Sleep score
Role: Benchmark · Current decision: Proposed milestones
In this measurement
What it represents
Oura’s Sleep Score is a composite of total sleep, efficiency, restfulness, REM, deep sleep, latency and timing. Contributors and the composite are graded on a 0–100 scale; 85–100 is the provider’s Optimal category. They are not percentages of a physiological process. Oura score, contributors.
The same night can receive different grades from different devices because both the underlying estimates and the scoring rules differ. Oura gave the first recorded night 75; COROS gave its selected episode 93. An 18-point difference is not an 18% difference in actual restoration.
The proposed target
The target specification proposes a 14-day average ≥85, then a 28-day average ≥90 with a consistency and coverage requirement. These are explicit project challenges. Their windows and thresholds are choices; no trial establishes that moving from 89 to 90 causes a discrete health improvement.
A score increase can be pursued and celebrated as a score increase. To establish a broader benefit, examine what changed and whether an appropriate outcome changed as well. Examples include obtained sleep, unwanted disruption, a validated vigilance task and reported functioning. A high score with worse daytime performance is a disagreement to investigate, not a reason to erase the latter observation.
What the first two nights suggest
Oura’s timing contributor was 12, total sleep 71, REM 81, restfulness 93, deep sleep 95 and efficiency 98; latency was 78. These locate weaker grading dimensions. They do not tell us which intervention will improve a meaningful outcome.
Timing needs two separate investigations: what Oura penalizes and whether the current sleep episode is suitably aligned with internal phase and daily demands. Deep sleep already received a high contributor score. There is no basis in this night alone for treating extra estimated deep minutes as the dominant opportunity.
The second night’s composite rose to 84. Total-sleep grading rose from 71 to 91 and timing from 12 to 29, while restfulness fell from 93 to 79 and efficiency grading from 98 to 90. These component changes describe a higher composite alongside worse grading of continuity. The account’s cold awakenings remain relevant even though COROS awarded 95. Second record.
This is why the project retains the score, its components and personal experience. The next decision concerns a comfortable routine and unwanted waking, with score improvement as a supporting result. A causal or sustained benefit has not yet been demonstrated.
Measurement and interpretation
The October 28 record returns Oura 91 / COROS 94. Oura’s composite stays 91 despite efficiency 97 → 100, latency 78 → 94, restfulness 91 → 92, timing 64 → 48, deep 100 → 95 and total sleep 100 → 97. REM is 91. The 100-point efficiency contributor differs from the 95% efficiency estimate. The follow-up reports an okay night, good rest and morning, possible few unremarkable awakenings and suboptimal pre-bed food/lights after a social day. These outcomes stay alongside the grades. The roughly 00:30 attempt also precedes Oura’s 00:44 main start; its 13-minute latency is not a complete measure of that reported attempt-to-sleep interval. The fixed October 13–26 assessment stays 82.31.
The October 27 record returns Oura 91 / COROS 90. Oura total-sleep/deep contributors are 100, REM 87, efficiency 97, restfulness 91, latency 78 and timing 64. Oura rises 85 → 91 while COROS falls 94 → 90; their grades remain separate. The follow-up reports feeling good during the night and in the morning, and almost all body pains gone, contrasting with October 26’s explicitly bad experienced night. The reported 23:10 attempt and initial inability to fall asleep also show why a 78-point latency grade and 7m30s main-episode latency do not describe the full experience of trying to sleep. The first October 13–26 assessment remains fixed at 82.31; the new date does not revise it or complete the 28-day window.
The October 26 main-night record returns Oura 85/COROS 94, superseding the earlier nap-only COROS response with invalid −1 grade. Oura contributors are total sleep 92, deep 98, REM 83, efficiency 98, restfulness 82, latency 86 and timing 36. Daily scores remain provider-native; neither extra detected episode is added as another nightly score. The follow-up reports bad sleep, waking sleepy, ongoing body pains and feeling unhappy overall despite the higher grades. It also reports no naps, disputing COROS’s detection. This is a clear score–experience mismatch; no period of recovery or good later-day function is inferred. The first fixed October 13–26 assessment has 13 eligible Oura scores, sum 1070 and observed mean 82.31. Coverage 13/14 meets the minimum 12, but the mean is below 85, so the milestone is not achieved. The battery-truncated October 22 remains excluded with its date visible; all other eligible dates, including difficult nights, remain included.
The earlier October 26 retrieval added Oura 86 on October 24 and 69 on October 25. October 24’s higher grade coexists with 78 minutes estimated awake; October 25 has timing 15, restfulness 67 and REM 60, while efficiency and latency contributors are88 and 91. Its high component grades do not erase lower grading elsewhere. At that earlier check, thirteen dated entries contained twelve eligible Oura scores; the battery-truncated October 22 remained excluded and no October 26 Oura response had yet been returned. Coverage alone does not complete a calendar window or award a sustained milestone. COROS was added after access became available: 94 on October 24 and 65 on October 25. Both providers grade the second night lower, while their exact duration, stage and waking estimates remain separate. The earlier COROS October 26 nap-only response carried an invalid −1 score; that response was not a main-night grade or a milestone observation. The complete main-night response is assessed above.
The October 23 record returns Oura 83 / COROS 88, with both episodes ending around 09:28. The diary reports a cleaner-triggered awakening around 09:30–09:40. Oura contributors are total sleep 81, deep 96, REM 88, efficiency 97, restfulness 85, latency 83 and timing 52. Today’s Oura score is eligible for whole-night milestones; yesterday’s incomplete 73 remains excluded rather than serving as the comparison for an apparent improvement. Eleven dated entries contain ten eligible Oura scores, insufficient to assess the established sustained windows.
The October 22 battery interruption illustrates a coverage failure: Oura stops at 06:30 while COROS continues to 10:07 and the diary recalls waking after 10:00. Oura returns 73, total-sleep 50, REM 36, timing 100 and restfulness 97, all retained as partial-recording outputs. The whole-night milestone excludes this score with an explicit reason; its high component grades are also ineligible as full-night achievements. No missing hours are imputed or borrowed from COROS. COROS’s 91 remains its separate series beside the middling subjective impression. Ten dates contain nine eligible Oura whole-night scores, with the battery-loss date visible as a coverage gap.
The October 21 record returned Oura 90 / COROS 91. Oura contributors were total sleep 99, deep 98, REM 88, efficiency 99, restfulness 85, latency 86 and timing 56. Oura’s score fell 92 → 90 and restfulness 88 → 85 while the diary reports fewer awakenings; COROS rose 89 → 91. Both estimated longer sleep and slightly less waking. This keeps improvement in remembered continuity separate from improvement in a provider grade. Nine recorded dates do not meet the existing sustained assessment windows.
The third night reversed part of the tradeoff: Oura’s score fell to 71 while restfulness rose to 92 and timing to 78. Total-sleep grading was 51 and REM 30; COROS scored 80. The main episode was shorter, although I felt much better on waking. A score decline does not erase that positive experience, and a positive experience does not establish sufficient duration. October 15 record.
On 16 October, Oura returned 90 after synchronization, with 8h37m30s estimated sleep. Total-sleep grading rose to 93 and REM to 97; efficiency and restfulness were 98 and 93. Not every contributor improved: latency grading was 67 and timing 72. COROS became available later and returned 98, versus 80 the previous night. Its eight-point gap from Oura is a difference between grading systems, not a measured difference in restoration. This is one encouraging grade, not achievement of either sustained milestone. The later sleepiness now reported for October 15 also keeps the full following day relevant to interpreting the next score.
The October 17 social exception returned Oura 64 and COROS 82 despite nearly identical estimated sleep duration, 6h22 and 6h21. Oura’s contributor grades were timing 18, total sleep 62, deep sleep 59, REM 70, restfulness 73 and latency 72, while efficiency remained 98 with an estimated 93% asleep. The high efficiency grade does not erase the shorter sleep or the report of an uncomfortable floor and repeated brief awakenings. The night remains in headline milestone windows rather than being removed because its conditions were exceptional.
The October 18 night returned Oura 90 and COROS 98 after much longer estimated sleep. Oura’s total-sleep and deep-sleep contributors reached 100, REM 97, efficiency 98, restfulness 84, latency 83 and timing 46. The latency estimate was 9m30s and deep sleep 2h33m30s. This is the first recorded 100 for deep sleep, demonstrating that the contributor ceiling can be reached without establishing an optimal biological deep-sleep quota. Timing is today’s weakest graded contributor; unknown composite weights prevent converting a possible contributor improvement into a guaranteed total-score gain.
The 90/98 pair matches October 16, although component grades and duration differ. The record now has two Oura 90 nights among six recorded dates, including the 64-point exception. This is progress toward the benchmark, with the sustained 14- and 28-day results still unassessed. Brief remembered awakenings and the follow-up’s report of feeling awake on waking remain alongside the high scores. October 18 later-day function is still unreported. The follow-up also describes feeling good during October 17 despite its lower grades.
On October 19, both providers returned 91, although their selected episode boundaries and waking totals still differ. Oura graded total sleep 100, deep 98, REM 98, efficiency 98, restfulness 88, latency 83 and timing 51. The latency estimate was 20m30s, versus 9m30s the preceding night, while its contributor stayed 83; a grade is not a direct conversion from minutes. Oura’s overall score rose 90 → 91 while COROS’s fell 98 → 91. Agreement on one headline grade does not make the grading systems interchangeable.
Today’s retrieval also reports Oura firmware 2.1.36, previously 2.1.25, with sleep algorithm v2 unchanged. Exact installation timing and any scoring effect are unknown. Seven recorded dates, including the social exception, remain in the sequence; neither sustained milestone has been assessed as achieved. The follow-up adds an approximately 00:10 sleep attempt, onset taking a while, uncertain final waking around 10:30 and initial sleepiness resolving to normal speed within 10–15 minutes. This subjective morning experience stays alongside the 91-point scores; later-day function remains unreported.
The October 20 record returned Oura 92 and COROS 89. Oura’s total-sleep and deep-sleep contributors were 97, REM 96, efficiency 99, restfulness 88, latency 89 and timing 66. Timing rose51 → 66, latency83 → 89 and efficiency 98 → 99; restfulness stayed 88, while total sleep fell 100 → 97. Oura estimated 8h55m30s asleep with 11 minutes latency and 94% efficiency. The overall Oura score rose 91 → 92 while COROS fell 91 → 89.
The diary still describes okay-ish sleep with many brief awakenings, so the highest recorded Oura grade is a benchmark result alongside a less-positive experienced night. Eight dates include a recent 90, 91, 92 Oura sequence, while the sustained 14/28-day criteria remain unassessed. Firmware 2.1.36 and sleep algorithm v2 match the preceding retrieval. The follow-up adds strong sleepiness on waking around 09:30, an unsuccessful 5–10-minute attempt to sleep more, no bathroom trips after awakenings and estimated under-30-second wake episodes. Dream recall and a suspected dream trigger do not validate exact REM or establish causation. A second follow-up reports feeling normal within 10–15 minutes, while sustained later-day experience remains unreported. The user finds COROS closer to the timing of awake periods but Oura probably closer to total awake time; this subjective comparison is not independent validation or a reason to choose one device as correct for every metric.
Sleep staging validation does not validate every composite-score threshold. A score is downstream of several estimates and an algorithmic judgment. Firmware or scoring changes can alter it without a lifestyle change. Keep hardware and algorithm information when exposed and mark known changes.
Use one named primary score series. Keep COROS as its own series and retain disagreement. Do not select the device with the better grade on each date, average the two scores into a purported biological measure, or merge Sleep Score with Readiness.
The useful record contains the score, all contributors, coverage, context and functional observations. It preserves difficult nights and missingness, so a sustained milestone means what the assessment rule says it means.