DataHub

Real-time view of the Reflex Bio Data Bank · Updated just now

Total recordings

12,069

▲ 94 today

Active contributors

229

▲ 7 this week

Total time recorded

907d 16m

Continuous monitoring

BEAT rewards issued

3,568

▲ 42 today

Recordings by type 9,260 typed recordings

Sleep 2,303 24.9%

Open 1,842 19.9%

Breathwork 2,414 26.1%

Neuromod 743 8.0%

Exercise 1,151 12.4%

Resonance 807 8.7%

Recordings by device 12,069 device-tagged recordings

Reflex R&D 7,842 65.0%

Reflex Vive 2,417 20.0%

Apple Watch 1,024 8.5%

Polar H10 483 4.0%

Polar Loop 303 2.5%

Recordings over time

Daily upload volume · last 30 days

30d

90d

All

Recording ID	User	Type	Duration ↓	Captured
Loading the data bank…

Showing 1–8 of 12,069

Page 1 / 1,509

Top contributors · last 7d

01 par_anon_7b1d 182

02 par_anon_4d2f 147

03 par_anon_a3c1 128

04 par_anon_6f4a 112

05 par_anon_2e9b 98

06 par_anon_8c1e 87

07 par_anon_5a7c 76

08 par_anon_1b6e 64

Insights

Saved queries and visualizations. 14 insights · 3 shared with you

Pinned

Manage pins

VLF power · 30-day cohort trend

Mean VLF power across all sleep recordings, day-binned, with 95% CI band.

@ogmc200 updated 2h ago

Recording mix by type

Proportion of recordings by type. Sleep / Open / Breathwork / Neuromodulation / Exercise / Resonance across the whole biobank.

@ogmc200 updated yesterday

RMSSD distribution · all users

Histogram of mean RMSSD across the cohort. Bin width 5ms, range 5–80ms.

@ogmc200 updated 4 days ago

All insights

Active contributors · cumulative

Unique participants contributing each week since launch. Cumulative count.

@ogmc200 updated 1 week ago

Sleep stage distribution

Light / Deep / REM / Wake — mean proportion across all sleep recordings.

@bsm_research shared with you

Breathing rate × VLF

Scatter of mean breathing rate (br/m) vs VLF power. Linear regression r² = 0.34.

@ogmc200 updated 2 weeks ago

Recovery score · 14d rolling mean

Cohort-mean recovery score over 14-day window, smoothed.

@ogmc200 updated 3 weeks ago

Resonance frequency · cohort

Distribution of resonance breathing frequency across users, 0.05 Hz bin width.

@bsm_research shared with you

Create a new insight

Build a query against the biobank and save it as a reusable visualization.

Studies

Research projects running on Reflex Lab · [6] active · [2] recruiting

[Autonomic biomarkers in post-acute COVID syndrome — longitudinal cohort]

[NHS Trust Partner] · PI [Dr. Name]

Cohort · [120] target

Duration · [12mo]

Recruiting

[Wearable-derived ANS signals for early CV risk stratification]

Innovate UK programme · PI [Dr. Name]

Cohort · [240] enrolled

Duration · [24mo]

Active

Resonance breathing & vagal tone — community analysis

Reflex Lab community · @ogmc200 + 3 contributors

Cohort · 1,247 participants

Analysis · Week 6

Analysing

VLF signatures of overtraining in endurance athletes

[Sports Science Department] · PI [Dr. Name]

Cohort · [60] athletes

Duration · [16wk]

Active

Sleep architecture & cardiac autonomic recovery

Reflex Lab community · @bsm_research

Cohort · 892 nights

Submitted · Apr 2026

Publishing

[Perimenopause ANS biomarker discovery]

[Clinical Partner] · PI [Dr. Name]

Cohort · [80] target

Duration · [18mo]

Recruiting

Schema

Data dictionary for the Reflex Bio Data Bank · v2.4.1

signals.vlf_power

vlf_power

Spectral power in the Very Low Frequency band (0.003–0.04 Hz) computed from RR-interval time series. The VLF band carries information about long-timescale autonomic regulation and is the primary signal Reflex was built to capture — most consumer wearables sample too sparsely to compute it reliably.

Type

float

Unit

ms²

Nullable

false

Typical range

5,000 ms²

Example

{
  "signals": {
    "vlf_power": 1842.5,
    "lf_power": 624.3,
    "hf_power": 412.8
  }
}

Query examples

-- Mean VLF across all sleep recordings, last 30 days
SELECT AVG(signals.vlf_power) FROM recordings
WHERE recording_type = 'sleep'
  AND timestamp_utc >= NOW() - INTERVAL '30 days';

Docs

Methodology, signal processing, and API reference for the Reflex Bio Data Bank

Methodology

Every record in the Reflex Bio Data Bank passes through a documented signal-processing and validation pipeline. The same methodology is used internally for Reflex Health app scoring, and is what allows researchers to cite this dataset with confidence.

Pipeline overview

Raw PPG signal from the Vive ring is sampled at 128Hz and streamed to the Reflex pipeline through the BLE upload chain. From there it passes through five processing stages — capture, gating, artifact correction, HRV computation, and consent-scoped persistence — before becoming available to researchers.

PPG capture & gating

The Vive ring captures photoplethysmography (PPG) through a green-light LED and photodetector. Motion artifacts are detected through accelerometer cross-reference and an inline motion-gating filter, which rejects samples where the device is moving above a documented threshold. RR intervals are then extracted from the gated PPG signal using a peak-detection algorithm calibrated against ECG.

Artifact correction

RR intervals pass through ectopic-beat detection and Berntson-style correction. Rejected segments are flagged in the quality.motion_artifacts field but the underlying data is retained for transparency — researchers can choose to include or exclude them depending on their methodology.

HRV computation

Both time-domain and frequency-domain HRV metrics are computed per Task Force 1996 standards:

Time-domain: RMSSD, SDNN, pNN50
Frequency-domain: VLF (0.003–0.04 Hz), LF (0.04–0.15 Hz), HF (0.15–0.4 Hz)
Derived: LF/HF ratio, total spectral power

Frequency-domain metrics are computed using Lomb-Scargle periodogram to handle unevenly-spaced samples without requiring resampling. Sleep-stage-aware aggregation is applied during sleep recordings.

Validation against ECG

Signal accuracy was validated against clinical-grade ECG in n=62 paired sessions during the Feb 2026 validation study. Headline results: median SDNN delta 2.5ms, median RMSSD delta 3.1ms. Full validation dataset and code are available to approved researchers on request.

Every recording in the biobank has been contributed under explicit, granular consent. K-anonymisation is applied at the participant-ID layer before any researcher access — there is no path from a research query back to an identifiable participant. The consent_scope field documents which uses each recording was consented to.