SYNTHETIC PROOF OF CONCEPT · Every health value and 30-night result on this page is invented for demonstration.

Research-to-product flow

Research tells us average sleep advice. Sleep Tight personalizes it.

GPT-5.6 research pushed the product away from risky real-time sleep-stage reactions and toward the moment before sleep, where daily context, routine consistency, stress signals, and heart rate are more actionable. The demo below shows that decision step by step.

1
Start with the researchGeneric guidance works on averages, but each person responds differently to light, sound, activity, and stress.
2
Use the pre-sleep windowThe phone sends the day's vitals. The watch captures a bounded 10-minute heart-rate window before bed.
3
Let Codex choose a bounded planAt 10:05 p.m., Codex Scheduled analyzes the latest data and selects a safe lamp-fade and background-sound routine.
4
Learn across nightsMonthly logs summarize what worked and how the next routine should be hyper-personalized.

Sleep Tight platform

From wrist signal to a better bedtime.

A complete, readable demonstration of what the watch sends, what Codex decides at 10:05 p.m., how sleep quality is analyzed when possible, how the bedroom responds, and what the system learns over 30 nights.

Open data dashboardOpen lamp simulationOpen sleep reportOpen future nudges
Demo nightNight 30 of 30All systems ready
1

Nightly automation

Codex receives the day on schedule.

Press replay to watch the demo capture occur. The watch sends its recent heart-rate window, the phone supplies the day and previous night, and Codex turns those inputs into tonight’s plan.

Open Mac dashboard
Scheduled job · every evening

Ready to replay the 10:05 p.m. run

W
Watch captureRecent heart-rate packets
P
Phone recordsSleep, activity and oxygen
C
Codex analysisSleep quality, signals and trends
✓
Plan readyBedroom command prepared
No replay running. The real service would wait for its scheduled trigger.READY
Watch capture

Recent body signal

Synthetic data

A compact 10-minute pre-bed window, not continuous medical monitoring.

Heart-rate packets received600
Median heart rate66 bpm
Observed range61–72 bpm
Capture window21:55–22:05
Delivery statusComplete
Samsung Health demo import

Sleep and daytime context

9 synthetic records

Non-zero demonstration values, stored separately from the genuine empty Samsung import.

Previous night’s sleep7 h 14 min
Sleep quality estimate82 / 100
Sleep stagesDeep 1:08 · REM 1:34
Overnight oxygen95% average
Steps and activity7,842 · 42 min
Active / total calories438 / 2,184 kcal

Logged at 21:55:12. Demo values are persisted in computer/data/demo/.

In plain language: the system received enough demo information to form a cautious bedtime plan. These numbers are not diagnoses, and one unusual heart-rate value would never control the bedroom by itself.
2

Codex bedroom control

The plan becomes light and sound.

Move the clock to see exactly what the generated plan does. In this proof of concept, the routine begins after the 22:05 analysis and reaches darkness and silence at 22:35.

Open full lamp simulation

Codex-generated room state

22:05
Wind-down begins

Warm amber light comes on gently. The selected sound begins quietly.

Light32%
ColourWarm amber
Sound16%
22:05 begin22:15 dimmer22:25 very low22:35 off
Important: this browser is simulating the command. A real lamp and speaker need their manufacturer APIs plus physical calibration. Screen percentage is not melanopic lux; browser volume is not dBA.
3

30-day learning report

The system learns what helped this person.

The month below is entirely synthetic. It demonstrates how single-factor comparisons, repeated nights, and plain-language conclusions could turn a month of records into a personal routine.

Open full sleep reportOpen future nudge simulator
Thirty synthetic nights

Sleep improved gradually

Outcome combines duration and the person’s morning rating. It is not a clinical score.

Outcome /100Hours slept
Experiment verdicts

What worked—and what did not

View all 30 synthetic nights
NightExperimentSleepOutcomeMorning note
The simple conclusion: in this invented month, a consistent 22:05 routine with a 30-minute amber fade helped most. Low-arousal instrumental music was useful on tense evenings; pink noise was only mildly helpful, and brown noise did not help.