Institutional collaboration one-pager

Research-grade longitudinal sleep interpretation, ready for data partnerships.

Butterfly Effect is a research system for detecting subtle sleep and physiological patterns that may precede changes in stress, symptoms, and health burden. The platform is strong within cohort, now has two nightly routes that clear promotion gates, is increasingly credible across cohorts for stress, and is ready for institution-facing pilot work focused on longitudinal data.

What is already real

The current stack already supports nightly routes, event stress models, daily challenge layers, portable-family research tracks, governance gates, calibration, and evidence-linked reporting. Two nightly routes now clear promotion gates under explicit governance. This is no longer a prototype shell. It is a structured research platform with explicit deployment boundaries.

Within-cohort routes operational 2 nightly routes clear promotion gates Event stress portable Night transport still the main bottleneck

Night routes12

Auto-promotable nightly2

Event cards6

Portable stress eventR² 0.316

Current strongest evidence

PMData stress D0

R² 0.238

LifeSnaps stress D+1

R² 0.346

Novartis fatigue D0

R² 0.435

Novartis pain D0

R² 0.371

Partner-specific versions and export

Use the tailored one-pagers for first outreach, and the print version when you need a clean PDF attachment.

Hospital version University version Wearable company version Technical data requirements Variable shortlist Pilot protocol Data intake + governance Target shortlist Print / PDF-ready Outreach templates

State of the project

Where the platform is strong, and where new institutional data matters most

The platform is strongest in structured nightly interpretation within matched cohorts and in portable event-level stress. The main unresolved frontier is generalization of nightly interpretations across cohorts.

Strong today

Nightly within-cohort routes with explicit held-out evaluation, including two routes that now clear promotion gates
Event stress portable family with positive cross-cohort performance
Calibrated reporting, governance gates, and PostgreSQL-backed route materialization

Improving

Nightly stress transport across cohorts, now positive in the open stack but still below deployment signal floor
Daily temporal layers for anxiety and depression
Research-only portable tracks such as `PHQ-9 only` depression

Not yet strong enough

Portable anxiety as a deployable clinical family
Portable depression across mixed instruments
Portable physical pain beyond somatic-burden proxies

Why now

Why this is the right stage to approach institutions

The project is mature enough to justify a serious collaboration, but still open enough that additional longitudinal datasets can clearly move the science.

What we can already bring

A working interpretation stack, not only exploratory notebooks
Existing open-cohort validation and a clear benchmark baseline
Governance layers separating accepted, experimental, and support-only claims
A reproducible data-to-report pipeline suitable for pilots

What institutional data would unlock

Stronger nightly transport across cohorts
Cleaner outcome equivalence for anxiety and depression
Validation of interpretable patterns against richer physiological measurements
Prospective evidence instead of only open retrospective harmonization

Data requirements

The minimum dataset for a good interpretation, and the ideal expansion path

EEG and ocular channels are highly valuable, but they are not the minimum requirement for the interpretation layer already being built. The core need is longitudinal nocturnal physiology plus repeated outcomes.

Tier 1

Required nightly core

Sleep timing, duration, continuity, awakenings or efficiency, nocturnal HR or HRV, and actigraphy or movement fragmentation.

Tier 1

Repeated outcomes

Stress, fatigue, anxiety, depression, pain, or symptom burden measured repeatedly, ideally daily or several times per week.

Tier 1

Context variables

Age, sex, medication, alcohol, caffeine, exercise, shift work, relevant diagnoses, and major acute events.

Tier 2

Highly valuable raw signals

PPG or IBI, minute-level HR, raw accelerometry, temperature, respiration, and SpO2 where available.

Tier 3

Gold-standard subset

PSG subset with EEG, EOG, EMG, plus aligned symptom outcomes for calibration and physiological validation.

Pilot collaboration

A practical first collaboration model

The right first study is not “all available data.” It is a sharply defined longitudinal pilot around one outcome family and one institutional dataset.

Recommended pilot design

One institution, one dataset, one primary outcome family
Longitudinal wearable sleep plus repeated self-report outcomes
Retrospective harmonization first, then prospective extension if feasible
Optional PSG subset for physiological anchoring

What the institution receives

Data schema and harmonization specification
Benchmarking against the current open stack
Interpretable route-level and family-level outputs
Feasibility results for transport, calibration, and subgroup robustness

Governance posture

Research-only framing unless a route is explicitly accepted
Pseudonymized identifiers and DUA-compatible ingestion
Clear separation between accepted, experimental, and support-only outputs
No diagnostic claim inflation

Best current institutional fit

The best collaboration target is a group with longitudinal wearable sleep, nocturnal cardiovascular signals, and repeated symptom data. The highest-impact contribution would be a dataset capable of testing nightly transport across cohorts rather than just adding another single-cohort benchmark.