Back to home
How the Lab works

Eight specialists.
Two rounds of debate.
Every claim cited.

NightLab's Lab Bench is not a chatbot. It is a structured research process where specialized AI agents with distinct expertise analyze your data, debate their interpretations, and produce insights that are grounded in peer-reviewed sleep science. The Peer Reviewer rejects anything that isn't backed by evidence.

The process

How a Lab Bench run works

01

Context assembly

Your phenotype profile, last 30 nights of biometric data, recent check-ins, active experiments, and relevant research are gathered into a shared context.

02

Round 1 — Specialist analysis

Five or six specialists analyze your data in parallel, each from their domain. The Data Analyst runs real statistical computations. Each specialist retrieves relevant research papers and cites them.

03

Round 1 — Peer Review

The Peer Reviewer examines every claim from every specialist. Clinical assertions without primary research citations are auto-rejected. Statistical claims are checked against the Data Analyst's actual computations.

04

Round 2 — Revision

Specialists see the Peer Review verdicts and revise their analyses. Rejected claims are either strengthened with proper citations or withdrawn. A second Peer Review pass follows.

05

Synthesis

The Lab Director reads the full debate transcript and produces the final insight: a headline, a narrative grounded in your data, a confidence score, cited sources, and one clear next action.

The specialists

Your Lab Bench team

Each agent has a distinct system prompt, domain expertise, and role in the debate. The two highest-stakes seats — Peer Review and Lab Director — run on the most capable model.

Sleep Physician

Claude Sonnet

Anchored on AASM clinical guidelines. Screens for OSA, PLMD, RBD, and narcolepsy red flags. Conservative by design — flags when to see a doctor.

Circadian Scientist

Claude Sonnet

Chronotype analysis, light exposure timing, phase shifting, melatonin pharmacokinetics, and shift work adaptation strategies.

CBT-I Specialist

Claude Sonnet

Evidence-based behavioral interventions: sleep restriction, stimulus control, cognitive restructuring, paradoxical intention. The first-line treatment for insomnia.

Physiologist

Claude Sonnet

HRV and autonomic balance, thermoregulation, exercise timing, nutrition, and the pharmacology of alcohol, caffeine, and supplements.

Data Analyst

Claude Sonnet

The only agent allowed to assert a number. Runs real statistical computations — lagged correlations, partial correlations, changepoint detection, effect sizes with confidence intervals.

Hormonal Specialist

Claude Sonnet

Activated for cycle-sensitive users. Models luteal-phase temperature elevation, perimenopausal vasomotor disruption, and hormonal sleep architecture changes.

Peer Review

Claude Opus

Adversarial agent. Challenges every claim the others make. Auto-rejects any clinical or physiological assertion without a primary research citation. Your defense against AI overconfidence.

Lab Director

Claude Opus

Reads the full two-round debate and produces the final insight: headline, narrative, confidence score, citations, and your suggested next action. Calm, precise, never condescending.

Self-experiments

Your Sleep Lab Notebook

NightLab doesn't just tell you what might help — it helps you test it rigorously and keeps a permanent record of what you've learned about your own sleep.

Structured protocols

Pre-registered hypotheses, adherence tracking, and confound control across 20+ evidence-based interventions.

Real statistics

Welch's t-test, Hedges' g effect sizes, bootstrap confidence intervals. Not vibes — science.

Permanent record

Every experiment, every finding, every confirmed personal insight — exportable, always yours.

First, do no harm

Orthosomnia-aware design

A 2026 study in Frontiers in Psychology found that 17% of sleep app users worry more about their sleep after using tracking apps. People with insomnia are especially vulnerable.

NightLab actively monitors whether its own data is making your sleep worse. When it detects elevated sleep worry, excessive pre-bed metric checking, or the paradox of high anxiety with normal objective sleep, it intervenes: suppressing anxiety-inducing metrics, recommending data rest days, and shifting focus from numbers to behaviors.

Sometimes the best thing a sleep app can do is stop showing you data.

Trust and safety

What we promise

Full transparency

Every insight includes a "see the debate" button. You can read the full multi-agent transcript, including what was rejected and why.

Every claim cited

Clinical and physiological claims must cite peer-reviewed research with a DOI. The Peer Reviewer enforces this automatically.

Your data, your control

Full data export. Full account deletion. We never train on your data. Encryption at rest. Row-level security on every table.

Not a medical device

Clear disclaimers everywhere. Automatic referral pathways for OSA risk, severe insomnia, and depression screening. Crisis routing for PHQ-9 item 9 is immediate and ungated.

Ready to understand your sleep?

NightLab is launching soon on iOS and Android.

Back to home

NightLab is a product of NorthPath Strategies LLC.

Not a medical device. Not a diagnosis. Talk to your physician about anything that matters.