A signal is observable. The quality it is meant to represent often is not. Good judgment begins by refusing to collapse those two facts into one.
We cannot inspect competence, commitment, trustworthiness, organizational health, or love directly. We observe actions, records, outcomes, and patterns, then infer what may lie behind them. That inference is necessary. It is also provisional.
The problem is not that signals are indirect. Almost all useful measurement is indirect. The problem begins when an indicator, score, or display is treated as the underlying reality without stating the assumptions that connect them.
Every signal contains a claim
A diploma can support a claim about preparation. A fast response can support a claim about availability. A low incident count can support a claim about operational health. None proves the claim by itself. Each observation becomes evidence only through an interpretation: this visible result is expected when the hidden quality is present.
That interpretation may be reasonable without being certain. The diploma may reflect relevant learning, selection effects, persistence, or access to opportunity. The fast response may reflect commitment, anxiety, spare capacity, or an interruption-heavy role. The low incident count may reflect reliability, low usage, narrow detection, or underreporting.
Separating observation from interpretation does not weaken a decision. It makes the decision auditable. “Response time was under ten minutes” is an observation. “The team is committed” is an inference. The second may be justified, but it needs more support than the first.
A proxy has a domain and an error
A proxy is useful when changes in the observable measure carry information about the thing of interest. That relationship has a domain: a population, context, time period, and set of conditions in which the inference was examined. Moving beyond that domain increases uncertainty even when the number remains precise.
Measurement error is not limited to a faulty instrument. The wrong construct can be measured perfectly. A dashboard may count tickets accurately while the decision concerns customer harm. An interview may score verbal fluency consistently while the role depends on diagnosis and execution. Precision in the proxy does not establish validity for the intended claim.
Signals also have base rates and competing causes. A rare condition can produce many false positives even with a useful test. One observed behavior can be consistent with several hidden states. Interpretation improves when the decision maker asks not only “does this signal fit my explanation?” but also “what else would produce the same observation?”
Interpretation changes the evidence
Signals do not arrive with fixed meanings. Their meaning depends on who produced them, who observes them, and what both sides know. A credential carries different information when access, standards, or selection processes differ. Silence can indicate agreement, uncertainty, exclusion, or no opportunity to speak. The receiver supplies context whether or not that work is acknowledged.
Under information asymmetry, signaling can also be strategic. A sender may choose what to display based on the response they expect, while the receiver interprets the display knowing that choice was possible. This does not make the signal false. It changes the inference: credibility depends in part on whether producing the signal is meaningfully connected to the hidden quality.
Once people know a measure will drive a decision, the conditions that made it informative can change. That does not make every measured result fraudulent. It means evidence collected under observation may not have the same interpretation as evidence collected before the measure mattered. The use of a measure becomes part of the measurement environment.
This is an epistemic caution, not a claim that reality is unknowable. Some signals are strongly diagnostic. Repeated performance under relevant conditions can be excellent evidence of capability. The discipline is to match certainty to the quality and independence of the evidence, not to the vividness of the display.
Evidence should survive another view
One signal can justify attention. Consequential decisions usually need convergence. Look for different observations that would fail for different reasons: outcomes over time, direct work samples, records from independent sources, disconfirming cases, and tests under realistic conditions. Repeating the same proxy in several formats is not triangulation.
Define the claim before choosing the measure. State what the measure captures and what it omits. Preserve uncertainty instead of hiding it behind a precise score. Check whether the relationship still holds for the people and conditions in front of you. Seek evidence that could make the conclusion less convenient, not only evidence that confirms it.
A signal remains useful when it helps us update a belief without pretending to settle it. This is the difference between evidence and proof in ordinary judgment. Evidence changes what we should believe. Proof, where it is available, closes a narrower question under stated rules. Most human and organizational decisions operate in the first category.
Key takeaway. A signal is an observation used to infer something that cannot be inspected directly. Treat it as evidence only after naming the claim, the context, the plausible alternative causes, and the uncertainty. The measure can guide a decision without becoming the reality it represents.
Source notes
The formal sources come from measurement and signaling research. Applying them to personal and organizational judgment is this essay’s interpretation, not a claim that every example follows one model.
- Construct validity. Cronbach and Meehl describe how evidence supports interpretation of a measure for an unobserved construct rather than proving the construct directly. Source: “Construct Validity in Psychological Tests”, 1955.
- Signaling. Spence models decisions made from observable education signals when productivity is not directly known. Source: “Job Market Signaling”, 1973.
- Measures under decision pressure. Campbell analyzes how quantitative indicators can become distorted when used for consequential social decisions. Source: “Assessing the Impact of Planned Social Change”, 1979.
Discuss this article
Thoughtful comments, corrections, and notes from real-world practice are welcome. Discussion is managed through GitHub.
This space is intended for meaningful technical discussion, useful corrections, and field experience. Spam, personal attacks, low-effort comments, and vendor pitches may be removed.