Reject a not-yet-backfilled archive instead of caching it as complete (#225)

A self-hosted Open-Meteo instance that isn't backfilled yet answers historical
requests with all-null values (or only its recent sync window), which
_finalize_frame reduces to a near-empty frame. That frame is not None, so the
loader would accept it as source=open-meteo and cache it indefinitely as a
complete record — never falling back to NASA and never self-healing after the
backfill lands. Require at least MIN_ARCHIVE_DAYS before accepting an archive
result; below that, fall through to the NASA backup and don't cache the short
frame. Guards the prod cutover window.

data/climate.py

@@ -23,6 +23,12 @@ CACHE_DIR = os.path.join(paths.DATA_DIR, "cache")
 MAX_ATTEMPTS = 3                   # per upstream call, before giving up
 START_DATE = "1980-01-01"          # ERA5 reaches back to 1940; 1980 = 45 yrs, fast + robust
 ARCHIVE_LATENCY_DAYS = 6           # ERA5 archive lags real time by a few days
+# A real archive spans decades (~16k daily rows from START_DATE). A self-hosted
+# Open-Meteo instance that isn't backfilled yet answers with all-null values or only
+# its recent sync window, which _finalize_frame reduces to a near-empty frame. Below
+# this many days we treat the archive as "not ready" — fall through to the NASA backup
+# and do NOT cache the short result as a complete, indefinitely-held record.
+MIN_ARCHIVE_DAYS = 3650            # ~10 yrs: far above any sync window, far below a full archive
 # History rarely changes, so the full multi-decade archive is fetched once and
 # cached indefinitely (refetched only to add new metric columns). Only the recent
 # tail is refreshed — a small incremental fetch, at most hourly.
@@ -585,7 +591,13 @@ def _load_history(cell: dict) -> tuple[pl.DataFrame, dict]:
         source = None
         if time.time() >= _archive_cooldown_until:
             try:
-                df = _fetch_history(cell)
+                fetched = _fetch_history(cell)
+                # Guard a self-hosted archive that isn't backfilled yet: an all-null or
+                # recent-only response reduces to a near-empty frame, which would
+                # otherwise be cached indefinitely as a complete record and never
+                # fall to NASA. Require a plausibly-full span before accepting it.
+                if fetched.height >= MIN_ARCHIVE_DAYS:
+                    df = fetched
                     source = "open-meteo"
             except Exception as e:  # noqa: BLE001
                 if is_rate_limit(e):

tests/data/test_climate.py

@@ -249,3 +249,49 @@ def test_daily_params_carry_no_model_by_default():
     p = climate._om_daily_params({"center_lat": 1.0, "center_lon": 2.0}, past_days=7)
     assert "models" not in p
     assert climate.ARCHIVE_URL.endswith("/v1/archive")   # default (no env override in tests)
+
+
+def _full_history_frame(n=4000):
+    """A plausibly-full archive frame (>= MIN_ARCHIVE_DAYS rows), standard columns."""
+    start = datetime.date(1990, 1, 1)
+    dates = [start + datetime.timedelta(days=i) for i in range(n)]
+    return climate._finalize_frame(pl.DataFrame({
+        "date": dates,
+        "tmax": [70.0] * n, "tmin": [50.0] * n, "precip": [0.0] * n,
+        "wind": [5.0] * n, "gust": [8.0] * n, "humid": [60.0] * n,
+        "fmax": [70.0] * n, "fmin": [50.0] * n,
+    }))
+
+
+def test_short_archive_is_rejected_and_falls_back_to_nasa(monkeypatch, tmp_path):
+    """A self-hosted archive that isn't backfilled yet returns too few days; the load
+    must reject it (not cache a near-empty history as complete) and use NASA instead."""
+    monkeypatch.setattr(climate, "CACHE_DIR", str(tmp_path))
+    monkeypatch.setattr(climate, "_archive_cooldown_until", 0.0)
+    cell = {"id": "short_cell", "center_lat": 47.6, "center_lon": -122.3}
+
+    short = climate._to_frame(_om_daily(3))                     # 2 usable days « threshold
+    full = _full_history_frame()                                # >= MIN_ARCHIVE_DAYS
+    assert short.height < climate.MIN_ARCHIVE_DAYS <= full.height
+    monkeypatch.setattr(climate, "_fetch_history", lambda c: short)
+    monkeypatch.setattr(climate, "_fetch_history_nasa", lambda c: full)
+
+    df, meta = climate._load_history(cell)
+    assert meta["source"] == "nasa-power"      # rejected the short archive
+    assert df.height == full.height
+
+
+def test_full_archive_is_accepted(monkeypatch, tmp_path):
+    """A backfilled archive (enough days) is accepted and served as open-meteo."""
+    monkeypatch.setattr(climate, "CACHE_DIR", str(tmp_path))
+    monkeypatch.setattr(climate, "_archive_cooldown_until", 0.0)
+    cell = {"id": "full_cell", "center_lat": 47.6, "center_lon": -122.3}
+
+    full = _full_history_frame()
+    monkeypatch.setattr(climate, "_fetch_history", lambda c: full)
+    def _nasa_should_not_run(c): raise AssertionError("NASA should not be called")
+    monkeypatch.setattr(climate, "_fetch_history_nasa", _nasa_should_not_run)
+
+    df, meta = climate._load_history(cell)
+    assert meta["source"] == "open-meteo"
+    assert df.height == full.height