leviathan-OVERWATCH/scripts/gen-le-aircraft.py

#!/usr/bin/env python3
"""
Regenerate app/src/main/res/raw/le_aircraft.csv — the bundled list of
law-enforcement aircraft, keyed by the ICAO hex that ADS-B actually broadcasts.

    python3 scripts/gen-le-aircraft.py

Source: ADSBexchange's `basic-ac-db.json.gz` (~14.5 MB, ~618k aircraft), which
is derived from the FAA registry and already keyed by ICAO hex. That matters:
registry.faa.gov itself is behind Akamai bot mitigation and answers 403, and
the raw FAA registry is keyed by N-number, which would need a separate
N-number -> hex conversion step.

Why not the community plane-alert-db lists: measured 2026-09-17, their
plane-alert-pol.csv holds 996 aircraft but only 255 US-registered, and matching
all 2,760 of their police+government hexes against 394 live aircraft over a
250 nm sweep of Washington DC produced zero hits. Filtering this database by
owner name yields ~1,950 US law-enforcement aircraft — roughly 7.7x the US
coverage.

This runs at development time and commits its output; the app never downloads
it. Re-run occasionally to pick up registry changes.
"""
import gzip, io, json, re, sys, urllib.request
from pathlib import Path

SRC = "https://downloads.adsbexchange.com/downloads/basic-ac-db.json.gz"
OUT = Path(__file__).resolve().parent.parent / "app/src/main/res/raw/le_aircraft.csv"

# Owner-name patterns. Word boundaries matter: a bare "MARSHAL" matches
# "MARSHALL SPACE FLIGHT CENTER", and a bare "POLICE" would be fine but
# "SHERIFF" needs to catch both "SHERIFF" and "SHERIFFS".
LE = re.compile(r"(\bSHERIFF|\bPOLICE\b|\bSTATE PATROL\b|\bHIGHWAY PATROL\b|"
                r"\bSTATE TROOPER|\bPUBLIC SAFETY\b|\bUS MARSHAL\b|"
                r"\bMARSHALS SERVICE\b|\bCONSTABLE\b|\bDEPT OF CORRECTION|"
                r"\bDEPARTMENT OF CORRECTION|\bBORDER PATROL\b|"
                r"\bCUSTOMS AND BORDER|\bCUSTOMS & BOR|\bHOMELAND SECURITY\b|"
                r"\bDRUG ENFORCEMENT\b|\bFEDERAL BUREAU OF INVESTIGATION\b)", re.I)

# Air-ambulance and firefighting outfits share a lot of vocabulary with police
# aviation ("PUBLIC SAFETY"), and a medevac helicopter overhead is not the
# thing this app exists to warn about.
EXCLUDE = re.compile(r"MARSHALL SPACE|\bFIRE\b|AMBULANCE|MEDICAL|\bEMS\b|"
                     r"AIR METHODS|LIFE FLIGHT|MEDEVAC|HOSPITAL", re.I)


def main() -> int:
    print(f"fetching {SRC} …")
    req = urllib.request.Request(SRC, headers={"User-Agent": "OVERWATCH-build/1.0"})
    with urllib.request.urlopen(req, timeout=180) as r:
        raw = gzip.decompress(r.read())

    rows, ladd, seen = [], 0, 0
    for line in io.BytesIO(raw):
        line = line.decode("utf-8", "replace").strip().rstrip(",")
        if not line or line in "[]":
            continue
        try:
            rec = json.loads(line)
        except ValueError:
            continue
        seen += 1
        own = (rec.get("ownop") or "").strip()
        if not own or not LE.search(own) or EXCLUDE.search(own):
            continue
        hexid = (rec.get("icao") or "").strip().lower()
        if len(hexid) != 6:
            continue
        # Commas would break the one-line-per-record format; the owner is
        # display text only, so squashing punctuation is harmless.
        owner = re.sub(r"[,\r\n]+", " ", own)[:48].strip()
        if rec.get("faa_ladd"):
            ladd += 1
        rows.append((hexid, owner, "1" if rec.get("faa_ladd") else "0"))

    rows.sort()
    OUT.parent.mkdir(parents=True, exist_ok=True)
    with OUT.open("w", encoding="utf-8") as f:
        f.write("# hex,owner,ladd — generated by scripts/gen-le-aircraft.py, do not edit\n")
        for hexid, owner, l in rows:
            f.write(f"{hexid},{owner},{l}\n")

    print(f"scanned {seen} aircraft -> {len(rows)} law-enforcement ({ladd} LADD-flagged)")
    print(f"wrote {OUT} ({OUT.stat().st_size // 1024} KB)")
    return 0


if __name__ == "__main__":
    sys.exit(main())