diff --git a/Sleep/2026-08-12.md b/Sleep/2026-08-12.md
index 4c09c95..c4e7620 100644
--- a/Sleep/2026-08-12.md
+++ b/Sleep/2026-08-12.md
@@ -12,7 +12,9 @@ tags: [sleep, recovery, sober, late-bedtime, low-rem]
 >
 > The three-night trend is the headline: **139 → 118 → 112ms.** Not alarming — 112 is a good number — but it's a gentle downward drift that lines up with a third consecutive ~midnight+ bedtime (00:24 → 00:07 → 00:05). The watch flagged it again: "you slept at 00:05, which is a little late," with the endocrine/nervous-system warning and the vitamin A/C + protein suggestion. The body is recovering well, but the timing keeps creeping late.
 >
-> The real anomaly is sleep architecture: **REM collapsed to 5% (24 min)** — the lowest in recent logs (previous sober nights: 11% on Aug 11, 11% on Jul 29, 11% on Jul 27) — while light sleep expanded to 66% (High, ref <55%). Deep sleep held at a normal 29%. REM that low on a clean, alcohol-free night is out of pattern. Most likely drivers: the late bedtime shifting the sleep window later (circadian misalignment hits REM first), possibly with some pre-sleep mental activation before bed. Not a red flag by itself — but worth a check-in: if the late bedtime pattern holds, REM stays the first casualty.
+> The real anomaly is sleep architecture: **REM collapsed to 5% (24 min)** — the lowest in recent logs (previous sober nights: 11% on Aug 11, 11% on Jul 29, 11% on Jul 27) — while light sleep expanded to 66% (High, ref <55%). Deep sleep held at a normal 29%. REM that low on a clean, alcohol-free night is out of pattern.
+>
+> **Post-hoc context (same morning, self-reported):** the 7:30 wake was attempted but failed — no run today meant no drive to get up. Gutca snoozed 7:30 → ~8:30, falling straight into dreams each time, then got up around 08:31-08:45. This reframes the architecture: the watch's recorded "sleep" (→ rise 08:31) **includes the entire snooze hour** — the 1h of dream-heavy snoozing was scored mostly as light sleep, which inflates the 66% light figure and likely hides some of the missing REM (watch REM detection is conservative during restless snooze). The dreams themselves are the brain grabbing the REM the architecture says it was missing. So: the light/REM split is partly a measurement artifact of the snooze loop — the underlying signal is late bedtime + a no-goal morning, not a biological REM problem.
 
 ## Details