diff --git a/Sleep/2026-09-04.md b/Sleep/2026-09-04.md
new file mode 100644
index 0000000..873adee
--- /dev/null
+++ b/Sleep/2026-09-04.md
@@ -0,0 +1,45 @@
+---
+date: 2026-09-04
+type: sleep
+tags: [sleep, sober, short-night, late-bedtime, sleep-onset]
+---
+
+# Sleep — 2026-09-04
+
+**5h 45min** · 02:11→07:57 · REM: 15% · Deep: 30% · HRV: **111** · Resting HR: 52 · Woke: 1× · Score: 80/100
+
+> "Got home just around midnight so I can get some sleep for my morning run, but I could not get myself to sleep, so I slept less. Eventually I fell asleep. In the morning I let myself sleep in." Sober night, home ~00:00 — but sleep onset dragged until 02:11 (watch bed time). The sleep-in to 07:57 was real, yet the wake-up still came too early to recover the lost hours: 5h 45min total, below the 6h reference.
+
+**Interpretation:** A short-night story where the only debt is **duration** — the architecture of the sleep he did get was healthy: deep 30% (1h 42min) and REM 15% (52min) both normal, woke only once, continuity 80. Score 80, "good," with the watch's verdict landing on sleep deprivation (5h45 < 6h). **HRV 111ms sits exactly in his sober-late-bedtime band** (Aug 29: bed 02:14 → 106ms; tonight: bed 02:11 → 111ms) — the clean no-alcohol signal, and a clear demonstration that the bedtime lever is the story: the 145ms all-time high (Sep 2) came with a 23:50 bedtime, while ~02:00 bedtimes land ~106-111ms regardless of sobriety. Resting HR 52 (vs 48 on the record night) is the short-sleep tax, mild. The one irregularity worth naming: **sleep-onset difficulty despite being home by midnight** — excitement-driven racing thoughts per the established pattern, not alcohol.
+
+**Prediction for the day:** this is the night before the Friday 3 km easy (Sep 4, 48h before the Sunday 6 km long run). Expect the short sleep to show as elevated effort if the run lands in the midday heat — keep it easy, slow first km, walking allowed. Tonight's priority is protecting a normal bedtime; the pre-midnight lever (23:50 → 145ms on Sep 2) is the proven fix.
+
+## Details
+
+| Metric | Value | Reference |
+|--------|-------|-----------|
+| Night sleep | **5h 45min** | 6-10h (Below average) |
+| Bed → Rise | 02:11 → 07:57 | — |
+| Deep sleep | 30% (1h 42min) | 20-60% (Normal) |
+| Light sleep | 55% (3h 11min) | <55% (High, borderline) |
+| REM sleep | 15% (52 min) | 10-30% (Normal) |
+| Sleep continuity | 80 points | 70-100 (Normal) |
+| Times woke | 1 time | 0-1 (Normal) |
+| Breathing quality | 98 points | 70-100 (Normal) |
+| HRV | **111ms** | 78-137ms (Within range) |
+| Resting HR | 52 bpm | 48-58 bpm (Within range) |
+| SpO2 | 98% | 96-99% (Within range) |
+| Respiratory rate | 15 brpm | 12-16 brpm (Within range) |
+| Score | 80 / 100 | Better than 71% |
+
+### HRV progression
+
+| Date | HRV | Context |
+|------|-----|---------|
+| Aug 29 | 106ms | Sick week, late bedtime (02:14) |
+| Sep 2 | **145ms** 🚀 | Sober, bed before midnight (23:50) — all-time high |
+| **Sep 4** | **111ms** | Sober, home ~00:00 but sleep onset 02:11, 5h45 |
+
+## Connected
+
+- Runs/2026-09-04 (planned Friday 3 km easy — not yet run)