Daily Report — 2026-08-12

Sleep & CPAP Report — Night of 2026-08-12

Sources: health-report-2026-08-12-to-2026-08-13.json (Apple Watch sleep stages, exported 06:17 on 08-13) + OSCAR_Greg_by_Day_2026-08-12.csv (ResMed AirSense/AirFlow 11 via OSCAR). Session start (23:05 on 08-12) is on the same calendar date as the CPAP Period, so the match is direct. CPAP usage (6.72 h) is close to Apple Watch time in bed (6.51 h) — well-aligned tonight (~13-minute gap).

⚠️ Flags

Obstructive apnea count elevated for the second consecutive night — 13 OAs tonight, following 12 last night. This is the first back-to-back run of OA ≥ 12 in the dataset. In the last 20 matched nights, 7 have logged OA ≥ 10 (07-24, 07-25, 07-30, 08-04, 08-06, 08-11, 08-12) — and the two most recent are consecutive. AHI 2.68 is up modestly from last night's 2.57 and is 32% above the 7-day average (2.03) and 41% above the 30-day (1.90). Central apneas returned tonight (2 CA) after two zero-CA nights. Pattern is now approaching the "3+ consecutive nights elevated vs. own baseline" threshold called out in the task spec — one more OA-heavy night tomorrow would formalize it.
Awake time spiked to 28.1 min — 4th-highest of 51 matched nights (only 08-01: 66.7 min, 08-04: 35.6 min, and 07-25: 30.6 min are higher). Well above both trailing averages (7-day 18.2, 30-day 18.0) and up sharply from last night's 21.1 min. Sleep efficiency dropped to 92.8% — 5th-lowest of 51 nights. This is a real fragmentation night, not a rounding-error miss.
Auto-PAP pressure ran hot: Pressure avg 7.99 cmH₂O, the highest since 08-05 (7.97) and rank 23/26 in the current-format-CSV era. The machine appears to be titrating up in response to the recurring obstructive events; 90th-percentile pressure (10.0) held steady. Worth watching whether this stabilizes at a new baseline or continues climbing.
Deep sleep jumped to 20.78% — 3rd-highest of 51 matched nights (behind 07-21's 28.44% and 07-23's 25.47%). A striking positive reversal from the recent soft downward drift (13.6%, 15.3%, 11.0%, 16.2%, 15.9%, 11.3%, 8.5% over the preceding week). REM took the opposite hit — 19.11%, well below the 30-day average of 24.09%. The 187% jump in deep vs. last night despite the OA/awake-time regression is unusual; the sleep-stage architecture reshuffled substantially.
Leak stays fully resolved — Leak avg 0.28 L/min, 90th 1.2. This is now 3 consecutive clean nights after the 08-05 to 08-09 elevated stretch. Rules out mask/fit as a driver of the persistent OA pattern. Whatever is driving the obstructive events across 08-11 and 08-12, it is not leak — the correlation would be inverse in the mask-fit hypothesis.

1. Summary Table — Night of 2026-08-12

Metric	Value
Bedtime (session start)	23:05 on 08-12
Time in bed	6.51 h (390.4 min)
Total sleep duration	6.04 h (362.3 min)
Sleep efficiency (TST/TIB)	92.8% — rank 5/51 from low (well below median 96.1%; awake-time drag)
Awake time	28.1 min — 4th-highest of 51 matched nights; +56% vs. 7-day (18.2), +56% vs. 30-day (18.0)
Core sleep	3.63 h (217.8 min) — 60.11%
Deep sleep	1.25 h (75.3 min) — 20.78% — rank 3/51, +59% vs. 7-day (13.11%), +42% vs. 30-day (14.67%)
REM sleep	1.15 h (69.2 min) — 19.11% — below 7-day (22.82%) and 30-day (24.09%)
CPAP usage	6.72 h — matches dataset median (6.72), in line with 30-day (6.66)
AHI	2.68 — +32% vs. 7-day (2.03), +41% vs. 30-day (1.90); 4th-consecutive night ≥ 2.34 barring 08-10
RDI	2.83
OA / H / CA / UA / RERA	13 / 3 / 2 / 0 / 1 — 2nd-consecutive OA ≥ 12 night; CAs return after 2 zero-CA nights
Pressure avg	7.99 cmH₂O — highest in 8 nights; rank 23/26 in current CSV era
Pressure 90th %ile	10.0 cmH₂O — flat vs. last night
Leak avg	0.28 L/min — 3rd consecutive clean night; low-leak bucket
Leak 90th %ile	1.2 L/min — normal band
Compliance flag	1 (compliant)

2. 8-Night Trend & Leak Context

Date	AHI	OA	H	CA	RERA	Leak avg	Deep %	Awake (min)	Efficiency
08-05	1.49	7	2	1	0	1.90	11.3%	10.5	97.3%
08-06	2.83	11	4	2	1	0.56	15.9%	9.5	97.3%
08-08	2.34	6	10	0	0	0.87	16.2%	15.6	96.1%
08-09	1.34	6	3	0	2	3.00	11.0%	15.1	96.2%
08-10	0.58	3	0	0	0	0.17	15.3%	20.1	94.7%
08-11	2.57	12	4	0	1 UA	0.42	13.6%	21.1	94.6%
08-12	2.68	13	3	2	1	0.28	20.8%	28.1	92.8%

Two-night composite (08-11 + 08-12): 25 OA + 7 H + 2 CA + 1 RERA in 13.3 hours of use with leak fully resolved on both nights.

Leak vs. sleep architecture — 51 matched nights (full history through 08-12)

Buckets on nightly leak average:

Low-leak (≤ 0.30 L/min): n = 17 — Mean deep: 18.07% | Mean awake: 15.9 min
Higher-leak (> 0.30 L/min): n = 34 — Mean deep: 14.76% | Mean awake: 18.6 min
Very-high-leak (≥ 1.0 L/min): n = 14 — Mean deep: 12.96% | Mean awake: 16.7 min

the user r (leak avg vs. deep %): −0.28 (n=51). the user r (leak 90th vs. deep %): −0.32. the user r (leak avg vs. awake min): −0.04. the user r (AHI vs. deep %): +0.06. the user r (AHI vs. awake min): +0.03.

Interpretation: Tonight fits the low-leak bucket's deep-sleep signature almost perfectly — the low-leak group averages 18.07% deep, and tonight came in at 20.78%. That's exactly what the leak/deep relationship predicts. But it also underscores that tonight's problem is not the CPAP mask: leak is optimal, deep sleep is elite, and yet AHI and awake time both regressed. The r-values reinforce this — AHI's night-to-night moves are essentially uncorrelated with deep-sleep % or awake time in the historical data, so tonight's fragmentation (28.1 awake min) is not being driven mechanically by the 13 OAs. The reshuffling between REM (down to 19.11%) and deep (up to 20.78%) at similar total-sleep hours is the more interesting architectural story — deep displaced REM tonight, possibly related to sleep-onset timing or arousal thresholds.

3. Comparison to Trailing Averages

Metric	Tonight (08-12)	7-day avg (08-04–08-11)	30-day avg (through 08-11)
AHI	2.68	2.03	1.90
Leak avg (L/min)	0.28	1.05	0.74
Leak 90th (L/min)	1.2	3.60	2.16
Pressure avg (cmH₂O)	7.99	7.51	7.71
Usage hours	6.72	6.42	6.66
Deep sleep %	20.78%	13.11%	14.67%
REM sleep %	19.11%	22.82%	24.09%
Awake time (min)	28.1	18.2	18.0

Split-personality night: leak, deep-sleep %, and pressure are all trending the right way (leak resolved, deep elite, pressure holding auto-band); AHI, OA count, REM %, and awake time are all trending the wrong way. The dominant miss is awake time (+56% vs. both trailing averages) — that's the metric to explain, not AHI. AHI's ~30–40% miss is entirely driven by the OA count (13 vs. 7-day mean of ~6.7), while hypopneas (3) sit below the recent 7-day mean.

Weekly AHI/Leak/Pressure trend

Week	AHI avg	Leak avg	Pressure avg	Deep %	Awake (min)	n
07-15 to 07-21	1.39	0.49	8.05	16.15%	16.4	5
07-22 to 07-28	1.68	0.87	7.41	16.30%	17.4	7
07-29 to 08-04	1.74	0.70	7.03	12.60%	24.7	7
08-05 to 08-11	1.86	1.15	7.53	13.87%	15.3	6
08-12 to 08-18 (partial, n=1)	2.68	0.28	7.99	20.78%	28.1	1

Prior weekly AHI has ratcheted up modestly for four consecutive weeks (1.39 → 1.68 → 1.74 → 1.86). Tonight's opening entry for the new week extends the direction. Leak has moved sharply back to the low range after last week's spike; pressure is climbing again toward the mid-July 8.05 baseline.

4. Outlook

The story shifts from "leak resolution" to "recurring OA cluster." With leak now cleanly resolved for a third night and central apneas mostly quiet, the OA count has emerged as the persistent signal — 08-11 (12) and 08-12 (13) are the first back-to-back OA ≥ 12 nights in the 51-night dataset, and they follow prior isolated OA spikes on 08-04 (14) and 08-06 (11). If tomorrow (08-13) logs OA ≥ 10 again, that formally triggers the "3+ consecutive nights elevated vs. baseline" flag in the task spec and would argue for one of: (1) positional shift (more supine sleep), (2) upper-airway congestion or allergic irritation, (3) alcohol/dinner-timing effect, or (4) the current pressure floor being too low for the recent nights' obstructive events (auto-PAP is already responding — Pressure avg 7.99 tonight — but may not be climbing fast enough).

Fragmentation is the second signal. Awake time of 28.1 min ties tonight to the small cluster of high-arousal nights (08-01, 08-04, 07-25) and drops efficiency to a bottom-decile 92.8%. This is not obviously event-driven (only 18 respiratory events in 6.72 h ≈ 1 every 22 min — not enough to explain 28 min of wake), so environmental or behavioral drivers (bedroom conditions, sleep-onset stress, hydration) are more likely.

Deep sleep is the counter-narrative. 20.78% is genuinely great, and it's exactly what the low-leak bucket predicts. If the OA pattern resolves without pressure change, tonight's architecture profile is what a "good" night looks like — deep and total sleep both healthy, just with the fragmentation and event count as the outstanding issues.

5. Caveats

CPAP usage (6.72 h) vs. Apple Watch TIB (6.51 h): matched within ~13 minutes tonight. No data-alignment concerns — the AHI, leak, and event counts reflect the full sleep window.
Matched-night count computed on this run: n=51. This is lower than the n=55 cited in the 08-11 report; the discrepancy reflects the current file set (52 unique Apple Health sleep dates + CPAP overlap = 51 matched), not a data-loss event. Historical files may have been consolidated or moved; historical values are recomputed from the currently-present files on every run, so figures can drift slightly vs. prior reports. All conclusions in this report are self-consistent within the current dataset.
Unmatched CPAP-only nights (no matching Apple Watch sleep session): 2026-06-15, 06-16, 06-18, 06-20, 07-03, 07-17, and 08-07. Excluded from matched-night comparisons. 06-15 and 06-16 also have Pressure_Avg = 0 (early-use unreliable readings).
Trailing averages are computed from the 7 and 30 most-recent matched nights before tonight (08-04 through 08-11 for the 7-day; 07-16 through 08-11, n=25 available for the 30-day window). The 30-day window is size-limited by dataset depth.
No SpO2 data in tonight's CPAP CSV export (columns present but blank) — consistent with every export in the current format.
Leak 90th/95th column naming: current-format exports (07-16 onward) report a 90th-percentile leak column; earlier exports (through ~07-15) used a 95th-percentile column. Both treated as the same "tail leak" metric for trend purposes.
Correlation coefficients and weekly averages are recomputed from the full historical file set on each run; small drifts vs. figures cited in earlier reports can occur as the dataset changes — not data errors.

See all past reports →