Cherish Health

Radar returns reflected energy and nothing else. Every meaningful output this product generates is inference performed on that signal, which makes the models not merely central but constitutive.

The difficulty of the inference is what distinguishes this from sensing products that simply threshold a reading. Separating a person lowering themselves deliberately to the floor from a person falling requires reading body position through the descent, since the endpoints are identical. Catching a fall where the person grabbed furniture on the way down is harder still, and is precisely the case a wearable accelerometer misses because the sharp deceleration never occurs. Extracting heart rate and respiration from micro motion at distance and through intervening walls is a signal recovery problem well below the noise floor of naive processing.

The hardware is a radar front end and an edge compute platform, and hiring for printed circuit board layout, field programmable gate array and digital signal processing roles indicates the signal chain is engineered in house rather than assembled from a module.

Unlike the wearable competitor in this lane, no substantial non predictive product sits alongside to carry value independently. There is no nurse call hardware, no access control, no location beacon infrastructure. Remove the models and a buyer is left with a radio that detects motion. Graded A on that basis, with the corresponding concentration of risk noted: every question about this vendor is a question about the models.

A clean notification model with a deliberate false positive control, and no published measure of whether either works.

The design is sound. The device detects and notifies; a caregiver or a monitoring response centre decides and acts. The system takes no clinical action and makes no clinical claim about what a detection means. Routing to a response centre rather than only to a family member is meaningful, because it provides a path when no relative is available.

The fall versus intentional descent discrimination deserves specific credit on this axis rather than only as engineering. Its purpose is to avoid raising an alarm every time someone sits on the floor, and building deliberate false positive suppression into a life safety product is the same category of design choice as an abstention class in a screening model: it accepts less apparent coverage in exchange for alerts that staff will still trust on the hundredth day.

What is missing is calibration, and for a life safety device that is the substance of the axis. No sensitivity, no false alarm rate, no alert volume figure, no time to notification, and no stated behaviour when a fall is detected and nobody acknowledges the alert. Nothing describes escalation ordering between family, dealer monitoring centre and emergency services, or what coverage hours a response path carries.

Connectivity failure is partially addressed by cellular defaulting rather than home broadband, which removes a common failure mode, and nothing states what happens when cellular itself is unavailable.

Ask for fall detection sensitivity and false alarm rate in occupied homes, the escalation ladder, and the behaviour on an unacknowledged alert.

The approach is explained with more technical specificity than most hardware vendors offer, and no number describes how well it performs.

What is disclosed is substantive. The modality is named as radar based spatial computing, the inference location is named as the edge, coverage is described as multi room and multi person from a single unit, and the classification targets are enumerated: body position, movement, falls, heart rate, respiration. The two hard cases the models are built to handle are described explicitly, namely separating deliberate descent from a fall and catching a fall broken by grabbing furniture. Explaining which failure modes a model is designed against is a more informative disclosure than a general accuracy assertion, because it tells a reader what the designers understood the problem to be. A range figure is given, at 40 feet through two walls, which is a testable physical claim.

Performance is entirely absent. No detection rate, no false alarm rate, no operating point, no model card, no architecture description and no statement of the radar band, bandwidth or resolution, which for a radar product are the specifications that bound what is physically achievable.

The strongest sensing claim carries a qualifier the marketing does not always repeat. Heart rate and respiration through two walls is described by the chief executive as proven in the company's own laboratory, which is a controlled demonstration rather than a field result, and a reader encountering the claim second hand would not know that.

Ask for the radar specification, detection performance in occupied homes, and whether the vital signs capability is shipping or laboratory only.

The partner layer is named openly and the component layer is not, with one structural fact working in the vendor's favour.

What is named is real. AT&T is identified as the connectivity provider, Alarm.com as the enterprise platform partner carrying the product to its service provider network, and security dealers including ADT as the installation and sales channel. Naming the parties that sit in the delivery path is more disclosure than most vendors in this index manage, and each is independently verifiable.

The structural fact is edge inference. Because classification runs on the device, no third party cloud model provider sits in the inference path, which removes the largest supply chain question this index usually has to ask. A vendor that processes locally cannot be sending patient audio or imagery to an undisclosed model provider, because it is not sending it anywhere to be classified.

The silicon is undisclosed and this is a hardware product. No radar chipset or front end vendor is named, and millimetre wave radar front ends come from a small number of semiconductor suppliers, so the dependency is real and concentrated. Hiring patterns across printed circuit board layout, field programmable gate array, digital signal processing and embedded software roles indicate the signal chain and board are developed internally, which is a meaningful signal about how much of the stack the company controls, but the underlying transceiver is almost certainly sourced.

No cloud provider, sub processor register or training data provenance is published, and nothing states whether data from deployed devices contributes to model development.

Ask for the radar front end supplier, the backend hosting provider, and whether deployed device data trains models.

A dedicated pass located no peer reviewed publication, no clinical study, no independent evaluation and no published accuracy figure for any function of this product.

What exists instead is channel breadth, and channel breadth is not clinical evidence. AT&T, Alarm.com, ADT and a senior care partner announced at a health technology conference are distribution and connectivity relationships. They establish that large organisations were willing to carry the product commercially, which is a signal about business development rather than about whether the device detects falls reliably. This index recorded the same distinction against a competitor that reached roughly 80 schools over nearly a decade and produced no outcome data at all.

The flagship technical claim is explicitly self reported. Detecting heart rate and respiration from 40 feet through two walls is described by the chief executive as demonstrated in the company's own laboratory. A laboratory demonstration under chosen conditions is a legitimate engineering milestone and it is not evidence of field performance with real bodies, furniture, pets and movement.

A design award recognises design. A veterans centre placement establishes deployment, not outcome.

The absence matters most for the primary function. Fall detection is a life safety claim, the population is people who fall, and no sensitivity, specificity, false alarm rate or comparison against any alternative was located. A buyer replacing a pendant or a wearable with this device has no published basis for believing the substitution is safe.

Ask for a field study of fall detection performance in occupied homes, with sensitivity and false alarm rate, ideally against a comparator.

Two architectural decisions that materially reduce risk, and no operational detail behind either.

The first is the modality itself. Radar produces no image, so no picture or video of a person undressed, in a bathroom, or asleep ever exists to be leaked, subpoenaed or viewed by staff. Category coverage notes that older adults frequently refuse in home cameras outright, and choosing a sensing method that cannot produce an image is a stronger privacy guarantee than a policy promising not to look at one. The second is edge inference. Cherish AI is stated to run on the device, which means classification happens locally rather than in a cloud holding a continuous feed.

Together those reduce both what data exists and what travels, and they are design choices rather than assurances. This index should credit that.

What is absent is everything operational. No encryption statement, no retention schedule, no access control model, no deletion process, and critically no statement of what actually leaves the device. Edge inference is compatible with transmitting only discrete events, and equally compatible with streaming derived signals continuously, and the difference is the whole privacy question. Nothing published resolves it.

One consequence of the technology is unaddressed anywhere and is specific to this vendor. Radar sensing through walls detects whoever is present, not whoever consented. Multi person capability is marketed as a feature, which means visitors, family members, care staff and, in a facility or apartment building, people in adjacent rooms may be sensed without any relationship to the purchase.

Ask what leaves the device and how often, what is retained centrally, and how people other than the subscriber are handled.

Two dedicated passes located no health privacy position: no compliance statement, no business associate agreement template, no execution requirement and no description of which entity contracts with whom.

The channel structure makes the omission harder rather than simpler. This product reaches consumers through security dealers and a smart home platform partner, and reaches facilities through that partner's service provider network. A consumer purchase from a security dealer sits outside health privacy law entirely, so protections over continuous in home sensing data are whatever the privacy policy grants. A nursing home deployment may fall inside it. Nothing distinguishes the two, and a family buying a health monitoring device from a home security company has no way to know which regime applies to them.

The agreement chain is the specific unanswered question. With a connectivity carrier, a platform partner and an installing dealer between the vendor and the monitored person, responsibility for the data can sit in several places, and no published material describes where the vendor believes it sits or who signs what.

The monitored data warrants the attention. Continuous presence, position, and respiration inside a bedroom or bathroom describes a person's private life in detail, even though no image exists.

Graded D because nothing at all was located after a dedicated search. Ask which deployments the company treats as covered, whether it signs agreements with facility operators, and what governs data handled by dealers and platform partners.

Two dedicated passes located no security page, no external attestation, no trust center, no penetration testing statement, no vulnerability disclosure policy and no documentation offered under agreement. No security credential of any kind was found.

The surface here is broader than an application. This is an always on radio frequency sensor placed in bedrooms, with cellular connectivity that bypasses whatever network protections a household or facility operates, running inference locally on hardware in an unsupervised location. Device authentication, firmware signing and update practice, and physical tamper handling are all unaddressed, and for a connected device in a private dwelling those are the primary questions rather than secondary ones.

The distribution channel adds an access dimension that most vendors in this index do not carry. Installation is performed by third party security dealers, which means installer accounts, provisioning credentials and dealer platform access exist between the vendor and the monitored person. Nothing published describes how dealer access is scoped, audited or revoked.

The consequence of compromise is unusually direct. A feed indicating whether a dwelling is occupied, whether the occupant is asleep, and whether anyone else is present is exactly the information that would be useful to someone intending a burglary or targeting a vulnerable adult, and it is being generated in homes reached through a security industry channel.

One pre emptive note: a privacy policy or a restated privacy by design claim cannot move this grade. Only an external attestation, or security documentation available under agreement, will.

Ask for any external security assessment, the device authentication and firmware update design, and how dealer access is controlled.

Safety positioning carrying a vital signs capability, with no determination published and two regulators potentially in scope.

No clearance, grant or approval was located and none is claimed. The consistent framing is safety monitoring, emergency detection and independent living, which is the recognised wellness route, and fall detection for emergency response sits comfortably on it alongside conventional personal emergency response products.

Heart rate and respiration are the complication. Those are vital signs, contactless measurement of them is an active area of device regulation, and press coverage has described the company as a medical sensor manufacturer and the product as medical driven technology. Whether the vital signs capability is shipped as a monitored parameter, held as a laboratory capability pending regulatory work, or presented only as wellness information is not stated anywhere, and those are three materially different regulatory positions. The associated claim is itself described as a laboratory demonstration, which may indicate the capability is not commercially released, but the public material does not say so.

A second regime applies here that does not apply to most vendors in this index. This is a radio frequency emitting device operating in an occupied dwelling, which brings spectrum authorisation and emission limits under communications regulation independently of any health determination. Nothing published addresses the authorisation the device holds.

The distribution channel adds a third question, since selling a health monitoring device through security dealers places it alongside products marketed under different rules.

Ask for the written device determination, whether vital signs are commercially available or laboratory only, and the radio authorisation the device carries.

A dedicated pass located no fairness testing, no subgroup performance, no calibration data, no validation methodology, no drift monitoring and no external audit.

The bias mechanism in radar sensing is physical and therefore unusually concrete. What the sensor receives is energy reflected from a body, and the strength and shape of that return varies with body size, body composition and posture. A very small person and a very large person present different radar cross sections. A person using a wheelchair sits inside a large metal reflector that dominates the scene. A walking frame, an adjustable bed, an oxygen concentrator and a mobility scooter all introduce moving or static metal into rooms where this device is expected to work. Clothing and bedding attenuate differently.

The population this product serves is defined by exactly those characteristics. Assisted living and nursing home residents disproportionately use mobility aids, and the people most likely to fall are the people least likely to move like a model's typical training subject.

Gait and descent patterns compound it. Someone with Parkinsonian movement, hemiparesis after a stroke, or a shuffling gait produces a descent signature unlike a typical fall, and a system that distinguishes falls from deliberate descent by reading body position through the movement is directly exposed to that variation.

Pets are a further confound specific to home deployment and are not addressed.

Nothing published states detection performance for any of these groups, or whether mobility aids and atypical gait were represented in development.

Ask for fall detection performance among residents using wheelchairs and walking frames, and how pets and metal furnishings are handled.

Nothing allocates responsibility for a product whose primary function is detecting an emergency.

A dedicated pass located no service level agreement, no detection warranty, no uptime commitment, no indemnity, no remediation position and no published terms of service for either the consumer or the enterprise offering.

The stakes are immediate rather than downstream. A missed fall in a home where the occupant lives alone produces harm measured in hours on a floor, and the population is specifically people for whom that outcome is severe. No detection rate is published, so neither a family nor a facility can characterise the residual risk they are accepting, and the substitution question is unanswerable: a household replacing a wearable pendant with an ambient sensor has no basis for knowing whether coverage improved or degraded.

The channel makes allocation genuinely unclear rather than merely unstated. A consumer buys from a security dealer, connectivity comes from a carrier, the enterprise platform comes from a third party, and the models come from the vendor. When an alert does not arrive, responsibility could rest with any of them, and nothing in public material describes how that is divided or who the customer's counterparty actually is. A family in that situation would be discovering the answer at the worst possible moment.

One pre emptive note: additional partnerships, awards or channel announcements cannot move this grade. Only a contractual commitment, or published detection and delivery performance for the emergency function, will.

Ask who the contracting party is in a dealer sale, what is committed on alert delivery, and what recourse exists when a detection fails.

A dedicated pass located no electronic health record integration, no interface standard, no published application programming interface, no marketplace listing and no clinical data exchange description.

The integrations that do exist are consumer and security infrastructure rather than clinical. A cellular carrier supplies connectivity, a smart home platform carries the enterprise product to its service provider network, and security dealers install it. Those are distribution and transport relationships. None of them moves a clinical observation into a system where a clinician works.

The consequence is specific and worth stating plainly. This device generates events that would be clinically meaningful in a record, particularly a fall and any longitudinal change in respiration or activity, and nothing indicates that any of it reaches a chart. A fall detected at three in the morning that never appears in the resident's record is available to whoever answered the alert and to nobody afterwards, which removes it from care planning, from incident review and from any analysis of whether the deployment reduced harm.

The care setting makes this more pointed rather than less. Senior living operators run care management and medication administration systems, and none is named.

A senior care solution with a partner was announced at a health technology conference in March 2026 and may represent a clinical path, and nothing published describes what it integrates with or how.

Graded D consistent with the treatment of a total absence of named systems across this index. Ask whether any record system integration exists in production, through what standard, and whether detected falls are written to a resident record.

The architecture is genuinely disclosed at the point where it matters most, and the backend behind it is not.

What is stated is more than a cloud label. Inference runs at the edge on the device, and connectivity is cellular by default with first responder network capability rather than depending on household broadband. Both are real architectural facts with consequences a buyer can reason about. Edge processing means the primary sensing data does not require a round trip to be useful, and cellular default removes the most common failure mode in home deployments, which is an older adult's router or wireless network failing silently. For facilities an Enterprise Hub exists as a distinct deployment form.

That combination materially narrows the residency question, because a device that classifies locally has less reason to move raw data anywhere. It does not answer it.

What is missing is the other end. No cloud or hosting provider is named, no region is stated, no residency commitment is made, no tenancy model is described for the enterprise product, and no retention position is given for whatever reaches the backend. Nothing states whether the vendor, the connectivity carrier or the platform partner operates the receiving infrastructure, which with three parties in the path is not a detail.

The carrier dependence also constrains geography, since a device defaulting to one national network is bounded by that network's footprint.

Ask what is transmitted from the edge and to whose infrastructure, where that sits, and how the enterprise deployment is isolated between facilities.

A real consumer price exists in the public record, reported rather than published, and everything above the consumer tier is undisclosed.

The figures are usable. Reporting at launch put the device at roughly 250 to 300 dollars with a subscription of 39 dollars per month, which gives a prospective family both the capital and running cost and a clear unit. Compared with the quote only wall most vendors in this index present, that is a genuinely more open position and it is graded accordingly.

Three weaknesses keep it from higher. The figures come from press coverage at the January 2024 launch rather than from a current published rate card, so they are two years old against a product that has since gained an enterprise variant. The dealer model means the number a buyer actually pays is set by an installing dealer rather than by the vendor, and installation, activation and monitoring service charges layered by that dealer are not visible anywhere. And the connectivity is cellular by default, which ordinarily carries a recurring cost, with nothing stating whether the subscription includes it.

Enterprise pricing is entirely absent. Facility deployments run through an Enterprise Hub distributed by a platform partner, and nothing indicates whether a nursing home is charged per device, per room, per resident or per building, nor how many devices a floor requires given the multi room coverage claim. That last question is the one that determines cost at scale and it is exactly the one the coverage claim makes impossible to answer from outside.

Ask for a current rate card, what the subscription includes, how many devices a typical facility floor requires, and the enterprise pricing basis.

Settings are broad and end deployments are almost entirely unnamed, which is the reverse of the usual pattern and just as limiting.

The stated range covers private homes for aging in place, senior living facilities, nursing homes and healthcare providers, with an Enterprise Hub existing specifically to serve the facility case. Multi person multi room sensing is a real coverage property rather than a marketing one, because it means a household with a spouse or a shared facility room is within scope where single occupant sensing would not be.

What is missing is the evidence of depth. Named end customers are scarce: a veterans centre placement through a non profit partner is the clearest one, and the rest of the named relationships are channels rather than deployments. Nothing states how many facilities run this, at what scale, or in what configuration.

Clinical scope is narrow by design and should be read that way. Falls, body position, movement, heart rate and respiration is the full set. There is no activity of daily living inference, no condition prediction, and no behavioural trend analysis of the kind the wearable competitor in this lane offers. That narrowness is defensible, since it is what the sensing modality supports, but a buyer comparing across this lane should not assume equivalent breadth.

Nothing addresses markets outside the United States, and the connectivity dependence on a single national carrier is a structural constraint on that.

Ask for facility deployments by name and scale, device count per unit of floor area, and whether any non United States deployment exists.