diff --git a/BUGS.md b/BUGS.md
index 8c8a1f6..6950b54 100644
--- a/BUGS.md
+++ b/BUGS.md
@@ -24,6 +24,23 @@ When CLOSED, move the entire entry to Resolved — don't duplicate.
   fills the cards immediately on load (and every 15 s) via REST. Needs in-car
   verification.
 
+### BUG-21 — OBD speed drops out mid-drive, replay shows 0 km/h (GPS-only)
+- **Status**: `OPEN` 2026-08-14
+- **Phase**: 29
+- **Symptom**: in the portal trip replay, a ~20-minute stretch of the 2026-08-13 evening
+  drive shows "GPS: 0.0 km/h" even though the car was moving (GPS positions advance).
+- **Cause**: the Pi's OBD adapter (ELM327) stopped returning speed mid-drive — the last OBD
+  speed reading in trip #9 was 22:23:50 UTC and it didn't return until 22:43:41 UTC (trip #10),
+  a ~20-minute gap. During that gap `_log_loop()` logged rows with GPS coords but NULL speed
+  (`insert_tracking(gps_x, gps_y, speed=None, …)`) — there is no GPS-derived speed fallback, so
+  the SPEED column stays NULL and the replay shows 0.0 km/h (`speed_source="gps"`).
+- **Data**: trip #9 has 723 points — 570 with OBD speed, 25 with speed=0, 128 with NULL speed.
+- **Fix (proposed)**:
+  1. GPS-speed fallback in `_log_loop()` — when OBD speed is None/0 but GPS coords exist,
+     compute speed from the previous position (haversine / elapsed time) and log that as SPEED,
+     so the replay shows the real speed (sourced "gps").
+  2. Investigate why the OBD adapter didn't recover for ~20 min (reconnect on stalled reads).
+
 ---
 
 ## Resolved
diff --git a/PLAN.md b/PLAN.md
index f47a7c9..a7564af 100644
--- a/PLAN.md
+++ b/PLAN.md
@@ -2282,6 +2282,10 @@ logged as `BUG-XX` entries in [BUGS.md](./BUGS.md) (BUG-16 onward) and fixed + v
 - **BUG-20 (server trip IDs inflate) — ✅ FIXED**: `trips.id BIGSERIAL` burned sequence values
   on every upsert (recompute after each upload batch). Fixed: sequence clamped to `MAX(id)`
   after each recompute + one-time renumber migration — trips now 1..10, sequence at 10.
+- **BUG-21 (OBD speed dropout mid-drive) — OPEN**: ELM327 stopped returning speed for ~20 min
+  on 2026-08-13 (22:23:50→22:43:41 UTC); Pi logged NULL speed (no GPS-speed fallback), so the
+  replay shows 0 km/h for that stretch. Proposed: GPS-speed fallback in `_log_loop()` + OBD
+  reconnect on stalled reads.
 
 Open items from the drive — fill in as reported:
 - (none clock-related — BUG-17/BUG-18 closed 2026-08-14).