World Sleep Industry ENCYCLOPEDIA

Product Efficacy Verification

6.3.2

Definition, objects, and verification boundaries

Product functional verification uses objective evidence to confirm that specified functional requirements have been implemented correctly by particular design outputs. In WSIC this entry belongs to 6.3.2 Third-party testing and evaluation services. It can address adjustment of bedding, thermal control, lighting, acoustics, sensing, alarms, connectivity, applications, cloud services, or the systems formed by these elements, at component, subsystem, and finished-product levels.

Functional verification principally asks whether the product does what its specification requires. It differs from design validation, which determines whether the requirements and product satisfy the right user needs, and from performance evaluation, safety evaluation, reliability testing, and product efficacy research. Starting a function does not show that speed, accuracy, or capacity meets its target; passing every planned test does not eliminate residual risk; and performing a programmed action does not demonstrate improvement in insomnia, sleep architecture, or health outcomes.

The word efficacy in the catalog's English label should not be interpreted as a guarantee of clinical efficacy; this entry follows the scope of its normative Chinese title, product functional verification. A conclusion applies only to the identified model, hardware revision, software and algorithm version, configuration, accessories, interfaces, data, and test environment. Laboratory, third-party, or automated testing status cannot turn evidence from limited samples into certification, regulatory authorization, or a guarantee for every use condition.

Requirements baseline, verifiability, and traceability

Verification begins with an approved requirements baseline. Each requirement should have a unique identifier and define the object, trigger, input, required behavior, output, timing, tolerance, abnormal response, and acceptance criteria, while remaining related to intended use, user needs, risk controls, and interface constraints. Phrases such as supports smart adjustment, accurately identifies sleep, or operates stably are too vague and must be decomposed into observable, measurable, and unambiguous statements.

Requirements verification and product functional verification should be distinguished. The former examines whether requirements are complete, consistent, feasible, and well formed; the latter examines whether the implementation satisfies them. Missing or conflicting requirements, or acceptance criteria written after results are seen, cannot support a reliable conformity decision even when the test execution itself is flawless. Regulatory, contractual, and product-specific standards should become verification inputs only after their applicability, edition, and relevant provisions are established.

A traceability matrix should connect user needs, system requirements, subsystem or software requirements, design outputs, risk controls, test cases, results, defects, and changes. Bidirectional traceability should show both that every requirement is covered and that every designed function has an authorized source, preventing omissions and unapproved extra functionality. One case may address several requirements, but each still needs its own disposition; an overall pass rate must not conceal an unverified critical requirement.

Verification strategy, levels, and case design

The verification plan should use a combination of review, analysis, inspection, demonstration, and testing selected for the architecture and risk instead of assigning every question to one finished-product test. Component-level work isolates logic and interfaces early, integration work examines data, control, and timing interactions, and system-level work confirms end-to-end behavior, with independent review where justified. Evidence from a lower level can be reused upward only when configurations and interfaces are shown to be equivalent.

Cases should be derived from requirements, state models, interfaces, use scenarios, historical defects, and risk analysis, and should include normal flow, boundary values, equivalence classes, state transitions, combinations, and negative testing. Coverage can reveal whether a requirement, code element, branch, state, or interface was exercised, but it does not prove that defects are absent. A percentage is meaningful only when the covered object, calculation basis, and rationale for omissions are reported.

Each case should define prerequisites, test data, steps, expected results, decision method, reset conditions, and evidence requirements in advance. Priority can reflect functional criticality, change scope, failure consequence, complexity, and previous problems, but low-risk functions should not remain uncovered indefinitely. Exploratory testing can reveal unforeseen behavior, yet it cannot replace reproducible and auditable verification against specified requirements.

Coordination of hardware, software, data, and interfaces

Sleep products often deliver one function through sensors, actuators, power systems, firmware, mobile applications, wireless networks, cloud services, and third-party interfaces. Separate passage of every component therefore does not establish end-to-end function. Verification should examine data units, ranges, timestamps, identity, protocol, buffering, synchronization, command acknowledgement, and error propagation, including startup, pairing, update, disconnection, reconnection, power loss, and recovery.

Software verification needs a fixed build, dependencies, configuration, feature flags, execution platform, and test data, and should distinguish static review, unit, integration, system, and acceptance evidence. Automated scripts require version control and checks against known outcomes, while differences between simulators, stubs, replay data, and real hardware should be recorded. A nominal pass can still be invalid evidence when logs are missing, the time base is wrong, or expected results are generated by logic that shares the implementation's defect.

Functions using rules, statistical models, or machine learning also require fixed model, parameter, training-data, and evaluation-data versions. They should be challenged with missing, out-of-range, shifted, and low-quality inputs and with conditions in which no valid result can be produced. Average correctness does not describe every scenario or population and cannot replace review of error types, abstention, and degraded modes. Privacy and cybersecurity need their own professional assessments, but authentication, authorization, data integrity, or update mechanisms must be included when they are functional requirements.

User tasks, environments, and abnormal paths

User-facing functions should be verified for specified users, goals, tasks, equipment, and physical and social environments. Installation, wearing, setup, nighttime interaction, comprehension of prompts, cleaning, maintenance, and switching among users may determine whether a function is actually usable. Completion by a developer does not demonstrate completion by the target user. Subjective satisfaction and ease of use can contribute relevant evidence but cannot replace an explicit functional decision.

Conditions should include applicable temperature, humidity, light, noise, load, posture, power, battery state, network quality, operating system, permissions, and concurrent activity, distinguishing intended use from justified boundaries or stress conditions. Acceleration, simulation, and controlled benches aid reproducibility, but extrapolation to whole-night, long-term, household, or institutional use needs additional support. An untested condition should be identified as a scope limitation rather than assumed to pass.

Abnormal paths should cover invalid input, foreseeable operating error, timeout, resource exhaustion, detached sensors, obstructed actuators, communication loss, unavailable services, failed updates, and corrupted data, while examining alarms, degradation, safe states, retry, recovery, and records. Foreseeable misuse does not require support for every improper action, but the product should not silently produce a plausible yet wrong output. After recovery, state, clock, settings, and unfinished operations should be checked for silent loss or duplicate execution.

Execution control, defect handling, and conclusions

Before execution, the team should confirm sample identity, versions, calibration status, test environment, dataset, personnel authorization, and plan approval, retaining enough information for another competent person to reproduce the work. The record should preserve actual results, time, operator, raw logs, screenshots or instrument output, and every deviation. A checked pass box without an evidence chain cannot support an auditable judgment.

When an actual result differs from expectation, a defect record should assess severity, affected scope, root cause, and related requirements before deciding to correct, defer, restrict, or accept it. The original failure and potentially affected neighboring functions both need retesting after a correction. Labels such as intermittent, user issue, or not reproducible are not by themselves grounds for closure. Deviations, temporary workarounds, and unresolved anomalies must enter the final review.

The conclusion should distinguish pass, fail, blocked, not tested, and not applicable for each item and identify versions, coverage, limitations, unresolved defects, and approval responsibility. Pass percentage, case count, or automation ratio should not alone determine release. A third-party report can improve independence, but the responsible product organization still has to decide whether the evidence supports the intended markets, uses, risk controls, and external claims.

Change, regression, and life-cycle maintenance

Changes to requirements, materials, components, suppliers, firmware, algorithms, applications, cloud services, operating systems, interfaces, configuration, or production processes can alter existing functions. Change control should assess impact first, update traceability, and then determine new, repeated, and regression cases. A supposedly minor edit can affect remote behavior through shared resources, timing, permissions, or dependencies.

Regression verification should select scope according to impact and risk while retaining stable critical end-to-end cases and cases for corrected defects. Automation suits frequent, deterministic, and repeatable checks, but assertions, data, environment, and failure diagnostics require maintenance; script count is not a substitute for test quality. When platforms, simulators, or reference data change, the verification toolchain should first be shown to continue producing trustworthy evidence.

Production sampling, final factory testing, installation acceptance, and post-market monitoring answer different questions and cannot replace complete development verification. Complaints, repairs, telemetry, update failures, and field anomalies should feed back into requirements, cases, or risk inputs. When new evidence invalidates an earlier conclusion, claims should be restricted, the product corrected, and verification repeated. This entry can be read together with “performance evaluation,” “safety evaluation,” “product reliability testing,” “product efficacy validation,” and “third-party product testing.”