Search full texts, projects, topics and technologies.
↑ ↓ chooseEnter openEsc close
Experiment · Running · 29 October 2026
SOMA//TRACE
A body, partially observed. A generative self-portrait in the space of data: four layers — movement, sleep, body composition, publications — one sculpture, sound, and the honest limits of what measurement cannot know
The band remembers my steps and estimates how I slept. A body-composition analyser keeps rare cross-sections of change, and the site keeps traces of what I chose to publish. I can overlay these series and build a sculpture out of them, but I cannot reconstruct myself from them. SOMA//TRACE is an attempt to make that boundary visible: to give the data a form and a sound, and then to show where measurement ends and interpretation begins.
This is not medical monitoring and not a health dashboard. I am at once the subject, the data source, the editor of my own disclosure and the researcher — and the question that interests me is how far a measuring device can know a person at all. The body is lived whole; it is recorded partially. The band registers sleep stages but does not know dreams. Steps know nothing of why one walks. InBody gives discrete slices over two months, not a continuous copy of the body. Publications leave a public trace that hides the volume of thought and labour behind it.
At school I listened to Lyosha domiNo. His 2011 album trюк opened with “Спродюсирую любовь” (“I'll Produce Love”); at the end, I discovered that the beat was actually his own heartbeat, recorded at a cardiology centre. It gave me goosebumps. That openness with an audience, however small, even from a small town, has stayed with me as a point of reference. SOMA//TRACE uses different material: its sound is synthesized from movement, sleep and publication data, not recorded from a heart.
What you can do here
In the experience itself — orbit the scene with mouse, touch or keyboard; move time from a week to the whole history; switch and isolate the four layers; enable the generative sound — it starts only on your action and changes with the period; open interpretation events and see what they were compared against and with what confidence; in SOURCE, read the whole path of the data, including the withheld. On supported devices you can enter AR (the sculpture on your desk) or VR (inside your own traces); elsewhere a screen version is offered, and the page says so honestly.
The body the device does not know
The central conflict of the work: the lived body and its digital trace are not the same thing. Sleep sits next to publications here not because a causal link has been proven, but because they are two different modes of the author's existence: the nocturnal, closed, band-measured one, and the diurnal, public one that leaves traces on the site. The scene lets you study co-occurrence — "the same week had a burst of publishing" — but it never asserts causality: never "published more because of bad sleep".
An aesthetic concept
The form of the work is not an illustration of the data but their "matter". The veils of sleep flow upward, like a dream one cannot fully recall; the fibers of movement wrap the silhouette, because activity lives around the body, not inside it; the golden core is the most discrete and the most corporeal — rare measurements, physical mass; the orbits of publications are the outermost, most public layer, where each event takes its place on the ellipse of time. Color here is grammar, not decoration: violet is sleep, cyan is movement, gold is body composition, amber is publication. The form is legible without color too: matter, lines, volume, nodes, the hatching of absence.
I deliberately choose glow over photographic accuracy: measurement does not look like a body — it looks like the trace of a body, the light left after a sensor is touched. Overexposure would be the worst possible mistake: it would mean interpretation drowning out measurement. So the sculpture carries exactly as much light as the data provides.
The problem of measurement objectivity
Every source in this work errs in its own way, and honest practice means showing that, not hiding it. A consumer band does not measure sleep stages — it estimates them from indirect signals; hence always "estimated share" in the wording, never "fact". InBody measures discretely, with per-device error; hence measurements are key states and everything between them is labelled interpolation. Publishing activity measures not thinking and not labour, only the decision to publish. The one thing known precisely is absence itself: a week without observations is shown as a pause, not an invented value.
The meaning of artistic interpretation
Interpretation events are not diagnoses and not "insights" — they are a way of seeing one's own norm. The device compares me with myself: the median of eight weeks, my own usual rhythm. When the scene marks "a week noticeably shorter than the personal rhythm", it does not say something is wrong — it asks what was different. That is the role of artistic interpretation: to give the data back its question mark, which algorithms usually lose on the way to a dashboard.
Materials and sources
Huawei Band 10 → Huawei Health: steps, distance, estimated sleep stages. In use. Weekly medians from BODY//DATA — a local archive without clouds or third parties, with a publication delay and explicit annotated exclusions.
How talking to Huawei became possible at all: the official bridges are unavailable for an RU-region account, and the official data export would have taken weeks. The solution turned up unexpectedly — in a repository integrating Huawei Health with a smart home (and7ey/huawei_health, MIT), where the cloud protocol had already been reconstructed and documented. BODY//DATA took that vendored client as its base. So one open source project, built for a "smart home", opened the road for an artwork about the body — to me, the best possible illustration of how public traces of information connect unpredictably.
InBody: discrete body-composition measurements. The public dataset retains absolute weight, skeletal muscle mass and body-fat percentage for the sculpture. Publication requires separate approval tied to the current dataset's SHA-256.
Public activity of staniverse: publications, articles, works, projects. In use — through the same canonical count as the garden calendar: hidden records excluded, the life channel only with an explicit opt-in.
Laboratory tests: absent for now; if they appear — a separate layer and a separate disclosure decision.
Method and the grammar of form
Every layer has a path: source → published feature → transformation → form. Normalization is fixed across the whole history (weeks are comparable; the window is never rescaled per view), and randomness is deterministic from a fixed seed.
Layer
Feature
Transformation
Form
MOTION
median steps/day, distance
intensity 0..1 on a fixed scale
density and branching of cyan threads
SLEEP
median duration, stage shares
duration → wave span; deep → slow swells; REM → fine ripples
violet translucent shells, pause gaps
MASS
InBody weight, skeletal muscle mass and body-fat percentage
the core's radius from absolute weight; between measurements — labelled interpolation; before the first measurement — UNKNOWN state (dashed outline, not a real body)
the golden core; before the first measurement — a smaller dashed outline
OUTPUT
the number and dates of a week's public events
voice waves across the disc; in the sound — murmured "voice" phrases (a publication is a voice on the internet)
amber waves and trace nodes of the outer constellation
Three readings of concrete parts: (1) an amber node: "a garden record → the week's public-event count → node brightness → this pulse"; below it: "this is a trace of publishing, not a measure of productivity". (2) a violet wave: "Huawei Band 10 → the estimated REM share → the ripple frequency of the shell → this wave"; a consumer band estimates stages and can be wrong, REM is not a dream count. (3) the golden core: "an InBody measurement → the absolute weight → the core's radius → this form"; between measurements the morphing is artistic and labelled as interpolation.
How to read the anomalies
Events are compared against the personal rolling baseline — the median of the previous eight valid weeks (a minimum of four; otherwise the event is honestly not computed). "A week noticeably shorter than the personal rhythm" — the night median below 75% of the personal median. "A higher estimated REM share" — above the personal median by more than one and a half robust spreads and at least five points. All thresholds are open in the event inspector together with the method version. These are artistic heuristics, not diagnoses: REM is not a "number of dreams", a short night is not "sleep debt", a quiet publishing week is not idleness.
What we do not know
The limitations are built into the work, not hidden. Consumer sleep staging errs — every wording says "estimated share". Stage intervals are not published — there is no invented hypnogram. InBody measures discretely — the morph between measurements is labelled as artistic interpolation. Weeks without observations are shown as pauses and gaps, not zeros; a missing post is not a conclusion about inactivity. Bedtime is deliberately withheld, so the "late onset" and "schedule shift" events are marked as not computable: the scene cannot derive them and does not pretend to.
Privacy as part of the art
These are controlled disclosures of a named self-portrait, not anonymisation. Bedtime and wake time, minute-level stages and geolocation are withheld. Heart rate, its variability and SpO2 are absent from the current public data. The dataset includes absolute weight, skeletal muscle mass and body-fat percentage; unrelated body-composition values and the unused tracker score are excluded. The site validates the dataset against a strict schema and falls back to a neutral demo if it is invalid. Publication requires the author's approval for the exact dataset; withdrawal blocks future publication, while already published and third-party copies must be removed separately.
Stepping inside the trace
In AR, the sculpture is placed on a found surface of your desk — you can walk around it, rescale it and toggle layers. In VR (Quest 3) you can step inside your own trace: the sky of sleep around you, the orbits of movement, the constellation of publications — and a timeline ring that moves time with your hand. The sound is spatial and born from the same data; it can always be muted. The language switches inside the session without losing your place or state. On devices without WebXR, a screen version is available with an honest label of why the immersive mode is not.
Context and credits
Quantified Self (Gary Wolf, Kevin Kelly) — the culture of self-tracking; SOMA//TRACE inherits its instrumentality but refuses its optimisation pathos.
Data criticism — the work argues against the idea that measurement equals knowledge of a person; data here is a material of narrative, not of truth (cf. Jenny Odell's "Data Fiction").
Art & Science as a field and Digital Humanities (ITMO master's programme): joining an artistic statement with scientific honesty about method — the context the author works in.
Parametrick — the nearest relative in the Lab: there, features of a song or a photograph become the sections of a vase; here, features of sleep, steps and measurements become the matter of a sculpture. Mnemoform continues the same line: the idea of "personal data → personal form", started in Parametrick and tested on living bodily data here, grows there into the memory of a place. All three works are one attempt: personal data can become a personal form without a single invented value.
Author and maker: Stanislav Ermolenko (staniverse). Body data source: BODY//DATA — a local archive with manual approval of every publication. Engine: three.js, Web Audio API, WebXR; the scene and audio method — soma-art-1. Update history: 2026-10-29 — first publication of the work.
Nearby connections
A compact neighbourhood from the archive corpus: the direct relations — Parametrick, Mnemoform, staniverse — plus shared topics.