Evidence checked October 31, 2026. This article evaluates research claims, not an individual's diagnosis or treatment. Numerical examples are hypothetical. Historical findings are dated and are not presented as new October 2026 discoveries.
1. A promising signal and a diagnostic test are different achievements
CDC's Long COVID clinical guidance, dated March 9, 2026 and checked for this article, states that no laboratory test can definitively diagnose or rule out Long COVID. It also says that laboratory and imaging findings should not be the sole measure of a patient's well-being. The guidance provides an essential frame for interpreting biomarker headlines: the absence of a definitive test is not evidence that the condition or the person's symptoms are unreal.
A biomarker study may identify a measurable biological difference between groups. That can be important progress. It can point toward mechanisms, help organize heterogeneous illness, or suggest a candidate measurement for further development. Yet a difference discovered in a study does not automatically become a test capable of making an accurate clinical distinction for a new patient.
Long COVID makes that distinction particularly consequential. People can have different symptoms, trajectories, infection histories, and pre-existing conditions. A marker associated with one subgroup might be valuable without being present in everyone. A marker absent in one person might say little about whether that person has the broader condition. The scientific task is to establish the marker's role rather than allowing the word biomarker to imply a universal answer.
This article examines that task through dated primary research and institutional definitions. It does not rank commercial tests or propose treatment. The aim is to show what a strong claim would need to establish, how apparently conflicting studies can be understood, and why biological discovery and clinical validation should support one another without being treated as the same stage of evidence.
2. The word biomarker needs a job description
A biomarker can serve several purposes. It might help identify a condition, estimate future risk, track a process over time, identify a subgroup likely to respond differently to an intervention, or measure a biological response. Success in one role does not establish success in another. A study must specify the question before the marker's performance can be judged.
The FDA's overview, drawing on the FDA-NIH BEST terminology, distinguishes diagnostic, prognostic, predictive, monitoring, response, safety, and susceptibility categories. It also emphasizes a defined context of use. The FDA explanation makes a useful distinction between a biological measurement and a direct assessment of how someone feels or functions.
For a hypothetical Long COVID study, a protein pattern associated with persistent symptoms could be investigated as a diagnostic candidate. If measured soon after infection, the same pattern might instead be studied for its ability to predict later illness. If it changes during recovery, it might be investigated as a monitoring candidate. Those studies require different timing, comparison groups, and outcomes.
The strongest headline names the role actually tested. Identifying a potential biological pathway is not equivalent to predicting recovery. Predicting recovery is not equivalent to selecting an effective treatment. And a marker that changes after an intervention is not automatically evidence that the intervention improves patients' lives. Each step needs its own bridge between measurement and meaning. Precision about the job prevents a promising result from being asked to do work that the study never evaluated.
3. A case definition shapes the signal
Before researchers can compare people with and without Long COVID, they need a way to define the groups. Different definitions may emphasize duration, symptoms, functional effects, infection history, or exclusions. Those choices can change who enters the analysis and therefore which biological differences are visible.
WHO's October 6, 2021 consensus definition described post-COVID-19 condition in people with probable or confirmed infection, usually present three months after COVID-19 onset, with symptoms lasting at least two months and not explained by an alternative diagnosis. The WHO definition also recognized fluctuating or relapsing symptoms. Symptoms can persist from the initial illness or appear after initial recovery. It is important to retain its date rather than presenting it as a new rule issued for the current study.
The 2024 National Academies definition describes an infection-associated chronic condition present for at least three months that may be continuous, relapsing and remitting, or progressive and affect one or more organ systems. Its definition summary does not require laboratory proof of the initial infection. These formulations share important features but should not be assumed to create identical research samples in every implementation.
For an original analytical example, imagine two studies using different symptom thresholds. One recruits people with severe multisystem impairment; the other includes a wider range of persistent symptoms. A marker concentrated in the severe subgroup may appear stronger in the first study. That difference need not mean either study is fraudulent or useless. It may reveal that the marker's association depends on the population being defined.
4. A research index is not a boundary around legitimate illness
Researchers sometimes use a symptom-based index to create reproducible categories. Such an index can make large studies more consistent and allow associations to be examined systematically. It is not necessarily intended to determine whether every individual does or does not have Long COVID in clinical care.
The 2024 update of the RECOVER-Adult research index explicitly describes a classification for research within the broader National Academies definition. The authors state that people who do not meet the index threshold may still have Long COVID. The primary paper therefore supplies an important guardrail: a research cutoff should not be converted into a universal test of whether a person's illness counts.
A hypothetical biomarker analysis using an index can ask whether a measurement distinguishes high-index participants from a comparison group. That is a defined and potentially useful question. It does not automatically establish performance across all clinical presentations outside those groups. The article reporting the biomarker should preserve that distinction rather than shortening the result to a universal Long COVID detector.
Changes in the index over time also matter. A later version may include additional information or classify some participants differently. Comparing studies then requires knowing which version was used. A difference in biomarker performance could reflect a changed target definition as well as a changed assay or cohort. Transparent reporting allows those possibilities to be investigated. It also protects participants from the mistaken implication that a research classification is a verdict on the reality or severity of symptoms that fall outside its selected pattern.
5. The control group determines what is being distinguished
A study can compare Long COVID participants with people who were never known to be infected, people who recovered without persistent symptoms, or people with other illnesses producing similar symptoms. Each comparison asks a different question. A marker that separates one pair of groups may not separate another.
Consider a hypothetical immune signal elevated after infection in both people with Long COVID and people who recovered. Comparing Long COVID participants only with uninfected controls could identify the signal as associated with the illness group. Adding recovered infected controls would reveal that the signal may reflect infection history rather than persistent symptoms specifically.
A second hypothetical marker might distinguish Long COVID from healthy recovery but also appear in other chronic inflammatory conditions. It could still be biologically informative. Its usefulness as a diagnostic test in people presenting with similar symptoms would require a more demanding comparison. Healthy controls help establish a difference; clinically relevant alternative conditions help establish specificity for the proposed use.
No single control group answers every question. A mechanistic study may deliberately choose a narrow comparison to isolate one process. The problem arises when the result is generalized beyond that comparison. Readers should ask who the controls were, how they were recruited, and whether their sampling conditions matched those of the cases. The label control is not a guarantee of comparability. It is a description of a role in the design, and that role must fit the inference the researchers want to make.
6. Immune profiling found differences, not a finished diagnostic service
Klein and colleagues' 2023 immune-profiling study enrolled people with and without Long COVID and examined immune features, antibodies, and hormones. The study reported group differences, including lower cortisol levels among participants with Long COVID, and used computational models to identify features associated with group status. The primary paper presents these findings as a basis for further investigation and biomarker development, while describing limitations in recruitment and sample size for modeling.
The scientific value is that the study generated specific biological hypotheses. It did not merely restate that participants felt unwell. But translating the result into a diagnostic claim requires additional work: independent evaluation, reproducible measurement, a defined operating rule, and testing against the relevant alternatives in the intended population.
It would also be a mistake to turn the observed hormone difference into a treatment instruction. A cross-sectional association does not establish that changing the measured quantity will improve the condition. The difference could occupy several positions in a causal chain, and an intervention can have effects beyond the measured marker. This article does not recommend hormone testing or treatment for Long COVID.
The appropriate interpretation preserves both sides of the evidence. The observed differences deserve investigation; the limits deserve equal visibility. Calling the study promising is justified when the promise is specified: better hypotheses about biology and candidate measurements to validate. Calling it a definitive blood test would claim a clinical capability that requires a different body of evidence.
7. Longitudinal proteomics can sharpen a hypothesis without proving causation
A 2024 study by Cervia-Hasler and colleagues examined blood proteins in participants followed after SARS-CoV-2 infection. It reported complement-system dysregulation and signs associated with thromboinflammation in active Long COVID. The authors' abstract describes these findings as grounds for further diagnostic and therapeutic investigation. That wording matters: a candidate pathway is not already a validated treatment target or a universal diagnostic signature.
Repeated observations over time can be more informative than a single snapshot. They may help show whether a signal accompanies persistence, changes with recovery, or appears before a later outcome. But temporal association alone does not settle every causal question. Other processes can influence both the marker and the symptoms, and participants who remain in follow-up may differ from those who do not.
A hypothetical marker that falls as symptoms improve could be part of the mechanism, a downstream response, or a correlated indicator of another process. These possibilities are scientifically different even if the plotted lines move together. The next study should be designed to discriminate among them rather than treating correlation over time as proof of one explanation.
The distinction is especially important for public discussion of blood-related findings. A research association should not become advice to use an unproven intervention. Demonstrating a biological pattern, showing that the pattern contributes causally, and showing that modifying it safely improves outcomes are separate achievements. The first can be valuable while the other two remain open.
8. A negative routine-laboratory study narrows one route
Erlandson and colleagues examined twenty-five routine clinical laboratory measures in the RECOVER cohort. Their 2024 paper found no evidence that the assessed values could serve as clinically useful biomarkers of post-acute sequelae in that analysis. The primary study also described important limits, including the lack of pre-infection results for most participants and the fact that it did not evaluate multivariate laboratory profiles.
That result should not be read as evidence that Long COVID has no biological basis. It concerns a defined set of routine measurements, a study population, timing, and research classification. A negative result can show that a proposed measurement strategy is insufficient without invalidating the condition it was intended to measure.
Nor does the result prove that every possible laboratory panel will fail. Specialized research assays may measure different processes. The appropriate conclusion is narrower: these routine values, evaluated in this way, did not provide the desired biomarker. That information can prevent overinterpretation of common tests and help focus future work.
An original analogy within the research problem is useful. If a study measures one compartment of a complex system and does not find a discriminating signal, it has learned something about that measurement. It has not measured every process in every tissue. The next question is which alternative measurement is justified by a specific hypothesis, not whether a broader search can be advertised as guaranteed to succeed. Negative evidence is most useful when its scope is retained precisely.
9. The time of sampling is part of the biological question
Long COVID can evolve, fluctuate, or relapse. A sample taken at one point may capture a different state from a sample taken months later. Studies that combine a wide range of times since infection can therefore average across processes that are not constant. The interval between infection, symptom onset, and specimen collection belongs in the interpretation.
Consider a hypothetical marker elevated early after infection in many people but persisting only in some. A study sampling everyone at a later stage could identify a different association from a study sampling soon after the acute illness. Neither pattern can be understood without the timeline. A marker might be useful for prognosis at one stage and unhelpful for diagnosis at another.
Symptoms and samples should also be linked closely enough to support the intended claim. If a blood draw is compared with a symptom assessment from a substantially different period, the study may be relating different states. That may be appropriate for prediction, but it should not be described as simultaneous correspondence. The timing defines the question.
Repeated measurements introduce another issue: observations from the same person are related. A study with many samples may have fewer independent participants than the sample count suggests. The analysis should preserve that relationship rather than treating every tube as a new person. Longitudinal data are powerful because they reveal within-person change, not because they manufacture independent people. Their value depends on distinguishing those two forms of information.
10. A pre-infection baseline can change the explanation
A biological difference observed after infection may have existed before infection, emerged during acute illness, or developed alongside persistent symptoms. Without earlier measurements, those explanations can be difficult to separate. A post-infection association alone cannot establish the direction of causation.
Imagine a hypothetical study finding higher levels of a molecule among participants with Long COVID. If the molecule was already higher before infection, it might be investigated as a susceptibility marker or a correlate of another risk factor. If it rises after symptoms develop, it might reflect a consequence or a contemporaneous process. If it changes before the persistent illness becomes apparent, it could support a different predictive hypothesis. The same cross-sectional result cannot distinguish all three histories.
Pre-infection data are not always available, and their absence does not make a study worthless. It defines the inference the study can support. Researchers can use other designs and evidence to investigate competing explanations, but the limitations should remain explicit. The language associated with, predicts, and causes should not be treated as interchangeable.
Baseline measurements also have their own conditions. A result collected for an unrelated clinical reason may not be directly comparable with a later research assay. Timing, platform, sample handling, and the participant's state can differ. The mere existence of an older number is not enough; its comparability needs evaluation. A useful baseline is an observation with a known relationship to the later measurement, not simply an earlier entry in a medical record.
11. Heterogeneity can weaken an average and strengthen a subgroup question
A broad condition can contain several biological pathways or symptom patterns. If a marker is relevant to only one subgroup, averaging across everyone may dilute its association. That does not make subgroup discovery easy: searching many possible partitions can also produce patterns that fail to reproduce. Both problems must be considered together.
A hypothetical study might find that a marker is elevated in participants with one cluster of symptoms but not in others. This could identify a useful research subgroup. It could also reflect a small-sample pattern, a correlated medication, or a selection effect. The subgroup's definition should be tested independently rather than adjusted repeatedly until the marker looks convincing.
A biomarker for a subgroup should be described as such. It need not identify every person with Long COVID to have scientific or eventual clinical value. But a negative result outside that subgroup should not be used to exclude the broader condition. The marker's scope and the condition's scope are different boundaries.
The most productive next step may be to specify the subgroup prospectively and evaluate whether the marker predicts a relevant outcome within it. That is more informative than repeatedly dividing the same dataset in search of the strongest association. Heterogeneity is a reason for careful design, not a license to explain every failed replication after the fact. A credible subgroup hypothesis makes a prediction that new data could prove wrong.
12. A machine-learning signature can learn the study rather than the illness
Biomarker studies may measure many variables and use machine learning to combine them. This can reveal patterns that are difficult to detect one variable at a time. It also creates opportunities for overfitting, especially when the number of independent participants is limited relative to the flexibility of the analysis.
A hypothetical model might distinguish cases and controls because their samples were processed in different batches. If batch-related features correlate with the labels, the model can achieve impressive classification without learning a generalizable illness signal. The problem may be invisible in a random split that preserves the same batch pattern on both sides.
Feature selection can create another source of optimism. If the full dataset is used to choose the most promising molecules before cross-validation, information from the evaluation observations can influence the model indirectly. A valid evaluation needs the entire selection process contained within the development procedure or tested in an untouched external dataset. Reporting only the final classifier can hide those earlier choices.
The essential question is what the model would do with samples from new people collected and processed independently. A high internal score is useful development evidence, but it is not the same as external validation. The report should identify the unit of splitting, the role of repeated samples, the handling of batches, and whether the final signature was fixed before evaluation. A model becomes clinically interesting when it survives a comparison it did not help design around its own answers.
13. A blood measurement has a pre-analytical history
The biological meaning of a measurement depends partly on how the specimen was obtained, handled, stored, and analyzed. A candidate marker can appear inconsistent across studies because the populations differ, because the measurement differs, or because both differ. Those explanations require different responses.
A hypothetical research assay might detect a group difference under a tightly controlled collection schedule. Before it can be used more broadly, investigators need to know whether the result remains interpretable under the conditions of the intended service. If timing or handling materially affects the value, the workflow must account for that dependence. Otherwise, technical variation may be mistaken for biological change.
This does not mean every study must use identical procedures forever. Methods improve. But comparisons should state how the measurements relate and whether a change in platform alters the scale or threshold. A marker's name is not always a complete specification of the assay. Different methods can target related but not identical features.
For Long COVID research, careful specimen records also help explain apparent disagreement. A failed replication may reveal that a biological association is weaker than first thought; it may reveal a measurement mismatch; or it may reveal a context-specific signal. Investigators should test these explanations rather than automatically dismissing the replication or defending the original result. The aim is to learn whether the marker can be measured reliably enough for its proposed purpose, not to preserve a preferred finding at the expense of comparability.
14. A marker can be biologically relevant and diagnostically nonspecific
A biological process can contribute to several conditions. Finding evidence of that process in Long COVID may be important without making the measurement unique to Long COVID. Diagnostic specificity is a different question from mechanistic relevance.
A hypothetical inflammatory marker might be elevated in a subgroup of Long COVID and in several other illnesses. It could help characterize an immune state or guide research into shared pathways. It would not, by itself, distinguish the cause of similar symptoms in a new patient. The next evaluation needs comparison groups that reflect the alternatives encountered in the intended setting.
Conversely, a marker that is highly specific to one exposure history may not track the severity or persistence of illness. It could distinguish prior infection while failing to distinguish people who recovered from those who did not. The interpretation depends on which biological relationship the marker captures.
These distinctions make the control-group question more than a methodological detail. They determine what the result means. A study can legitimately discover a shared process, an exposure-associated signal, a subgroup marker, or a candidate diagnostic signature. The problem is describing all four as proof of a definitive Long COVID test. Clear categories allow research to accumulate without forcing every useful finding into the narrowest and most demanding clinical role. A marker does not have to be unique to be informative, but its lack of uniqueness must be considered before it is used to assign a diagnosis.
15. Changes in a biomarker are not automatically changes in well-being
A laboratory measurement and a patient's experience can move together, move separately, or change on different timescales. A study evaluating a biomarker should therefore examine what relationship it has to symptoms, function, and other outcomes relevant to the intended use. A change in the number alone does not establish a meaningful improvement or deterioration.
Imagine a hypothetical intervention study in which a candidate marker decreases while participants report no improvement in daily function. The result could show a biological effect of the intervention without establishing clinical benefit. Alternatively, participants might improve while the marker remains unchanged, suggesting that the marker does not capture the relevant pathway or outcome. Neither pattern should be concealed by selecting only the favorable endpoint.
This is why a surrogate-endpoint claim requires more than an association. A marker proposed to stand in for patient benefit must be shown to support that interpretation in the relevant context. It cannot inherit that status merely because it is objective, measurable, or mechanistically appealing.
For Long COVID, preserving patient-centered outcomes is particularly important because the condition can affect activities that a routine laboratory panel does not represent. A research program should not treat the absence of a measured abnormality as the absence of impairment. Nor should it treat normalization of a research marker as proof that the person has recovered. The scientific goal is to understand the relationship between biological processes and lived outcomes, not to replace one with the other before that relationship has been established.
16. Recruitment can exclude the people most affected
Participation in a biomarker study often requires travel, appointments, repeated sampling, and sustained engagement. Those requirements can shape who is able to enroll and remain in follow-up. If people with greater impairment face greater barriers, the study may underrepresent the part of the condition it most needs to understand.
A hypothetical clinic-based cohort might include people able to attend an academic center while missing those who are housebound, live far away, or lack resources for repeated visits. The resulting sample can still produce valuable findings. It should not be assumed to represent the full range of illness severity and circumstances without examining those access patterns.
Loss to follow-up creates a related uncertainty. Participants who improve may stop attending; participants who worsen may become unable to attend; others may leave for unrelated reasons. Each pattern can alter the observed relationship between a marker and recovery. A longitudinal graph of the people who remain is not automatically a graph of the original cohort's experience.
The report should describe recruitment routes, participation requirements, and missing follow-up. Where feasible, researchers can compare observed characteristics of those retained and those lost, while acknowledging what remains unknown. This is not merely a fairness addendum. Selection affects the biological inference. A marker evaluated in a narrow, accessible sample may behave differently in people with other trajectories. Understanding that possibility helps define the next validation study and prevents an apparently universal conclusion from resting on a restricted view of the condition.
17. Infection history is an imperfect research variable
Studies often classify participants by prior SARS-CoV-2 infection, but infection history may not be documented equally across people or time periods. Testing access, asymptomatic infection, and incomplete records can create uncertainty. A group labeled without known infection is not necessarily a group proven never to have been infected.
The National Academies definition's lack of a requirement for laboratory proof of the initial infection reflects the distinction between clinical recognition and a narrow documentation rule. In research, the important task is to describe how infection history was established and how uncertainty was handled. Treating uncertain history as certain can distort comparisons.
A hypothetical control group could include people with unrecognized prior infection. If a marker reflects infection exposure rather than persistent illness, that misclassification could weaken or alter the observed association. The direction and size of the effect depend on the design; it should not be assumed to explain every negative finding.
Reinfection adds another layer. A new exposure can complicate the timeline between the initial illness, persistent symptoms, and the measurement. Studies should state how recent infections were identified and incorporated into the analysis. The goal is not to demand impossible certainty from participants. It is to align the strength of the conclusion with the certainty of the underlying history. A careful study can work with imperfect information, but it must not hide that imperfection behind a binary label that appears more definitive than the evidence.
18. Replication should preserve the claim while changing the opportunity for error
An independent study is most informative when it tests a clearly specified claim using new participants and a procedure that makes the result comparable. If the marker, threshold, outcome, and population are all changed simultaneously, disagreement can be difficult to interpret. Replication needs enough continuity to test the claim and enough independence to challenge its original failure modes.
A hypothetical validation could use the original fixed signature in a new cohort, with analysts unaware of the clinical labels until the procedure is complete. That design asks whether the signature transfers. A different study that searches the new cohort for another favorable signature may produce useful discovery, but it is not the same test of the original claim.
Replication also benefits from reporting failures and inconclusive results. If only the best-performing subsets appear in publications, the evidence base can become more confident than the total research experience warrants. A marker that works in one subgroup and not another should be described in those terms, including uncertainty, rather than presented only through the successful comparison.
The broader aim is cumulative knowledge. Which parts of the signal are robust? Which depend on timing, phenotype, or assay? Which disappear when a confounder is addressed? These questions can turn apparently inconsistent studies into a more precise account of biology. They require a willingness to narrow claims as well as expand them. A smaller, reproducible claim can be more useful than a sweeping association that cannot survive a new opportunity to be wrong.
19. A useful biomarker need not become a universal blood test
The public discussion often treats one outcome as the measure of success: a simple test that identifies every person with Long COVID. Research could produce valuable tools before, or without, achieving that goal. A marker might help characterize a subgroup, support trial enrollment, track a defined process, or improve understanding of a symptom pattern.
Those uses still require validation, but the required evidence should match the purpose. A research-enrichment marker may be judged by whether it identifies a population relevant to a particular hypothesis. A monitoring marker needs evidence about within-person change. A treatment-selection marker needs evidence that response differs in a way the marker can identify. None can be assumed from a single cross-sectional group difference.
This more specific view can reduce both hype and disappointment. A finding does not fail merely because it does not solve all diagnosis. Nor should it be advertised as clinically ready because it has some research value. The useful middle ground is a defined capability with known limits and a clear next validation step.
For people living with the condition, scientific precision should not sound like withdrawal of recognition. The research question is how to measure and understand a complex illness better, not whether a person deserves to be taken seriously before a molecule is identified. A responsible account can acknowledge unmet needs while explaining why candidate tests require scrutiny. The urgency of the need strengthens the case for reliable evidence; it does not make an unreliable shortcut more helpful.
20. Recovery needs an outcome definition of its own
A study that asks whether a biomarker predicts recovery must define recovery carefully. Improvement in one symptom, return to a previous level of activity, absence of a research-index threshold, and a participant's own assessment of recovery are not necessarily the same outcome. A marker can appear predictive or uninformative depending on which endpoint is chosen.
Imagine a hypothetical participant whose breathing improves while cognitive symptoms remain disabling. A study centered on respiratory measures might classify substantial improvement. A broader functional outcome might show continuing impairment. Neither observation has to be wrong, but calling both complete recovery would erase an important distinction. The biomarker's interpretation should track the endpoint actually measured.
A single follow-up visit may also miss a fluctuating course. If a participant feels better on the assessment day but experiences recurrent limitations afterward, a one-time classification may not represent sustained recovery. The appropriate observation window depends on the research question. The report should state whether it concerns short-term change, sustained improvement, or another trajectory rather than using recovery as an undefined endpoint.
Baseline severity matters in this comparison. A large improvement from severe impairment can leave someone substantially affected, while a small change near a cutoff can alter a binary classification. An analysis that uses only the cutoff may obscure both experiences. Examining the underlying measures and their changes can help explain what the prediction means in practical terms.
Finally, a prognostic marker should be evaluated using information available at the time the prediction is supposed to be made. Including later measurements that already reflect the outcome can make a model look predictive when it is partly describing what has happened. The timeline should be explicit: when the marker was measured, when the prediction was fixed, and when the outcome was assessed. That sequence distinguishes forecasting a future course from recognizing a current state. For a condition with varied trajectories, the distinction is central to whether a biomarker can genuinely help anticipate what comes next.
21. Read the next biomarker headline as a chain of claims
The next announcement may report a protein panel, an immune-cell pattern, a metabolic signature, or another measurable feature. To understand it, separate the claims. Was a group difference found? Was a fixed classifier tested in new people? Was the comparison clinically relevant? Was the measurement reproducible? Did using the result improve a decision or an outcome? These are stages of evidence, not interchangeable descriptions of one result.
The studies reviewed here show why both positive and negative findings matter. Immune profiling and proteomic research can generate specific biological hypotheses. Routine-laboratory studies can show that familiar measurements do not provide the desired distinction. Research definitions can become more reproducible while remaining narrower than clinical recognition. Each contribution advances a different part of the problem.
As of the CDC guidance checked on October 31, 2026, a definitive laboratory test for diagnosing or ruling out Long COVID was not established in that guidance. That statement should coexist with serious attention to biological research and patient experience. It is not a claim that nothing is known, nor a prediction that a useful marker cannot emerge.
The most convincing progress will make the links between measurement, population, mechanism, and purpose clearer. It will also report the people for whom the proposed measurement remains uninformative, rather than letting an average conceal important gaps. A biomarker earns its role through evidence about what it can establish for new people under defined conditions. Until that role is demonstrated, a promising signal should remain a promising signal: worth investigating carefully, neither dismissed nor promoted beyond the work it has actually done.