Learning from personal experiments

Updated 2 Nov 2026

A convincing result is a useful effect that survives repetition under described conditions. This page connects the targets and strategy queue to a decision: what to change, how to test it and whether to keep it. The first entry is a baseline observation; no personal intervention has yet been validated.

The current routine proposal

November 2 — poor experienced sleep, with different score directions. Oura 83 / COROS 88 estimate 8h11m / 8h37m asleep, ending at 09:44 PST with starts three minutes apart. Waking estimates differ 30 / 7 minutes. Oura timing and REM contributors fall to 50/74; both HRV summaries are 44 ms. The curves add the new native sequence. The follow-up reports attempting sleep around 01:00, waking around 09:45 and not sleeping well. Those approximate boundaries fit both devices, but their awake totals remain different. Oura falls 96 → 83 while COROS rises 87 → 88. Exact onset, recalled awakenings, morning/later-day function and current food, light, walking, social and thermal conditions remain unknown. November 1’s good experience and proposed ranking of optimization priorities are not assumed repeated or refuted by today’s readings. The further follow-up reports many thoughts before bed and identifies timing as the main suspected contributor. This is the user’s working explanation; Oura’s timing grade is not a measured causal share of poor sleep.

November 1 — a good night with several changes together. Oura 96 / COROS 87 estimate 9h59m30s / 10h26m asleep. Oura timing reaches 96, total sleep 100, but restfulness falls to 83 and awake time grows to 52 minutes. The curve preserves elapsed time through the clock rollback. COROS’s returned clock window has a one-hour offset ambiguity; the diary keeps its native duration and labels separate. The follow-up supplies bed around 23:00–23:30, feeling kind of falling asleep from 22:30 and waking around 09:30, close to Oura’s 09:26:53 ending. This does not confirm continuous sleep from 22:30 or resolve COROS’s clock ambiguity. The latest clarification separates the conditions: the short sleeping surface and restricted turning were uncomfortable; the pillow was better, the blanket comparable, and temperature much better, perfect and stable. Background noise felt helpful as steady background sound. The user still views pre-bed light and late food as drawbacks, but suspects they are not the highest-impact priorities. That is a working hypothesis about relative importance, not an effect-size ranking established by this night. Main meal 19:30–20:00, small desserts 20:00/21:30, about 60 minutes walking after food and a social evening are also reported. Main-meal-to-bed and last-dessert-to-bed intervals were roughly 3–4 and 1.5–2 hours. Several factors changed together; this night identifies neither a cause nor an optimal food cutoff. Morning/later-day alertness, fluids and cold awakenings remain unreported. The recurring cold-feet account below remains earlier context.

A condition is not an on/off switch. In a within-person laboratory study of 55 adults aged 18–30, evening light caused widely varying melatonin responses across individuals during five-hour exposures. The outcome was melatonin suppression, not proof that every bright-room exposure prevents good sleep. This night does not measure personal light sensitivity or isolate the effects of noise and lighting. Phillips and colleagues, 2019.

October 31 — reported timing supplied, provider grading still differs. The user reports last food 19:00, onset around midnight, waking 09:15–09:30 and no naps. That gives roughly five hours from last food to reported onset; actual attempt, fluid timing and later-day outcome remain unknown. Overall sleep and the morning are now reported good, with time lying in bed thinking. Oura 93/COROS 83 estimate 8h57m30s/8h47m asleep. COROS’s09:16 ending fits the reported range, while Oura ends 09:42. The curve retains the original Oura main sequence; its separate two-minute evening detection is disputed and kept outside the main night. Earlier detected starts and more estimated sleep accompany opposite score changes. No causal food or bedtime effect is isolated.

Recurring cold feet and excess starting warmth

The user identifies the problem as the room cooling overnight. One blanket is comfortable at bedtime but insufficient by morning, when the feet become cold around 07:30 while still covered. Starting with two blankets is too hot: it can prevent falling asleep or lead to removing one early, leaving the same insufficient warmth once the room cools. The room’s temperature drop is the user’s observation; its numerical size and timing have not been measured. This recurring account spans several unspecified nights. Overall sleep and the morning were reported good.

Room-temperature control is unavailable. Thermostat or heating changes are therefore unavailable options. Adding more starting bedding has already proved counterproductive. The user’s proposed earlier sleep-and-wake schedule is a feasible workaround to test: preserve enough sleep duration while aiming to finish before the usual cold period. It avoids exposure to that part of the night rather than fixing the room’s cooling. No successful schedule intervention is claimed yet.

Track bedtime comfort, cold awakenings and morning functioning after the schedule change. The reported cooling is accepted as the user’s observation; a thermometer trace is optional and would not create control over the room. Thermal-environment evidence.

Earlier routine observations

October 30 — a good experienced outcome, with a reported meal gap. Oura 85 / COROS 89 estimate 8h10m/8h20m asleep with starts 00:45–00:49, about 51–53 minutes later than October 29. Oura timing is 61, total sleep 87, restfulness 81, REM 96 and efficiency 99 (estimated efficiency 94%). HRV is 69/68 ms. The new curves retain labels and pulse without assigning a cause. The follow-up supplies last meal 18:00, sleep attempt 00:30, waking around 09:15–09:30 and feeling good. The meal-to-attempt gap is about 6h30m. Both device endings fit the reported waking interval, but exact onset and individual awake labels remain unverified. Oura’s five-minute latency concerns a main episode beginning 00:45:33, rather than the full interval from the reported attempt. Fluids, remembered awakenings and sustained later-day function remain unspecified. One favorable experience does not establish an optimal food cutoff or isolate a meal effect.

October 29 — revised return, confirmed sleep in the previously missing interval. Earlier bedtime and food finish were planned; actual timings remain unreported. The user confirms being asleep 08:27–09:19. Oura’s recheck now extends through 09:37, score 90, asleep 8h57m30s, awake 45m30s. COROS remains 86 / 9h09m, ending 09:19. Updated contributors include timing 73 (versus 48 the prior night), REM 80, restfulness 90 and efficiency 97 (estimated efficiency 92%). The original shorter response and its 88/100-efficiency values are superseded, with the revision retained in the diary. Curves use the new native data. The confirmed interval establishes missing coverage in the first response, not a battery failure or a causal effect of the intended routine.

October 28 — later main window, less estimated waking, okay night and good morning. Oura 91 / 9h and COROS 94 / 9h16m begin around 00:44–00:47, later than the preceding night. Estimated awake totals fall to 30/10 minutes; Oura timing falls to 48, efficiency contributor reaches 100, and restfulness is 92. HRV is 55/54 ms. Curves retain the actual estimates. The follow-up reports trying to sleep around 00:30, later than desired after meeting people all day, eating and lights before bed, and pulse not as settled as wanted. The night was okay, rest and morning good, with possibly a few unremarkable awakenings. Exact exposures, actual onset, body pains and later-day functioning remain unspecified. Oura’s 13-minute latency concerns its main episode beginning 00:44, not the full delay from the reported attempt. This supports repeating a more favorable feasible routine without isolating any cause or inventing a fixed food, fluid or bedtime cutoff.

October 27 — earlier attempt, good experienced sleep and morning. Oura 91 / 9h39m30s and COROS 90 / 9h50m have main starts around 23:50–23:56, roughly 1½ hours earlier than the previous night. Oura timing rose 36 → 64 and restfulness 82 → 91, while awake estimates remain different (47/15 minutes). HRV is 64/63 ms. Curves and samples show chronological detail without reconciling the devices. The follow-up reports lying in bed and trying to sleep around 23:10, initially being unable to fall asleep, then feeling good during the night and in the morning, with almost all body pains gone. Exact onset, final waking, sustained later-day functioning and the preceding routine remain unspecified. Oura’s 7m30s latency concerns its main episode starting 23:55:58, not the full delay from 23:10. The separate 23:25–23:51 detection is not explicitly confirmed as actual sleep. This is a positive experienced night alongside the measurements, without isolating a cause or establishing complete recovery. No new intervention follows from this night alone.

October 26 main-night follow-up — readings improved, but sleep felt bad. Oura now returns 85 and 8h33m main sleep, COROS 94 and 8h55m. The record keeps daily sleep 9h19m separate from the COROS 24-minute nap detection and retains Oura’s different evening rest episode. The follow-up explicitly reports no naps, disputing COROS’s detection; neither extra window is confirmed as actual sleep. Oura HRV rose 37 → 53 ms and temperature deviation fell +1.34 → +0.01 °C; COROS HRV rose 40 → 51 ms. The follow-up reports ongoing body pains, bad sleep, waking sleepy and feeling unhappy overall. Sleepiness duration, later-day functioning and the evening routine remain unspecified; the October 25 exposures are not carried forward. This is a mismatch between higher grades and experienced sleep, not a reported recovery. No new intervention is adopted.

The first 14-date score window now has 13 eligible Oura values, with mean 82.31 below 85. This supports continuing the comparison while retaining the difficult nights and known battery gap, rather than changing the inclusion rule after seeing the result.

26 October retrieval — October 24–25 added; no new routine inferred. Oura returned 86 / 8h26m30s asleep for October 24 and 69 / 7h19m for October 25. A later successful COROS retrieval adds 94 / 9h06m asleep and 65 / 7h55m, respectively. Its official HRV falls 55 → 40 ms. October 26 has no Oura record yet; COROS has only an unconfirmed detected nap spanning 23:30–00:01, without a valid main-night score. The newest available record includes a later 02:48–11:15 window, higher sleeping pulse and temperature deviation, and lower HRV.

The follow-up explicitly attributes alcohol, sickness, late bedtime and emotions before bed to October 25. These coincided with the measured changes; no single exposure is established as the cause. Experienced sleep quality, morning sleepiness and recovery remain unknown. Earlier bedtime remains the prior direction; a device window alone does not establish a completed change in intended schedule. No food, drink or training intervention is adopted from these measurements.

23 October 2026 — cleaner-triggered awakening after a late gym. Reported last food around 20:00, gym 21:30–23:00, shower around midnight and bed around 01:00. The cleaner woke me around 09:30–09:40. This supplies an external waking trigger, not a count of earlier interruptions. In the follow-up, I confirmed staying awake, feeling sleepy on waking and remaining a little sleepy at the time of reply. The elapsed interval, eventual resolution and effects on work or training remain unspecified. The prior dates’ 10–15-minute recovery is not assumed repeated.

The record retains Oura 83, 7h44m30s asleep, 39 minutes awake and COROS 88, 8h13m asleep, nine minutes awake. Their windows end around 09:28. Oura coverage now extends through the main night, so its score is eligible; October22’s battery-truncated 73 remains separate. The meal-to-bed interval was roughly five hours, gym-to-bed two hours and shower-to-bed one hour, from approximate reports. No isolated effect of those conditions is established. Earlier bedtime remains the preceding stated direction, rather than a completed change or newly fixed clock target.

22 October 2026 — middling subjective sleep; Oura coverage interrupted. I reported bed around 00:40 and final waking sometime after 10:00, perhaps 10:15, without clear recall. I want to go to bed earlier; no new exact clock target was specified. COROS returned 91, 9h15m asleep, eight minutes awake, and 00:44–10:07, broadly consistent with the diary.

The record preserves Oura’s episode ending 06:30 and returned 73 with an incomplete-recording warning after reported low battery. The score and its timing-100 grade cannot establish the full night’s result. Restore coverage before the next comparison; retain today as a measurement failure alongside the middling impression, rather than an observed fall in whole-night Oura sleep. No meal, drink, bathroom, awakening-count or daytime-function details were supplied for this date.

21 October 2026 — fewer remembered awakenings, still several. Bed around 00:10 and final waking around 10:15 were approximate diary times. I felt sleepy initially, then normal within 10–15 minutes. Last food around 19:00–20:00 gives roughly 4h10m–5h10m before bed. No water timing, bathroom trips, exercise, stress or other routine changes were specified for this night.

The record retains Oura 90 / 9h09m30s asleep / 34m58s awake and COROS 91 / 9h33m asleep / six minutes awake. Oura restfulness fell 88 → 85 despite fewer recalled awakenings; its efficiency estimate remained 94%. Remembered waking frequency, total estimated awake duration and the restfulness grade remain separate. The shorter meal interval than some earlier nights is another observation, not a controlled result for food timing. Keep adequate opportunity, comfort and sustained daytime function in the next comparison; no new meal or liquid cutoff is adopted.

20 October 2026 — okay-ish sleep with many brief awakenings; Oura 92, COROS 89. Last food was around 18:30 and bedtime around midnight, a roughly 5½-hour interval using approximate reported times. Lights went down approaching sleep. Water sips continued during the intervening hours, described as not much, without a quantified volume or final time. I remembered many short awakenings to turn or make small adjustments and dreamed a lot. The follow-up reports waking around 09:30 feeling super sleepy, a 5–10-minute unsuccessful attempt to sleep more, then deciding to get up. No awakening led to a bathroom trip. I estimated the short episodes at under 30 seconds and suspected dreams as triggers, without a timed sequence establishing causation. A second follow-up clarifies feeling normal within 10–15 minutes, vivid dream recall, and no awareness of snoring, choking/gasping or observed breathing pauses. Exact out-of-bed time and sustained later-day experience remain unreported. No new stress/exercise report was supplied; earlier-night details are not assumed repeated.

The record retains 8h55m30s asleep on Oura and 9h04m on COROS, with waking estimates 36m45s and 9 minutes and different windows. Oura’s timing grade rose51 → 66, latency83 → 89 and efficiency 98 → 99; restfulness stayed 88. Its 92 is the highest recorded composite, but it does not overrule the okay-ish impression or identify a cause. Dream recall and device REM estimates remain separate observations.

The next useful comparison keeps comfort, short awakenings, strong initial sleepiness and subsequent daytime functioning alongside grades. The second follow-up explicitly reports feeling normal within 10–15 minutes; this is separate from the earlier 5–10-minute attempt to sleep more. I find COROS’s awake-period timing closer to experience and Oura probably closer to total awake duration. Those two judgments stay separate and have not been independently validated. Three recent Oura scores 90, 91 and 92 are progress on that benchmark, while eight dates do not complete either sustained milestone. Firmware 2.1.36 and exposed sleep algorithm v2 match the preceding retrieval. No new meal, water or bedtime rule is adopted from this single night.

19 October 2026 — slept okay; both providers 91. I remembered multiple short awakenings to turn or make small adjustments, with no exact count. Pre-bed activity was stressful and lights went down approaching sleep. Last food was around 20:00; last water was about 500 mL at 22:30–23:00. I explicitly reported no waking to urinate. The follow-up reports an approximately 00:10 sleep attempt, onset taking a while and uncertain final waking around 10:30. Initial sleepiness cleared to the usual subjective pace within 10–15 minutes. Total daily drinks, out-of-bed time and later-day experience remain unreported.

The record retains Oura’s 9h29m asleep and COROS’s 9h27m, but different windows and waking estimates:00:13:36–10:24:59 with 42m23s awake versus 00:25–10:00 with 8 minutes awake. Oura estimated 2h10m30s deep sleep and 2h18m REM. HRV was 69 ms on Oura and 67 ms on COROS; the latter returned Above normal against its own 46–66 ms range. Oura’s observed firmware is now 2.1.36, previously 2.1.25; installation time and any measurement effect are unknown.

The late water without a urinary awakening is a useful comparison with earlier bathroom nights; it does not determine an optimal cutoff or explain the brief position-change awakenings. Stressful activity, sleep opportunity and other conditions have not been held constant. No new fixed bedtime or fluid rule follows from this observation. The follow-up supplies the initial waking experience; later-day functioning remains unreported. The recalled 00:10–10:30 interval is about 10h20m of opportunity, not a measured sleep duration. The water-to-attempt interval was roughly 1h10m–1h40m, and food-to-attempt roughly 4h10m, using approximate diary times. Timing 51 is the weakest Oura contributor, while total sleep is 100; the existing earlier-window comparison remains eligible without changing the sustained milestones.

18 October 2026 — long sleep after a midnight attempt, Oura 90; COROS 98. Last food was around 20:00, last water 21:30, gym 21:00–22:00, shower around 22:30 and a family phone call until around midnight. Small lamp and curtains preceded the shower; lights were off with curtains afterward. Five minutes of breathing preceded sleep, with no pattern supplied in this report. I remembered probably three or four brief awakenings to turn, adjust a pillow or put on the nearby blanket without standing, then quick returns to sleep. I dreamed a lot, and the new blanket felt just right. The conversation was less relaxed, and I suspected a possible influence; no cause is established.

The record retains Oura’s 9h59m30s asleep and COROS’s 10h27, with waking estimates 41m46s versus 7 minutes. Oura graded total sleep and deep sleep 100, REM 97 and timing 46. HRV was 59 ms on Oura and 61 ms on COROS, returning toward the earlier series after the social exception. Neither a remembered awakening count nor a high grade validates the device’s exact waking or stage totals.

This night matters to the next comparison: a roughly four-hour food gap, 2½-hour water gap and two-hour gym-to-bed gap coexisted with high grading and long sleep. The earlier six-hour food interval remains a candidate; it has not been established as a necessary condition. Sleep opportunity, exercise timing, conversation and bedding also differ, and alcohol/caffeine exposure was not supplied. No new fixed cutoff follows from this observation. The follow-up reports feeling awake on waking October 18; later-day experience remains unreported. It also retrospectively describes October 17 as feeling good, with no later fatigue reported. The next useful observations after the October 18 follow-up were that day’s later experience and whether comfortable, well-graded sleep repeats.

Earlier observations and proposals

17 October 2026 — a social-night exception, Oura 64; COROS 82. I reported a hard workout and a refreshed preceding day, then a later evening with chicken around 20:00, approximately 2–3 oz vodka, a burger/fries and bread/Brie around 01:00–02:00, and a roughly 03:00 bedtime on someone else’s floor/carpet without a good pillow. I woke repeatedly to shift position because of discomfort and recalled falling back asleep quickly. Both providers estimated about 6h21–22 asleep; the full record retains their different waking estimates and lower HRV. This is a departure from the earlier routine, with several simultaneous changes, rather than a controlled test of the meal gap. I plan to return to the usual optimization over the next days. The night remains in the chronological and headline milestone record; any routine-only analysis would be labeled separately.

The refreshed-day report also supplies a later outcome after the October 16 sleep. A separate October 17 follow-up reports feeling already awake and sharp on waking, doing work and feeling good so far. Possible later fatigue was initially anticipated. The October 18 follow-up retrospectively describes October 17 as feeling good, with no later fatigue reported. These positive experiences stay alongside the lower device grades and HRV.

16 October 2026 — an encouraging routine, Oura 90; COROS 98. Workout finished at 15:00; last meal 17:00–18:00; sips until around 19:30; walk 21:00–22:00. Light and curtains changed through the evening. Red-filtered laptop work ended 23:50, followed by a bathroom trip and five minutes of breathing with six seconds in, four seconds out. I felt close to falling asleep and chose to sleep, roughly around midnight.

I remembered no interruption until around 07:30. The follow-up clarified just dry mouth, without urinary urge; water resolved the dryness and I returned to bed. Final waking was 09:05; the last period felt light and included dreams. I also reported a very hot workout without water and feeling dehydrated afterward, with uncertain replenishment. Usual drinks are roughly 1.5 L/day plus water from foods; the exact day’s water balance was not measured. Dry mouth alone does not identify its cause; mouth breathing is another possibility. NHS dry-mouth guidance.

After synchronization, Oura estimated 8h37m30s asleep, 2h12 REM and 93% efficiency, with a score of 90. Awake-labelled samples included 47–48 bpm just before midnight. A later COROS retrieval supplied 98, 8h59 asleep and 7 minutes awake, with official sleep HRV of 56 ms. Both record much more sleep; their waking estimates differ by about 30 minutes. The roughly six-hour meal gap is a promising experienced condition to repeat. Earlier workout, longer sleep opportunity, walk, light, water timing, breathing and the preceding short night also differ. This supports investigating the routine without prematurely deciding that food timing either caused the improvement or could not have helped. In the follow-up I explicitly rejected dozing during the extra Oura-detected evening episode; it is retained as a classification disagreement.

The next reported plan

In the October 16 report, I proposed shifting the routine about two hours earlier, aiming for a 22:00–07:00 opportunity. It offers nine hours, not nine guaranteed hours asleep. Retaining a 4½-hour liquid interval places it around 17:30; the reported 19:30-to-midnight interval already approximated that duration. The hot-workout follow-up makes adequate drinking during and replenishment after exercise relevant before tightening a cutoff. Food contributes water; needs can rise with heat and activity, so usual drinking volume is not a fixed adequacy verdict. NHS hydration guidance.

With a 22:00 bedtime, finishing food at 16:00 retains six hours; finishing at 15:00 extends it to seven. These are proposed options, not completed conditions or established optima. Preserve nutrition and training recovery. A full two-hour shift changes several conditions together; it can identify a useful routine while leaving the isolated effect of each part unresolved.

The October 16 report also adds later sleepiness to October 15 despite that short night’s initially alert morning. On October 16 I felt sharp within minutes in a meeting I had joined late. Keep those observations distinct: the urgent meeting did not measure sustained daytime performance or cortisol/adrenaline. The dated record and function guide retain both.

The preceding October 15 proposal

15 October 2026 — one follow-up, with a shorter main sleep. The window was closed and temperature felt comfortable. I remembered no interruptions before the final awakening and felt very alert, but Oura estimated 5h33 asleep and COROS 5h55. This adds a useful comfort observation; it does not complete the three-night proposal below or establish sufficient duration.

The comparison proposed on October 15

The next proposed question is whether an earlier sleep opportunity preserves comfort while increasing obtained sleep and sustaining useful daytime function. A midnight-to-07:20 window would offer 7h20 if that waking time persisted, versus the diary’s approximate 5h40 interval. It is additional opportunity, not guaranteed sleep or an adopted wake-time target.

Actual and proposed changes stay distinct:

Dinner around 17:00–18:00 and liquids around 16:00 were suggested alongside midnight bedtime. The follow-up clarified that the dinner water came from food, with no drinks at dinner. Those earlier times remain hypotheses, not evidence-derived six/eight-hour requirements. The next comparison should preserve daily food, hydration and training recovery; this night does not establish a need to restrict water-rich foods or impose an eight-hour drinking ban. Meal and fluid evidence.

The record retains the short Oura-detected evening episode, now disputed by my report of no dozing, and source-stratified heart-rate samples. No caffeine or alcohol was reported for calendar 14 October. The device classification is not counted as a confirmed nap, and the declining pulse is not proof of a successful intervention.

Testing earlier food and water timing

15 October discussion: waking with an urge leaves bladder fullness as a plausible trigger. The absence of drinks at dinner does not exclude water from the meal. A 4h40 food-to-bed gap is not a ceiling on improvement. The Bryan Johnson source review supplies a longer-interval hypothesis worth testing, without establishing that his schedule transfers here.

Two comparisons answer different questions:

1. Timing of water-rich foods: move the same half-tub yogurt, tomato and pepper earlier in the day while keeping other dinner foods and timing similar. Retain total daily nutrition, water and salt intake. This also moves calories, nutrients and food volume, so it tests the food shift rather than isolating water as the mechanism. A useful result would be a later first unwanted awakening or more sleep, rather than merely a smaller bathroom trip.
2. Whole-meal timing: compare the same daily foods with the last meal ending around six hours before intended sleep. If practical, explore eight hours and then ten as separate conditions. These are candidate steps, not validated thresholds, compulsory progression or a newly adopted schedule. Do not remove food needed to maintain intake or training recovery just to achieve a longer gap.

An earlier bedtime is also a reasonable immediate change; do not deliberately retain a short opportunity to make a cleaner experiment. Once a workable bedtime is reasonably stable, hold it similar across meal-timing comparisons. Changing both at once can identify a useful routine, but cannot isolate the contribution of either.

Compare several nights per condition and repeat a promising contrast, rather than declaring an optimum from one morning. This is an exploratory personal comparison, not a diagnostic protocol or a predetermined sample size. Keep exercise, room conditions and intake reasonably comparable; mark departures. Record attempted sleep, first awakening, urge on waking, whether a bathroom trip occurred, return to sleep, total sleep and later-day functioning. Approximate urine volume can be an optional additional measure if the bladder question remains important; it does not by itself establish the waking cause. NIDDK bladder-diary guidance.

If earlier food repeatedly delays unwanted waking and adds sleep without worsening hunger, recovery or daytime function, that is a reason to retain it. If the urge or urine volume falls but waking remains at the same time, the change may help the bathroom symptom without explaining the awakening. Keep evening pulse and wearable scores as supporting observations; the desired result is better sleep and functioning. Preserve hydration and drink when needed rather than treating a cutoff as an endurance challenge.

The original October 14 proposal

14 October 2026 — original proposal, retained for comparison. The question was whether a comfortable, earlier routine reduced unwanted disruption while retaining sufficient sleep and useful daytime function. This feasibility check is separate from the 14-day score milestone.

The night ending 14 October gives the reason. I attempted sleep around 01:30 rather than my usual 03:00–03:30. Oura estimated more sleep and a higher score, but also more waking and lower efficiency. I reported cold awakenings, a bathroom trip and possibly excessive warmth after adding a blanket. The two-provider comparison also widened. Neither device identifies the cause of each awakening.

The 14 October follow-up clarified that the bathroom trip was a precautionary choice after waking without strong urinary urgency. I reported no caffeine, alcohol or naps on calendar 13 October, and feeling awake and ready within about 20 minutes after initially waking sleepy. These reports narrow the exposure and morning questions. They do not establish the initial waking cause or later-day functioning. The dated correction retains the change in interpretation.

What is already planned

The October 14 plan was to move most food and liquids earlier, bring dinner forward, ventilate during the day and close the window from sunset, do stressful work in the afternoon, and keep the workout before dinner. These were reported intentions. The dated follow-up above shows partial implementation, without isolating a benefit from any one change.

The proposed additions are modest: begin with comfortable leg coverage and adjustable bedding, draw curtains against city light during the settling period, and finish demanding work away from bed. Retain adequate food and hydration. A red screen filter records a setting; actual light exposure and mental activation remain unmeasured. Environment and light evidence.

A first checkpoint

Repeat the planned routine for three further nights, if practical, before introducing another enhancement. Three is a proposed review interval chosen to keep the first check small; it is not an evidence-derived sample size or a completed protocol. Retain every date, including departures from the plan, and note travel, illness or other exceptions.

The primary observation is remembered awakenings involving cold or excessive warmth, with a brief description of unwanted waking. Supporting observations are each device’s own estimated sleep, total awake time, efficiency when supplied and episode boundaries, plus morning restfulness and daytime sleepiness. Sleep score remains visible; stage totals are not the success criterion for this question.

Practical success at this checkpoint would be comfortable nights without repeated thermal awakenings, enough obtained sleep and no apparent deterioration in daytime experience. A quieter night supports continuing the routine. Persistent cold or heat calls for adjusting the thermal conditions. A higher grade accompanied by worse disruption or daytime experience calls for examining the tradeoff. None of these three-night outcomes establishes an isolated treatment effect.

Several changes are bundled, and there is no stable pre-change baseline. Even repeated improvement cannot yet distinguish window closure, bedding, stress, drink timing, increased opportunity or ordinary variation. There is no reason to deliberately reproduce the uncomfortable night merely to create a control condition.

The short note that makes the comparison useful

Record the morning experience before inspecting the device score when possible. For this first check, a concise note can retain:

The two 0–4 ratings are unvalidated project diary ratings, not diagnostic scales or a vigilance test. Unknown remains unknown. A wearable cannot fill in an unreported bathroom trip or subjective rating. The diary can expand when a specific question warrants the added burden.

The bathroom trip needs particular care in interpretation. Late fluids can matter, but sleep disruption can also occur before the urge is noticed. Moving substantial drinks earlier while retaining adequate hydration is different from reducing food or total fluids to chase a bathroom-free night. A short drink/awakening note helps distinguish these possibilities; fruit water content has not been established as this night’s cause. NHS nocturia guidance, NIDDK fluid guidance.

Once the routine is practical and the relevant disruption is understood, a narrower repeated comparison can test one remaining question. The following sections explain how that stronger design differs from this first feasibility check.

Start with an answerable question

“Improve sleep” is an ambition. “Does moving my usual last caffeine exposure earlier increase obtained sleep without reducing useful daytime performance?” is a testable question.

The specification should contain:

A proposed twenty-to-thirty-minute duration gain could be a practical decision threshold, distinct from a clinical cutoff. Other questions need other thresholds.

Count measurement time as well as intervention time. A ten-minute daily task takes seventy minutes a week before setup; a three-minute version takes twenty-one but may detect smaller changes less reliably. The appropriate cadence and sensitivity depend on the decision. More precise data earns its recurring cost when it can distinguish worthwhile improvement from ordinary variation.

Establish ordinary variation

An initial seven-to-fourteen-night baseline can reveal ordinary variation and recording problems. It is a practical proposal, not sufficient evidence for every question. Retain device identity, algorithm changes, timezone and episode boundaries; separate planned changes from completed actions.

Subjective restfulness adds information, but can miss accumulating performance impairment during restriction. Include a functional measure alongside the sensor and morning note. Laboratory restriction experiment

The smallest useful context record

Record what changes interpretation, while keeping the recurring burden visible.

Add room observations for an environment experiment or intake/timing for a meal question. Expand the record when a question requires it. Unknown must remain distinct from zero exposure.

Choose a design suited to the intervention

Single-person crossover guidance emphasizes explicit periods, sequence, outcomes, carryover and within-person correlation. It suits reversible interventions whose effects can be separated in time. CENT guidance

Randomized or counterbalanced nights can suit a reversible room adjustment, reducing confounding by weekday patterns or a gradually changing training state. Expectation and adherence still matter.

Longer repeated blocks suit accumulating or phase-shifting effects: sleep opportunity, regularity and coherent light patterns. Nightly alternation can create carryover or erase the contrast. Repeating blocks helps assess time trends when returning to the comparison is sensible.

Specify stabilization and washout from the intervention’s evidence. They differ for caffeine, phase shifting and a room adjustment. Compare usual conditions with an improvement or use naturally occurring exposures; deliberate sleep loss is unnecessary.

A caffeine timing comparison

After actual use is recorded, a candidate comparison is usual versus earlier timing at the same dose. Keep other caffeine exposures and intended sleep opportunity comparable; record timing, training and adherence. Define the interval from planned opportunity and retain actual onset as an outcome. A caffeine-induced delay could otherwise make the dose look farther from sleep. No additional late dose is needed.

Gardiner’s acute trial separated conditions by 48 hours and maintained habitual morning caffeine. This is protocol context, not a universal personal washout. Trial protocol

Prior-night debt can persist after the stimulant’s immediate effect diminishes. Changing total intake adds adaptation and withdrawal. For a daily habitual-pattern comparison, use stabilized repeated blocks and prespecify how transition nights are retained or flagged.

Test next-day performance at a consistent relationship to morning caffeine. Mixing caffeinated and uncaffeinated tests would confound direct stimulation with the preceding night’s effect. Retain the daytime and training tradeoff.

Interpret the result as the net effect of moving caffeine. Residual stimulation and a changed sleep episode can both affect performance. Showing that the routine works is a useful achievement; attributing all of its benefit to better sleep requires further evidence.

Measure function consistently

A validated psychomotor vigilance task measures waking performance. A custom reaction-time game needs its own validation.

A three-minute PVT predicted performance in a laboratory luggage-screening experiment. A later 83-person study found inadequate interchangeability with the ten-minute version during sleep loss/recovery. A shorter test reduces burden but changes sensitivity. Choose one protocol before baseline collection. Brief-test study, convergence study

Keep task, device, duration and stimulant condition consistent. Allow familiarization and retain lapses and response distributions. Motivation, latency and practice can move results independently of sleep.

Choose testing time from the question. A duration comparison can match time since waking. A phase experiment targeting performance at a real clock-time demand should test that window and record the changed time since waking. That change is part of the intended practical effect; holding it constant would answer a different question.

Recording the morning rating before viewing the grade is a useful expectation control. In a randomized 63-person insomnia experiment, false positive/negative sleep feedback changed symptom appraisal but not vigilance performance. Its clinical population does not establish the size of this effect here. NHS/Wellcome/Sackler funding was reported; complete conflict review remains unverified. Gavriloff et al.

Training comparisons need a specified outcome and comparable load. Accumulated fatigue or an easier block can explain change; “recovery” should refer to repeatable performance or another named outcome. Rationale for outcome selection.

Keep the whole eligible result

Publish adherence, all eligible nights and prespecified exclusions. Flag illness or sensor failure without removing inconvenient results. Report the effect and its spread; sequential nights are correlated, and a before/after mean can hide missingness and time trends.

Consider a hypothetical nightly standard deviation of 45 minutes. Fourteen independent nights per condition give a mean-difference standard error near 17 minutes: 45 × √(2/14). An approximate 95% interval extends roughly 35 minutes either way. These are invented inputs, not measured variability or a sample-size prescription. Carryover and correlation can widen uncertainty, so a small worthwhile effect may need more data.

Biological variation and device error also affect apparent stage changes. Measurement and validation explains the relevant limits.

A milestone and a decision

Keep three decisions separate: was the benchmark achieved, should the intervention be retained, and which benefit was established? The targets define award rules; the planned experiment defines the retain/reject rule.

A score milestone can be achieved while the change is rejected because functioning reproducibly worsened. A functional gain can justify retaining a change even when the score stays flat. If function is unmeasured, report it as unknown. An unresolved comparison remains unresolved rather than proving universal ineffectiveness.

After a repeatable benefit, assess transfer to heavier training or travel. Expansion can strengthen the outcome or validation as well as add instrumentation.

Realistic timescales

Definitions and hypotheses can be established within hours. One or two weeks can characterize ordinary variation; several weeks can support a focused repeated comparison. Small effects, carryover and changing training load may require longer. Personal sleep or function gains are attainable much sooner than demonstrated disease prevention or lifespan extension.

All sleep investigations