Questions

What is Atum actually building?

Atum is not another wellness app, respiratory score, or wearable dashboard. The thesis is narrower and stranger: that respiration may become a continuous computational layer for physiology. The current product path exists to test whether that layer is real:

Why breathing?

Respiration is potentially continuous, passive, multi-system linked, and observable through commodity microphones — a rare combination. No other physiological signal offers all four at once. Whether it also carries enough stable information density to support a new computational layer is the open question this work exists to answer.

Why is breathing different from most biomarkers?

Most biomarkers are sparse snapshots — a value at a point in time. Respiration is continuous structure. It is simultaneously connected to autonomic regulation, cardiovascular dynamics, nervous system activity, metabolic demand, sleep architecture, and stress regulation. The question is whether that structure stays computationally meaningful across long stretches of time, not just within a single recording.

Why now?

Three things changed at roughly the same time: microphones became ubiquitous and high quality, edge inference became cheap, and sequence models became capable of handling long temporal structure. Continuous respiratory observation at scale was unrealistic ten years ago. It may be realistic now.

Why microphones?

Microphones already exist everywhere — phones, earbuds, laptops, homes, cars. The thesis is not “invent a new sensor.” It is that physiology may already be partially observable through infrastructure humanity already carries.

Why not simply use wearables?

Wearables matter, and Atum is not a replacement for them. But most wearable systems observe physiology indirectly — through pulse, HRV, movement, temperature. Respiration may be different because it is continuous, mechanically expressive, and tightly coupled to autonomic state. Whether that makes it a richer temporal substrate, rather than just another sensor stream, is part of what the research has to settle.

What can respiration already reveal?

Existing research already associates respiratory patterns with:

Atum is not claiming to diagnose any of these. The point is narrower and more interesting: respiration already appears to carry surprisingly rich physiological information across multiple systems.

Why is Parkinson’s research important here?

Several recent studies suggest respiratory changes may appear before many visible neurological symptoms. The significance for Atum is not “Parkinson’s detection.” It is the broader implication — that respiration may carry early information about systemic physiological change long before classical symptom layers emerge.

Why does continuity matter?

Most medicine still operates through snapshots — tests, visits, isolated measurements. But physiology itself is continuous. The thesis behind Atum is that many important physiological shifts may appear first as changes in temporal structure, before symptoms emerge. Whether respiration is stable enough to capture those changes reliably is exactly what longitudinal data has to show.

What does the world look like if Atum is right?

In the strongest version of the thesis, respiratory continuity becomes a foundational layer for preventive medicine, health infrastructure, biometric systems, AI health agents, adaptive environments, and personalized medicine. Breathing becomes continuous, persistent, and computationally meaningful — and respiratory history itself begins functioning as infrastructure. This is the upside case, not a prediction.

What will users actually experience in the first product?

The first version is intentionally simple. Not dashboards, not scores, not optimization loops. The goal is more basic: to let people observe their own physiology as a continuous process for the first time — recovery, shifts, continuity, changes after sleep, stress, training, flights, food, meditation. Whether that kind of awareness becomes genuinely valuable over time is something the first cohort of users will show, not something a demo can prove.

What are people actually paying for?

Probably not “respiratory analytics.” The strongest near-term value appears to be recovery awareness, stress visibility, physiological continuity, and a deeper sense of self-observation. The open question is whether “observation without interpretation” can become a durable product category rather than a temporary novelty — and that is a question about retention, which only time answers.

Why start with sleep and recovery?

Because that is where respiration currently appears strongest, cleanest, most continuous, and least noisy. Sleep is the most likely place for continuous respiratory observation to prove useful first. What remains open is whether it expands naturally into a larger physiological layer, or stays primarily a sleep and recovery product. Both outcomes are live.

What exactly is “Breathing State”?

Right now, it is best understood as a computational representation of respiratory structure over time — not a diagnosis, not a disease label, not a score. Early observations show recordings can cluster, repeat, and partially separate across individuals. Whether that representation becomes stable, persistent, reusable, and biologically meaningful across contexts and time is unproven — and it is the single dependency the rest of the system rests on.

Does respiratory history actually compound in value?

This is one of the most important unanswered questions. The thesis requires that longer physiological history materially improves the system — not slightly, but enough that forecasting improves, baseline confidence improves, and downstream systems become dependent on continuity itself. There is precedent for this pattern in other continuous signals, but for respiration specifically it has not been demonstrated. Establishing it is the central goal of the longitudinal dataset.

What if the baseline is always changing?

This is one of the hardest unresolved problems. Healthy adaptation, fatigue, stress, disease, environmental shifts, and behavioral change may all look structurally similar. If the system cannot reliably tell them apart over time, the “observation without interpretation” thesis weakens considerably. Distinguishing these is an explicit research target, not an assumption — intra-person versus inter-person variance and drift behavior are exactly what the validation work measures.

What is engineering, and what is unresolved science?

The engineering problems are real but bounded:

The existence of a stable Breathing State is not engineering. It is an open scientific question. Part of the thesis cannot simply be built — it first has to be shown to exist. Keeping that line clear is part of how the work stays honest.

Why doesn’t Apple simply win?

If Atum is only respiratory AI — embeddings, anomaly detection — then large incumbents likely win quickly, and little remains defensible. The thesis only survives if respiratory continuity becomes a new representation layer, a reusable substrate, something deeper than a feature. Whether it does is unresolved — and it is the specific thing the first datasets are built to test.

Why should Atum exist as a standalone company?

This is an open question internally, and it is treated as one. If respiratory continuity becomes a foundational computational primitive, Atum may become infrastructure. If it does not, the market will naturally absorb these capabilities into wearables, health assistants, and multimodal AI systems. The company is built to find out which world it is in, not to assume the answer.

What happens if multimodal systems outperform respiration-only models?

Then Atum risks becoming a feature supplier — a signal layer, a component inside larger systems. Avoiding that outcome requires showing that longitudinal respiratory continuity creates information multimodal systems do not naturally recover without it. This has not been proven. It is one of the clearest pass/fail tests for the thesis.

Why haven’t incumbents already won the “time moat”?

Apple, Google, Whoop and others already have years of data, continuity, devices, and behavioral integration. So the burden is on Atum to show that respiratory continuity is fundamentally different from generic longitudinal health data. If it is not different in kind, the “time moat” simply belongs to whoever already owns the user — and that is a risk the thesis has to answer directly, not around.

What breaks the thesis?

One especially dangerous possibility: respiration may prove useful, predictive, and scientifically rich — but not stable enough, persistent enough, or non-redundant enough to support a new infrastructure layer. In that world, Atum could still become a strong respiratory intelligence company, but not a foundational computational substrate for physiology. Naming this outcome clearly is part of taking the thesis seriously.

When does this stop being a thesis and become a category?

Probably when three things happen at once:

That is the point where continuity itself starts behaving like infrastructure rather than narrative. None of the three is established yet — but each is measurable, which is what makes the question answerable rather than rhetorical.

What is the minimum proof required for institutional conviction?

Not demos. Not respiratory AI. The minimum proof likely looks like:

Without this, Atum remains compelling, intellectually coherent, and scientifically interesting — but unproven as infrastructure. The roadmap is organized around producing exactly this evidence, in this order.