Observed streams
Raw life data stays separatedSystem
A layered research system, not a single predictive surface.
Butterfly Effect only stays credible if the system remains stratified in public: nightly routes, event stress, daily challenge layers, portable families, participant return, professional review, and diagnostic readiness planning each sit under different evidentiary rules.
Engine Console
The motor reads like a controlled pathway from raw life data to reviewable interpretation.
This visual layer makes the architecture more explicit: intake, quality control, route selection, evidence attachment, review surfaces, and future validation readiness stay separate instead of collapsing into a single black-box output.
Quality gate
The system decides whether it can speakRoute selection
Different questions use different routesEvidence boundary
No output leaves without its limitsReview surfaces
The answer changes with the audienceValidation readiness
Future diagnostic paths remain conditional not diagnostic todayControl layer
Quality before inference Missingness, timing, participant history, preprocessing contracts, and leakage checks decide whether a route can speak.gate first
Interpretation layer
Routes stay separated Night, event, daily, portable family, and diagnostic readiness outputs carry different evidentiary rules.stratified
Review layer
No naked score Every output keeps uncertainty, confidence, route provenance, evidence rank, and future validation needs visible.bounded
Future layer
Diagnostic path remains conditional The motor can prepare a future clinical pathway, but only external validation, ethics, and clinical ground truth can move it there.not yet
Control Board
A practical operating layer for deciding what the motor is allowed to say.
This is where a dashboard-style UI helps: it turns the architecture into visible controls, so quality, route choice, confidence and claim boundary can be read at a glance.
01
separated
Intake registry
Daily states, sleep physiology, movement, context and events enter as separate streams before any interpretation is built.
daily recordsleepcontext
mapped
02
blocking
Quality gate
Missingness, timing, preprocessing contracts and leakage checks decide whether a route can produce an output.
missingnesstimingleakage
gate first
03
routed
Interpretation router
Night, event, daily, portable family, professional review and diagnostic readiness remain different lanes.
nighteventportable
stratified
04
bounded
Review boundary
Confidence, uncertainty, provenance, audience and evidence rank stay attached to the interpretation.
confidencelimitsreview
attached
Engine
The motor is an interpretation stack, not one model.
The system now covers the full chain from data collection to review planning: intake, feature contracts, quality gates, route specific inference, route stability, feature dependency audit, confidence scoring, evidence attachment, governance, participant return, professional companion, diagnostic readiness, validation ladder, and clinical ground truth planning.
01 · Intake
Longitudinal participant record
Daily self reports, subjective cognitive state, context, intervention events, and later sleep and physiology linkage are treated as repeated evidence, not as isolated answers.
02 · Control
Quality and preprocessing gates
Missingness, timing, signal quality, feature contracts, leakage controls, and portable preprocessing checks constrain what can be interpreted.
03 · Inference
Route-specific reports
Each interpretation keeps its route, target horizon, baseline, drivers, domain findings, confidence, interval logic, caveats, and provenance attached.
04 · Stability
Route stability suite
Registered routes are audited across validation signal, missingness, participant spread, interval coverage, feature complexity, and temporal segments before they are treated as dependable.
05 · Dependency
Feature and domain ablation
The motor tests which domains each model depends on, separating sleep, activity, physiology, self report context, portable domains, and measurement quality without turning dependency into causal claims.
06 · Confidence
Engine confidence score
Each route and report can carry a research confidence score that combines data quality, model validation, personal history, uncertainty, stability, and domain dependency.
07 · Governance
Evidence rank and release gates
Route cards, held-out replay, calibration, robustness audits, portability status, and release gates decide whether a signal stays experimental, support only, or can be used in a bounded surface. External open research datasets are used as technical benchmarks for model behavior, missingness, portability, and participant-facing interpretation; they are not presented as clinical validation.
08 · Return
Participant and professional outputs
The same governed payload can become a participant research return or a professional companion, while preserving limits and avoiding consumer scoring.
09 · Validation
Diagnostic readiness and validation ladder
Internal gates now classify review readiness, map signals to research phenotypes including neurocognitive, neurological, and rehabilitation contexts, and identify external anchors needed before any future clinical claim.
Access
Operational pages stay visible without crowding the main route.
The header now carries the strongest public reading path. Built, Participant, and Views remain available as working surfaces for people who want to inspect what is already implemented.
Implemented stack
Built
What is already materialized across routes, evidence cards, governance outputs, and support layers.
Research return
Participant
The participant-facing interpretation layer and professional companion, kept inside the same claim limits.
Live surfaces
Views
Direct entry into the dashboards and layer-specific surfaces already generated from the system.
Layers
Architecture determines what can be claimed.
The project is public-facing only if the structural boundaries stay explicit. Night, event, daily, and portable-family layers can coexist on the site without being flattened into one false universal model.
Layer composition
Accepted, experimental, and support surfaces by layer.
Night Layer
Nightly routes are the core operational layer. They carry held-out signal, interval calibration, governance, and deployment status.
Event Layer
Event models remain separate from nightly interpretation. They are portable for acute physiological stress, not for next-day sleep claims.
Daily Challenge Layer
Daily challenge cohorts show where temporal memory helps within cohort, even when transport still fails across cohorts.
Non-Stress Daily Layer
This layer keeps anxiety, depression, and somatic support visible without overstating clinical portability.
Non-Stress Portable Families
Portable non-stress families remain research references. Anxiety is closest to usable; depression and somatic burden are still below transport readiness.
Atlas
Each layer sits on a different methodological timescale.
This atlas keeps the system legible at a glance. Acute event models, nightly operational routes, daily challenge layers, and portable families do not belong to the same analytic timescale or communication frame.
Layer
Primary timescale
Interpretive output
Role in surface
Event Layer
Acute stress portability
Acute
Portable acute physiological stress outputs, kept separate from next-day sleep claims.
Night Layer
Within-cohort interpretation
Nightly
Governed nightly route interpretations with held-out metrics, calibration, and deployment status.
Daily Challenge Layer
Temporal challenge layer
Daily
Temporal challenge outputs that show where within-cohort memory adds signal.
Non-Stress Daily Layer
Supportive daily reading
Daily
Supportive daily readings for anxiety, depression, and somatic burden without portable clinical claims.
Non-Stress Portable Families
Portable research family
Portable family
Research-only family portability outputs used to study cross-cohort transport outside route deployment.
Structure
The public surface should read like a ranked atlas, not a product pitch.
This page is intentionally more structural and technical. It explains method, scope, and rank before any layer is opened or any output is interpreted.
night
Night Layer
Nightly routes are the core operational layer. They carry held-out signal, interval calibration, governance, and deployment status.
event
Event Layer
Event models remain separate from nightly interpretation. They are portable for acute physiological stress, not for next-day sleep claims.
daily
Daily Challenge Layer
Daily challenge cohorts show where temporal memory helps within cohort, even when transport still fails across cohorts.
nonstress_daily
Non-Stress Daily Layer
This layer keeps anxiety, depression, and somatic support visible without overstating clinical portability.
nonstress_family
Non-Stress Portable Families
Portable non-stress families remain research references. Anxiety is closest to usable; depression and somatic burden are still below transport readiness.
Principles
The site should carry the method, not decorate it.
The strongest public surface preserves structure, evidentiary rank, and limitations in the main reading path rather than hiding them in technical footnotes.
Subtle signal, delayed effect
The project starts from a simple premise: small variations in sleep and physiology can accumulate into visible changes in stress, symptoms, and day-to-day burden.
Interpretation before diagnosis
The surface is built to organize and explain signal while preparing a cautious future validation route, not to pretend that current models already justify diagnostic language or universal clinical claims.
Evidence before expansion
Where signal is strong, the site says so. Where portability fails or outcome equivalence is weak, those limits stay visible instead of being hidden behind presentation.