Butterfly Effect

Institution pilot protocol

The default protocol for a first serious institutional pilot.

This protocol turns a collaboration conversation into a narrow, falsifiable, operational pilot: one cohort, one endpoint family, one benchmark path, and one explicit decision gate.

Default first frame

The first pilot should usually be one institution, one cohort, one endpoint family, and retrospective harmonization first. The default endpoint family should be stress, because that is where the current stack is strongest within cohort and where nightly transport is still the main unresolved scientific question.

One cohort One endpoint family Retrospective first Prospective continuation only if feasibility is positive

Pilot frame

The first pilot should answer a narrow question well

The objective is not to prove universal generalization in one step. It is to decide whether one real institutional cohort strengthens the current stack in a defensible way.

Primary question

  • Can the cohort be harmonized without breaking participant-night linkage?
  • Does it support valid nightly within-cohort interpretation?
  • Does it improve the transport question or only confirm within-cohort signal?

Recommended minimum

  • `>= 40` participants
  • median `>= 20` nights
  • repeated endpoint on `>= 30%` of nights or `>= 3` times/week
  • stable participant-night linkage

Preferred

  • `>= 75` participants
  • median `>= 45` nights
  • HR / HRV or IBI-derived physiology
  • at least one raw or semi-raw sidecar

Pilot phases

The protocol is intentionally four-step

Each phase has a different job. Intake should not collapse directly into modeling.

Phase 0 — Pre-screen

  • fit / no-fit memo
  • first field map
  • date linkage and device lineage check

Phase 1 — Intake and harmonization

  • schema mapping
  • field and join audit
  • canonicalized draft tables

Phase 2 — Feasibility benchmark

  • within-cohort benchmark
  • subgroup robustness review
  • calibration summary where applicable

Phase 3 — Decision gate

  • stop / continue / redesign
  • route-level, family-level, or protocol recommendation
  • prospective continuation only if justified

Success criteria

A first pilot succeeds by reducing uncertainty, not by proving everything

A strong first pilot returns a clean harmonization path, a defensible benchmark, explicit subgroup review, and a clear continuation decision.

Success

  • clean canonical harmonization
  • defensible benchmark for at least one endpoint family
  • explicit subgroup and quality review
  • clear decision on continuation or redesign

Stop criteria

  • no stable participant-night linkage
  • endpoint too sparse
  • units or lineage cannot be reconstructed
  • only aggregate analysis possible, with no nightly structure