Evidence typing & weighing

A claim is only as trustworthy as the kind of study behind it. Before anything reaches your brief, the agent labels every finding by evidence type, marks which claims stand on more than one independent source, and separates what the literature has settled from what it still disputes or has never answered.

#Why evidence type matters

Two sentences can look identical on the page and mean completely different things. “Compound X lowers tumor growth” carries a different weight when it comes from a randomized controlled trial than when it comes from a dish of cultured cells. A conventional summary flattens that difference — it strings claims together in fluent prose and leaves you to guess how much each one should count. The result reads as confident whether the underlying study was decisive or merely suggestive.

BioSkepsis refuses to flatten it. As the agent reads full text across the life-science literature, it classifies the design behind each finding and attaches that label to the claim. You never have to reverse-engineer the strength of a statement from its wording, because the evidence type travels with it. That lets you calibrate: treat a causal result as a candidate to act on, treat a preclinical one as a lead that still needs human data, and never confuse the two.

Typing is per claim, not per paper

A single paper can produce several findings of different types — a mechanistic experiment in cells (preclinical) alongside an observational cohort (associational). The agent types each finding on its own merits, so one source can contribute claims at more than one evidence level.

#The four evidence types

Every finding is assigned exactly one of four types. They are ordered roughly by how directly they support a cause-and-effect claim, but each answers a different question and each has a legitimate role in a research brief.

#Causal

An intervention or a demonstrated mechanism establishes that one thing produces another. This is the evidence you get from controlled experiments and randomized trials — a variable is deliberately changed and the effect is measured against a comparison. Causal findings are the strongest basis for deciding what an intervention actually does, and the agent flags them as the claims most suitable to act on.

#Associational

A correlation is reported, but causation has not been established. Observational studies, cohorts, and epidemiological analyses typically produce associational evidence: two variables move together, yet confounding factors or reverse causation remain possible. These findings are genuinely informative — they generate hypotheses and describe real-world patterns — but the agent is careful never to let an association masquerade as proof of cause.

#Synthesis

The finding is drawn from a source that has already pooled or reviewed multiple studies — a systematic review or meta-analysis. Synthesis evidence carries the weight of an entire body of work rather than a single experiment, which is why it often anchors a well-supported conclusion. The agent treats it as high-value context, while still noting when a review itself reports heterogeneity or conflicting primary results underneath its headline conclusion.

#Preclinical

The evidence comes from in vitro systems (cells, tissues, biochemical assays) or animal models rather than from humans. Preclinical work is where most mechanisms are first uncovered and where the earliest signals appear, but its results do not always translate to clinical outcomes. The agent labels these findings plainly so that a promising bench result is never quietly read as a clinical fact.

Type	Typical source	Answers	Read as
Causal	RCTs, controlled interventions, mechanism experiments	Does X cause Y?	Strongest basis to act
Associational	Cohorts, observational and epidemiological studies	Do X and Y move together?	Signal, not proof of cause
Synthesis	Systematic reviews, meta-analyses	What does the whole body of work say?	Weight of many studies
Preclinical	In vitro assays, animal models	Does the mechanism work in a model?	Early lead, needs human data

#Synthesized findings

Evidence type describes the kind of study behind a claim. Independent corroboration is a separate axis — and just as important. When two or more independent sources support the same finding, the agent tags it as Synthesized. A synthesized claim is not resting on a single result that could turn out to be a fluke, an outlier, or a lab-specific artifact; it reflects agreement across studies that did not depend on one another.

Independence is the point. Two papers from the same group reporting the same experiment, or a review that simply restates its own included study, do not add up to independent corroboration. The agent looks for genuinely separate lines of support before applying the tag, so a Synthesized label means the finding has cleared a real replication bar — not merely that it was mentioned twice.

Type and synthesis stack

The two labels compose. A finding can be, for example, Causal and Synthesized — a cause-and-effect result reproduced across independent trials. That combination is the strongest thing a brief can carry, and it is called out explicitly so it stands apart from a lone unreplicated result.

#Contradictions & gaps

Real literature disagrees with itself, and a brief that hides that is worse than useless. The agent surfaces disagreement rather than averaging it away.

#Contradictions

When the studies the agent reviewed reach conflicting conclusions on the same question, that is reported as a contradiction — both sides shown, with their sources, instead of a single tidy answer that quietly picks a winner. Contradictions are signal: they often mark exactly the places where the science is unsettled, where an effect depends on conditions the studies defined differently, or where a result has failed to replicate. Seeing them lets you judge for yourself rather than inherit the agent’s pick.

#Gaps

A gap is a question the literature has not answered — an aspect of your research question where the agent searched, read, and found no adequate evidence either way. Naming a gap is a deliberate, useful result: it tells you where the frontier sits, and it is frequently the most valuable line in a brief because it points directly at what to investigate next.

Absence of evidence is stated, not implied

A gap means the agent found no adequate evidence — not that the claim is false. The brief marks it as unestablished so an open question is never silently rounded up into a settled conclusion.

#Coverage of each sub-question

The agent breaks your research question into its component sub-questions and reports how well the literature covers each one, so the brief’s completeness is visible at a glance rather than buried. Every sub-question is marked with one of three coverage states.

Covered. The literature adequately answers this sub-question, with sources — often synthesized across more than one — behind the answer.
Limited evidence. Some evidence exists, but it is thin, preliminary, conflicting, or drawn from weaker designs. There is a partial answer, and it should be read with caution.
Not established. The agent found no adequate evidence to answer this sub-question. It is an open gap, flagged so you know the brief does not, and cannot, resolve it.

Reading these states together with the per-claim evidence types gives you the full shape of what the agent found: which parts of your question are settled, which rest on weak or contested footing, and which the field has simply not answered yet. That map — not a wall of confident prose — is what lets you trust the brief and decide where to look next.

← PreviousHow the landscape feeds your brief Next →Citations & verification