Built

The public face has to explain the work, not just announce it.

This project already has operational nightly routes, portable event stress, daily challenge layers, governance, support cohorts, evidence linkage, route stability checks, feature dependency audits, and research confidence scoring. The page should make that visible without compressing it into one claim.

Current stack

What already exists underneath the landing.

The work is already materialized. This page exposes the stack in a way that is easier to read than a single dense home, including the checks that keep the engine from over-reading weak outputs.

Nightly operational routes

12 nightly route cards are already materialized, tiered, and governed. This is the operational spine of the project: cohort-matched interpretation, held-out metrics, interval calibration, and deployment status.

Portable event stress

6 event cards keep acute physiological stress separate from nightly sleep claims. This is currently the cleanest open portable layer in the project and should be communicated as its own stack.

Daily and non-stress layers

8 daily challenge cards and 11 non-stress daily cards show where temporal memory helps within cohort, especially for anxiety and depression, without pretending those signals already transport cleanly across cohorts.

Governance, evidence, and support

3 non-stress family cards sit on top of support cohorts, transport probes, route cards, evidence-linked reporting, and confidence governance. The landing is only the visible layer of a deeper validation stack.

Reliability controls

Route stability, feature ablation, and dependency checks are now part of the engine contract. Outputs can be downgraded when the signal is unstable, too dependent on one domain, or not strong enough for interpretation.

Delivery footprint

The build already resolves into live analytical surfaces.

This index shows the visible footprint of each layer before you open any dashboard: how many units are exposed and how much of the layer is accepted, experimental, or support.

Night Layer 12 visible units
4 accepted 7 experimental 1 support
Open layer
Event Layer 6 visible units
3 accepted 2 experimental 1 support
Open layer
Daily Challenge Layer 8 visible units
2 accepted 4 experimental 2 support
Open layer
Non-Stress Daily Layer 11 visible units
6 accepted 4 experimental 1 support
Open layer
Non-Stress Portable Families 3 visible units
0 accepted 1 experimental 2 support
Open layer

Materialized layers

What exists is already mapped to a visible analytical surface.

This is the build map of the public site: each layer already corresponds to a live dashboard, a role in the system, a concrete implementation footprint, and a review boundary for how far the output can be trusted.

Night Layer Routes, governance, held-out evaluation, interval calibration, and release gates.
governed routescalibratedstability audit
Open layer
Event Layer Portable event families, challenge cohorts, and binary/continuous acute stress models.
portable eventacute stressdependency check
Open layer
Daily Challenge Layer Daily self-report challenge cohorts, temporal probes, and within-cohort gains.
temporal probesdaily challengeconfidence gated
Open layer
Non-Stress Daily Layer Daily anxiety and depression support layers, plus open somatic anchors.
anxiety supportdepression supportsomatic anchors
Open layer
Non-Stress Portable Families Portable research families, outcome mapping, transport probes, and support cohorts.
research onlyoutcome mappingtransport probes
Open layer

Reliability layer

The engine now carries its own caution system.

These additions make the motor more defensible because they do not simply ask whether a model predicts. They ask whether the route remains stable, what it depends on, and how much confidence should be attached to the output.

What is now built

  • Route stability suiteCompares replay behavior across route variants so fragile gains can be marked as support rather than treated as strong signal.
  • Feature ablation suiteRemoves feature domains to test whether the model is broadly supported or overly dependent on one narrow input family.
  • Research confidence scoreCombines data quality, validation strength, personal history, uncertainty, stability, and dependency into an interpretable trust layer.

What it protects against

  • OverclaimingLow-signal and portable routes can be capped even when they produce a numeric output.
  • Hidden fragilityA route that only works under one split, one feature family, or one cohort context becomes visible before it reaches a public-facing interpretation.
  • Diagnostic driftThe system can prepare future clinical pathways while keeping the current public surface research only and not diagnostic.

Portable-family contract

The nightly portable layer now supports a governed blend, but only where it genuinely helps.

The blend support is now part of the technical stack rather than a one-off experiment. That does not mean every nightly family should use it. The contract exists so that a route can adopt it only when transport and replay both improve under the same governance frame.

What changed

  • Artifact contract`portable_family_blend_v1` is now supported end to end across inference, refresh, governance, and provenance.
  • Deployed example`stress_d1_family_portable_experimental` now runs as a governed `weighted_pipeline_blend`.
  • Why it mattersThe nightly portable stack can now adopt a blend without leaving the single-estimator inference contract that the rest of the system expects.

What did not change

  • Stress D+1 statusTransport improved, but the route still remains `experimental` because the held-out signal floor is not yet crossed.
  • Stress D0 stance`stress_d0` was re-probed with a blend and did not improve materially, so it stays single-model.
  • Methodological ruleNo family should adopt a blend just because the contract exists; it has to earn it in both transport and replay.

Participant-facing layer

The stack can now return value to people without collapsing into a sleep app.

This new layer is framed as participant research return rather than consumer wellness. It uses approved nightly stress routes and translates weekly longitudinal output into a form that a participant can read without losing the system’s quality gates, caveats, or provenance.

Current form

Institution Materials

The institution-facing layer is also prepared.

Today’s work is not only inside the analytical stack. The project now also has partner-ready one-pagers, a technical data requirements pack, a print-ready version, outreach templates, and a PDF export path for first institutional conversations.