fix(uap): stop HF fallback from serving 3-year-old NUFORC sightings

The UAP sightings layer is sourced from a live scrape of nuforc.org with a
static Hugging Face CSV mirror (kcimc/NUFORC) as a fallback. The fallback
parsed every row, sorted by occurred-desc, and took the top 250 — with no
date cutoff. The HF mirror is a third-party snapshot that hasn't been
refreshed in years, so the "newest 250" rows it returns are from ~2022-23.
When the live path fails (Cloudflare 403, curl disabled on Windows, wdtNonce
regex stale, etc.) users see a map full of sightings from 3 years ago,
labeled as the "last 60 days" layer.

Changes:

* HF fallback now applies the same 60-day cutoff the live path uses. Rows
  outside the window are dropped before take-top-N. If the mirror has
  nothing inside the window the fallback returns [] (don't serve stale).
* When the HF mirror is fully stale a loud ERROR log fires with the count
  of dropped rows so the operator can tell the mirror's the problem, not
  a network issue.
* When BOTH live AND HF fallback produce 0 rows, fetch_uap_sightings now
  trips assert_canary("uap_sightings", 0) so the health registry shows
  the layer as broken instead of "fresh and empty for days."
* Scheduler moved from daily 12:00 UTC to weekly Mondays 12:00 UTC. The
  layer is a rolling 60-day digest; refreshing once a week is enough
  cadence for human-readable map exploration and keeps nuforc.org load
  light.

6 new tests cover the cutoff filter, the doomsday-log path, the mixed-age
path, the both-paths-empty health failure, the positive fallback path, and
the scheduler cadence.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

backend/services/data_fetcher.py

@@ -960,16 +960,19 @@ def start_scheduler():
         misfire_grace_time=600,
     )
 
-    # UAP sightings (NUFORC) — daily at 12:00 UTC
+    # UAP sightings (NUFORC) — weekly on Mondays at 12:00 UTC. The layer is a
+    # rolling last-60-days digest; refreshing once a week is enough cadence
+    # for human-readable map exploration and keeps load on nuforc.org light.
     _scheduler.add_job(
         lambda: _run_task_with_health(
             lambda: fetch_uap_sightings(force_refresh=True),
             "fetch_uap_sightings",
         ),
         "cron",
+        day_of_week="mon",
         hour=12,
         minute=0,
-        id="uap_sightings_daily",
+        id="uap_sightings_weekly",
         max_instances=1,
         misfire_grace_time=3600,
     )

backend/services/fetchers/earth_observation.py

@@ -1383,10 +1383,21 @@ def _build_uap_sightings_from_hf_mirror() -> list[dict]:
     This is a resilience fallback for local/Windows runs where nuforc.org is
     Cloudflare-gated and the Mapbox token is not configured. It is not as fresh
     as the live NUFORC AJAX feed, but it keeps the layer visible and cached.
+
+    Date-cutoff guard: the kcimc/NUFORC HF dataset is a static snapshot whose
+    maintainer refreshes it sporadically. Without a cutoff, sorting by
+    occurred-desc and taking the top N rows returns whatever the mirror's
+    newest rows happen to be — which can be years old if the snapshot is
+    stale. We apply the same ``_NUFORC_RECENT_DAYS`` window the live path
+    uses (60 days). If the HF mirror has nothing inside the window we return
+    ``[]`` rather than silently serving 3-year-old "newest" rows.
     """
     from services.fetchers.nuforc_enrichment import _HF_CSV_URL, _parse_date
     from services.geocode_validate import coord_in_country
 
+    cutoff_dt = datetime.utcnow() - timedelta(days=_NUFORC_RECENT_DAYS)
+    cutoff_str = cutoff_dt.strftime("%Y-%m-%d")
+
     try:
         response = fetch_with_curl(_HF_CSV_URL, timeout=180, follow_redirects=True)
         if not response or response.status_code != 200:
@@ -1400,6 +1411,7 @@ def _build_uap_sightings_from_hf_mirror() -> list[dict]:
         return []
 
     candidates: list[dict] = []
+    stale_rows_dropped = 0
     try:
         reader = csv.DictReader(io.StringIO(response.text))
         for row in reader:
@@ -1410,6 +1422,9 @@ def _build_uap_sightings_from_hf_mirror() -> list[dict]:
             )
             if not occurred:
                 continue
+            if occurred < cutoff_str:
+                stale_rows_dropped += 1
+                continue
             raw_location = _normalize_uap_location(
                 row.get("Location", "")
                 or row.get("City", "")
@@ -1444,6 +1459,19 @@ def _build_uap_sightings_from_hf_mirror() -> list[dict]:
         logger.warning("UAP sightings: HF fallback parse failed: %s", e)
         return []
 
+    if not candidates:
+        # HF mirror returned rows, but none inside the rolling window. This is
+        # the smoking gun for "the public HF dataset hasn't been refreshed in
+        # years" — log loudly so the operator sees it instead of guessing.
+        logger.error(
+            "UAP sightings: HF fallback yielded 0 rows within last %d days "
+            "(dropped %d stale rows). HF mirror is likely stale; the layer "
+            "will be empty until the live NUFORC path recovers.",
+            _NUFORC_RECENT_DAYS,
+            stale_rows_dropped,
+        )
+        return []
+
     candidates.sort(key=lambda row: (row["occurred"], row["posted"], row["id"]), reverse=True)
     candidates = candidates[:_NUFORC_HF_FALLBACK_LIMIT]
 
@@ -1515,13 +1543,29 @@ def fetch_uap_sightings(*, force_refresh: bool = False):
 
     sightings = _load_nuforc_sightings_cache(force_refresh=force_refresh)
     if sightings is None:
+        live_error: Exception | None = None
         try:
             sightings = _build_recent_uap_sightings()
         except Exception as e:
+            live_error = e
             logger.warning("UAP sightings: live NUFORC rebuild failed, using fallback: %s", e)
             sightings = _build_uap_sightings_from_hf_mirror()
         if sightings:
             _save_nuforc_sightings_cache(sightings)
+        elif live_error is not None:
+            # Both paths failed: live raised AND HF fallback returned empty
+            # (either the HF mirror is stale beyond the cutoff or the network
+            # is gone entirely). The previous code silently set the layer to
+            # ``[]`` and kept marking it fresh; that masked the failure for
+            # days. Surface it via assert_canary so the health registry shows
+            # the layer as broken instead of "fresh and empty".
+            from services.slo import assert_canary
+            assert_canary("uap_sightings", 0)
+            logger.error(
+                "UAP sightings: both live NUFORC and HF fallback produced 0 "
+                "rows; layer is unavailable. Live error: %s",
+                live_error,
+            )
 
     with _data_lock:
         latest_data["uap_sightings"] = sightings or []