Knock Level (V)",
    "iat":    "Intake Air Temp (C)",
    "ect":    "Coolant Temp (C)",
}
LEAN_PCT = 3.0         # flag sustained error above this %
MIN_RUN = 5            # samples in a row to count as sustained
KNOCK_V = 0.8          # knock level considered an event
WIDEBAND_LAG_S = 0.10  # shift lambda earlier by this much
SKIP_TIPIN_S = 0.4     # ignore this long after throttle goes > 90%


def main(path):
    df = pd.read_csv(path)
    c = {k: v for k, v in COLUMNS.items() if v in df.columns}
    missing = [v for k, v in COLUMNS.items() if k not in c]
    dt = df[c["time"]].diff().median()
    print(f"File: {path}")
    print(f"Rows: {len(df)}  ·  ~{1/dt:.0f} Hz  ·  {df[c['time']].iloc[-1]:.1f} s")
    if missing:
        print(f"Missing expected channels: {', '.join(missing)}")

    # sensor sanity
    for k in ("lam", "map", "iat", "ect"):
        if k in c:
            s = df[c[k]]
            if s.nunique() <= 2:
                print(f"! SENSOR: {c[k]} is flat ({s.iloc[0]}). Check wiring before tuning.")

    # wideband lag correction
    if "lam" in c:
        shift = max(0, round(WIDEBAND_LAG_S / dt))
        df["lam_adj"] = df[c["lam"]].shift(-shift)

    # full-load segment, minus tip-in
    wot = df[df[c["tps"]] > 90].copy()
    if wot.empty:
        print("No full-load (TPS > 90%) data found.")
        return
    start = wot[c["time"]].iloc[0]
    wot = wot[wot[c["time"]] >= start + SKIP_TIPIN_S]
    print(f"\nFull load: {len(wot)} samples, "
          f"{wot[c['rpm']].min():.0f}-{wot[c['rpm']].max():.0f} rpm, "
          f"peak MAP {wot[c['map']].max():.0f} kPa")

    if "target" in c and "lam_adj" in wot:
        wot["err"] = (wot["lam_adj"] / wot[c["target"]] - 1) * 100
        for label, mask in (("LEAN", wot["err"] > LEAN_PCT), ("RICH", wot["err"] < -LEAN_PCT)):
            runs = (mask != mask.shift()).cumsum()
            for _, g in wot[mask].groupby(runs[mask]):
                if len(g) >= MIN_RUN:
                    pk = g["err"].abs().idxmax()
                    print(f"  {label}: {g[c['time']].iloc[0]:.2f}-{g[c['time']].iloc[-1]:.2f} s, "
                          f"{g[c['rpm']].min():.0f}-{g[c['rpm']].max():.0f} rpm, "
                          f"peak {wot.loc[pk, 'err']:+.1f}% at {wot.loc[pk, c['rpm']]:.0f} rpm / "
                          f"{wot.loc[pk, c['map']]:.0f} kPa")

    if "knock" in c:
        k = wot[wot[c["knock"]] > KNOCK_V]
        if len(k):
            print(f"  KNOCK: {len(k)} samples > {KNOCK_V} V, "
                  f"{k[c['rpm']].min():.0f}-{k[c['rpm']].max():.0f} rpm, peak {k[c['knock']].max():.2f} V")
            if "ign" in c:
                before = wot[wot[c["time"]] < k[c["time"]].iloc[0]].tail(10)[c["ign"]].median()
                print(f"         ignition {before:.1f}° just before → {k[c['ign']].min():.1f}° during event")

    for k, lim in (("iat", 55), ("ect", 100)):
        if k in c and df[c[k]].max() > lim:
            print(f"  TEMP: {c[k]} reached {df[c[k]].max():.1f} (limit {lim})")
    if "iat" in c:
        print(f"  IAT over pull: {wot[c['iat']].iloc[0]:.1f} → {wot[c['iat']].iloc[-1]:.1f} °C")


if __name__ == "__main__":
    if len(sys.argv) != 2:
        sys.exit("usage: python3 log_summary.py <log.csv>")
    main(sys.argv[1])