Fix treadmill/indoor distance over-measure and clean up polluted PRs

- Garmin Connect sync now applies Garmins corrected summary distance/moving time,
  overriding the raw wrist-estimated FIT distance for treadmill/indoor runs
- Exclude indoor (no-GPS) runs from distance personal records (bogus fast splits)
- backfill_indoor_distances task: re-fetch corrected distance for historical indoor runs
- recompute_personal_records_all task: wipe and rebuild PRs from valid activities

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

backend/app/services/garmin_connect_sync.py

@@ -181,7 +181,14 @@ def sync_activities(garmin, user_id: int, since: Optional[datetime],
         dest = dest_dir / f"{garmin_id}.fit"
         dest.write_bytes(fit_data)
 
-        process_activity_file.delay(str(dest), user_id, "fit", garmin_id)
+        # Pass Garmin's corrected summary so treadmill/indoor wrist-distance is
+        # overridden by the authoritative displayed distance/moving time.
+        summary = {
+            "distance": act.get("distance"),
+            "moving": act.get("movingDuration"),
+            "elapsed": act.get("elapsedDuration") or act.get("duration"),
+        }
+        process_activity_file.delay(str(dest), user_id, "fit", garmin_id, summary)
         queued += 1
 
         if status_callback and (queued % 5 == 0 or queued == total):

backend/app/workers/tasks.py

@@ -52,10 +52,38 @@ def is_wellness_file(file_path: str) -> bool:
     return any(name.endswith(s.upper()) for s in WELLNESS_SUFFIXES)
 
 
+def _apply_garmin_summary(parsed: dict, summary: dict):
+    """Override FIT-parsed distance/time with Garmin Connect's corrected summary
+    values. Treadmill/indoor runs report wrist-estimated distance in the raw FIT
+    that Garmin later corrects; the summary is the authoritative displayed value.
+    Scales the per-point distance stream so splits/charts stay consistent."""
+    if not summary:
+        return
+    corr = summary.get("distance")
+    raw = parsed.get("distance_m")
+    if corr:
+        if raw and raw > 0 and abs(corr - raw) / raw > 0.01:
+            scale = corr / raw
+            for p in parsed.get("data_points", []):
+                if p.get("distance_m") is not None:
+                    p["distance_m"] = p["distance_m"] * scale
+        parsed["distance_m"] = corr
+    if summary.get("moving") is not None:
+        parsed["moving_time_s"] = summary["moving"]
+    if summary.get("elapsed") is not None:
+        parsed["duration_s"] = summary["elapsed"]
+    dur = parsed.get("duration_s")
+    if parsed.get("distance_m") and dur:
+        parsed["avg_speed_ms"] = parsed["distance_m"] / dur
+
+
 @celery_app.task(bind=True, name="process_activity_file")
 def process_activity_file(self, file_path: str, user_id: int, source_type: str,
-                          garmin_activity_id: str = None):
+                          garmin_activity_id: str = None, summary: dict = None):
-    """Parse a FIT/GPX file. Routes wellness files to health parser."""
+    """Parse a FIT/GPX file. Routes wellness files to health parser.
+
+    `summary` (optional, from Garmin Connect sync) carries Garmin's corrected
+    distance/moving/elapsed values which override the raw FIT figures."""
 
     if is_wellness_file(file_path):
         parse_wellness_fit.delay(file_path, user_id)
@@ -80,6 +108,9 @@ def process_activity_file(self, file_path: str, user_id: int, source_type: str,
     if not parsed.get("start_time"):
         return {"status": "skipped", "reason": "no start_time", "file": file_path}
 
+    # Prefer Garmin's corrected distance/time (treadmill/indoor wrist-distance fix).
+    _apply_garmin_summary(parsed, summary)
+
     # Reject activities whose average speed is implausible for the sport (e.g. a
     # car journey accidentally recorded). Surfaced to the upload UI as the reason.
     if parsed.get("rejected_reason"):
@@ -387,6 +418,12 @@ def compute_personal_records(activity_id: int, user_id: int, parsed: dict):
     start_time_str = parsed.get("start_time")
     start_time = datetime.fromisoformat(start_time_str) if start_time_str else datetime.now(timezone.utc)
 
+    # Treadmill/indoor runs have no GPS and an unreliable distance stream, which
+    # produces bogus fast splits — exclude them from distance records.
+    has_gps = any(p.get("latitude") is not None and p.get("longitude") is not None for p in data_points)
+    if sport == "running" and not has_gps:
+        return {"status": "skipped_indoor", "activity_id": activity_id}
+
     # GPS can over/under-measure relative to the activity's official distance
     # (e.g. a 5 km run whose GPS track sums to 5.8 km), which would otherwise
     # produce a bogus "best 5 km" split. Scale the distance stream so its max
@@ -865,3 +902,131 @@ def backfill_moving_time(user_id: int = None):
         db.commit()
 
     return {"status": "ok", "updated": updated, "skipped": skipped}
+
+
+@celery_app.task(name="backfill_indoor_distances")
+def backfill_indoor_distances(user_id: int):
+    """Re-fetch Garmin's corrected distance/moving-time for historical indoor
+    (no-GPS) running activities whose raw FIT held the wrist-estimated distance.
+    Scales the stored per-point distance stream to match. One-off maintenance."""
+    import time
+    from app.services.garmin_connect_sync import authenticate_garmin
+    from app.core.database import SyncSessionLocal
+    from app.models.user import GarminConnectConfig, Activity
+    from sqlalchemy import select, text
+
+    updated, skipped = 0, 0
+    with SyncSessionLocal() as db:
+        cfg = db.execute(
+            select(GarminConnectConfig).where(GarminConnectConfig.user_id == user_id)
+        ).scalar_one_or_none()
+        if not cfg:
+            return {"status": "no_garmin_config"}
+
+        garmin, new_token = authenticate_garmin(cfg.email, cfg.password_enc, cfg.token_store)
+        if new_token:
+            cfg.token_store = new_token
+            db.commit()
+
+        acts = db.execute(
+            select(Activity).where(
+                Activity.user_id == user_id,
+                Activity.sport_type == "running",
+                Activity.polyline.is_(None),
+                Activity.garmin_activity_id.isnot(None),
+            )
+        ).scalars().all()
+
+        for a in acts:
+            try:
+                summ = (garmin.get_activity(int(a.garmin_activity_id)) or {}).get("summaryDTO", {})
+            except Exception:
+                skipped += 1
+                continue
+            corr = summ.get("distance")
+            if not corr:
+                skipped += 1
+                continue
+            raw = a.distance_m or 0
+            if raw > 0 and abs(corr - raw) / raw > 0.01:
+                db.execute(
+                    text("UPDATE activity_data_points SET distance_m = distance_m * :s "
+                         "WHERE activity_id = :aid AND distance_m IS NOT NULL"),
+                    {"s": corr / raw, "aid": a.id},
+                )
+            a.distance_m = corr
+            if summ.get("movingDuration") is not None:
+                a.moving_time_s = summ["movingDuration"]
+            if summ.get("elapsedDuration") is not None:
+                a.duration_s = summ["elapsedDuration"]
+            if a.duration_s:
+                a.avg_speed_ms = corr / a.duration_s
+            db.commit()
+            updated += 1
+            time.sleep(0.4)
+
+    return {"status": "ok", "updated": updated, "skipped": skipped}
+
+
+@celery_app.task(name="recompute_personal_records_all")
+def recompute_personal_records_all(user_id: int):
+    """Wipe and rebuild all personal records from stored activity data, excluding
+    indoor (no-GPS) runs whose distance is unreliable. Fixes records polluted by
+    treadmill over-measurement and any duplicate current-record rows."""
+    from app.services.route_matcher import compute_best_splits, STANDARD_DISTANCES
+    from app.core.database import SyncSessionLocal
+    from app.models.user import Activity, ActivityDataPoint, PersonalRecord
+    from sqlalchemy import select, delete
+
+    dist_for_label = {l: d for d, l in STANDARD_DISTANCES}
+    with SyncSessionLocal() as db:
+        db.execute(delete(PersonalRecord).where(PersonalRecord.user_id == user_id))
+        db.commit()
+
+        acts = db.execute(
+            select(Activity).where(
+                Activity.user_id == user_id,
+                Activity.distance_m.isnot(None),
+            )
+        ).scalars().all()
+
+        best = {}  # (sport, dist_m) -> {dur, aid, at, label}
+        for a in acts:
+            if a.sport_type == "running" and a.polyline is None:
+                continue  # indoor/treadmill — unreliable distance
+            rows = db.execute(
+                select(ActivityDataPoint.timestamp, ActivityDataPoint.distance_m)
+                .where(ActivityDataPoint.activity_id == a.id)
+                .order_by(ActivityDataPoint.timestamp)
+            ).all()
+            dp = [{"timestamp": r[0], "distance_m": r[1]} for r in rows if r[1] is not None]
+            if len(dp) < 2:
+                continue
+            total = a.distance_m or 0
+            gps_max = max(p["distance_m"] for p in dp)
+            if total > 0 and gps_max > 0 and abs(gps_max - total) / total > 0.02:
+                f = total / gps_max
+                for p in dp:
+                    p["distance_m"] *= f
+            for label, dur in compute_best_splits(dp, total).items():
+                dist_m = dist_for_label.get(label)
+                if dist_m is None:
+                    continue
+                key = (a.sport_type, dist_m)
+                if key not in best or dur < best[key]["dur"]:
+                    best[key] = {"dur": dur, "aid": a.id, "at": a.start_time, "label": label}
+
+        for (sport, dist_m), b in best.items():
+            db.add(PersonalRecord(
+                user_id=user_id,
+                activity_id=b["aid"],
+                sport_type=sport,
+                distance_m=dist_m,
+                distance_label=b["label"],
+                duration_s=b["dur"],
+                achieved_at=b["at"],
+                is_current_record=True,
+            ))
+        db.commit()
+
+    return {"status": "ok", "records": len(best)}