The three layers answer different questions
Evidence asks: what was observed under this source’s actual methods? Interpretation asks: what claim does that observation support, how directly, and with what uncertainty? Community signals ask: what are people reporting in lived or research-adjacent settings that may deserve closer attention?
None of those questions is lesser than the others. Confusion begins when an answer moves into a different layer without being relabeled—for example, when a measured biomarker becomes a promised outcome, or when several personal reports become an estimated rate.
The evidence layer preserves the original record
Relay’s evidence layer begins with the source: study, registry record, laboratory report, regulatory document, or other identifiable record. It keeps the population or model, exposure, comparator, outcome, timeframe, methods, and reported result close to the claim.
A result remains bounded to what was measured. A cell experiment is evidence about that model. A randomized trial is evidence about its assigned comparison and measured outcomes. A database analysis is evidence generated from its data and design. The source type matters, but no label replaces reading the actual question and methods.
The interpretation layer makes the reasoning visible
Interpretation is the bridge between a source result and a plain-language conclusion. It considers design fit, risk of bias, directness, precision, consistency, missing evidence, and whether the proposed conclusion travels beyond the studied population, exposure, outcome, or timeframe.
That bridge should be inspectable. Relay therefore separates what the source reported from what Relay concludes, uses calibrated language, names important limitations, and links back to the supporting record. Interpretation may change when better evidence arrives; the historical source result does not change with it.
The community-signal layer preserves observations without overclaiming
A community report can contain information that formal research did not measure: an unexpected experience, timing pattern, usability problem, or outcome that matters to participants. Structured reports can reveal repeated themes and generate questions worth investigating.
But the reporting process has its own boundaries. Relay may not know how many people had no event, who chose not to report, whether duplicate or incomplete reports exist, what else changed, or whether the reported exposure matched its label. Without a denominator and an appropriate comparison, report counts cannot establish incidence, relative risk, efficacy, or causation.
Agreement across layers strengthens a question—not every conclusion
Suppose a mechanism is plausible, a study reports a related measured change, and community members describe a similar experience. Those pieces may converge on a useful hypothesis. They do not automatically support the same claim because each may involve a different population, measurement, exposure, timeframe, or source of bias.
Relay treats convergence as a reason to inspect the claim more closely: Are the outcomes actually the same? Is the timing compatible? Are independent sources repeating the finding? Is there direct human evidence? Triangulation can strengthen confidence when differently limited sources point toward the same bounded conclusion, but it is not vote counting.
Disagreement is information Relay should preserve
A trial may show little average change while a subset of community reports describes a strong experience. An animal model may predict an effect that human studies do not reproduce. Two credible studies may disagree. Flattening those differences into one score would hide the most useful question.
Disagreement can reflect chance, bias, measurement differences, population or exposure differences, selective reporting, inadequate power, or genuine variation in response. Relay should show where the records diverge, identify explanations supported by evidence, and label the remainder unresolved instead of selecting the most appealing story.
Updates move through the layers at different speeds
A new report can update the community-signal layer immediately as a new observation. It should not automatically change an efficacy or safety conclusion. A new study can update the evidence record, but its effect on interpretation depends on its relevance, quality, and relationship to the existing body of evidence.
This is why Relay attaches provenance and review dates to its conclusions. A source can remain accurately summarized even when the broader interpretation changes. A signal can remain visible after later evidence weakens its proposed explanation. Good updating preserves the history of what was observed while revising only the claims that new evidence justifies.
Key terms
- Evidence record
- A traceable source and its relevant methods, measurements, results, and limitations.
- Interpretation
- A reasoned, claim-specific explanation of what evidence supports and how uncertain or transferable that conclusion is.
- Community signal
- One or more structured reports that suggest a pattern or question worth evaluating; a signal is not a confirmed causal conclusion.
- Provenance
- The traceable origin and history of a source, report, transformation, or conclusion.
- Denominator
- The total population or exposure count from which a rate or frequency is calculated.
- Triangulation
- Comparing differently designed sources to test whether they converge on the same carefully defined claim despite different limitations.
Relay diagram
Connected by provenance, separated by claim
What this can—and cannot—tell us
What it can tell us
- What a specific source measured and reported under its actual design.
- Which plain-language claims are directly supported, indirectly supported, or still uncertain.
- Where an interpretation depends on assumptions beyond the source record.
- What experiences or patterns community members are reporting.
- Whether different evidence streams converge, diverge, or address different outcomes.
- What new evidence changed a conclusion and when the conclusion was reviewed.
What it cannot establish alone
- That every cited source supports every nearby claim.
- That a plausible explanation is the same as a demonstrated outcome.
- That repeated reports establish causation, incidence, efficacy, or safety.
- That agreement across different layers removes each layer’s limitations.
- That one overall score can faithfully replace population, outcome, timeframe, and uncertainty details.
- That absence of a signal, study, or documented report proves absence of an effect or harm.
Go deeperOptional · about 2 minutes
Signal detection is the beginning of evaluation
Formal pharmacovigilance systems use reports and other data to detect possible new safety information. A signal is then validated, prioritized, assessed, and—when warranted—investigated with additional data. The initial signal is valuable precisely because it can direct attention before the causal explanation is settled.
Relay borrows the discipline, not the regulatory authority. Community Intelligence can surface structured patterns and missing questions, but it should label report quality, duplicates, exposure uncertainty, alternative explanations, and missing denominators rather than presenting a crowd count as a safety verdict.
Triangulation is claim-matched, not source-matched
Different methods can compensate for different weaknesses. Randomization may strengthen causal attribution, observational data may capture broader populations or longer follow-up, laboratory work may clarify mechanism, and reports may reveal unexpected outcomes. Their value depends on whether they address the same claim.
Three sources about three adjacent outcomes do not create three confirmations. Relay should first align the population, exposure, comparator, outcome, and timeframe. Only then can it ask whether independent evidence truly converges or merely forms a persuasive narrative.
A living interpretation needs versioned reasoning
Scientific conclusions can change without implying that the earlier source record was fabricated. A new, larger, more direct, or better-controlled study may reduce or increase confidence, reveal subgroup differences, or show that an earlier signal was misleading.
A trustworthy living library keeps the source record stable, dates the interpretation, documents why the wording changed, and leaves unresolved community signals visible with their status. Versioned reasoning makes correction part of the product rather than a quiet replacement of yesterday’s certainty.
The takeaway
If you only remember one thing from this guide:
Evidence shows what was observed. Interpretation states what it can support. Community signals show what people are reporting. Relay connects all three without pretending they are the same.
Sources and support8 sources
- Real-World Evidence U.S. Food and Drug Administration
Distinguishes routinely collected real-world data from clinical evidence generated through analysis of those data.
- Results Data Element Definitions ClinicalTrials.gov
Defines outcome measures, outcome data, statistical analyses, and other elements that keep a reported result tied to what was measured.
Explains transparent presentation of effect magnitude, available evidence, and outcome-specific certainty.
Shows why conclusions must consider certainty, applicability, benefits, harms, and the boundary between evidence interpretation and recommendations.
Supports transparent, consistent reporting of objectives, methods, outcomes, results, caveats, and conclusions.
- Adverse Event Monitoring System Public Dashboard U.S. Food and Drug Administration
Explains that reports may be incomplete or duplicated and cannot by themselves establish causation, safety profile, or incidence.
- Good pharmacovigilance practices Module IX: Signal management European Medicines Agency
Defines a safety signal as information warranting further investigation and describes validation, prioritization, assessment, and recommendation steps.
- Reproducibility and Replicability in Science: Summary National Academies of Sciences, Engineering, and Medicine
Distinguishes reproducibility, replicability, and generalizability and emphasizes transparent reporting of data and methods.