Methodology

The method behind every brief

Evidensity runs the stages of a systematic literature review (SLR) — search, screening, extraction, synthesis — as one structured, reproducible pipeline. Every figure is bound to the source quote it came from, and the whole method is benchmarked against published meta-analyses.

01
Search
10+ databases
02
Screen
Explicit criteria
03
Extract
Full-text schema
04
Ground
Quote + section
05
Validate
Credibility
06
Synthesise
Meta-analysis

Six stages, each answerable to a standard

1

Search & discovery

A structured query runs across more than ten academic databases — including PubMed, Europe PMC, Semantic Scholar, OpenAlex, Crossref and arXiv — to assemble the candidate corpus. Recall comes first: the aim is to surface every potentially relevant study, not a convenient sample.

PRISMA · Identification
2

Screening & eligibility

Every record is screened against explicit, pre-stated inclusion and exclusion criteria. On-topic studies pass, off-topic ones are excluded, and ambiguous ones are flagged for review rather than silently dropped — so the resulting corpus is defensible and reproducible.

PRISMA · Screening & eligibility
3

Full-text extraction

For each included paper the full text — not just the abstract — is parsed into a structured schema: methods, datasets, sample sizes, outcomes, effect sizes and confidence intervals. Findings are captured as data, not prose, so they can be compared and pooled downstream.

Structured data extraction
4

Source grounding

Every extracted figure and claim is bound to the verbatim sentence and section it came from. Nothing enters the synthesis without a traceable source span — this is the mechanism behind “every claim, cited.”

Traceability · audit trail
5

Validation & credibility

Extractions are cross-checked across stages to catch inconsistencies, and each study receives a multi-dimensional credibility assessment — methodological soundness, reporting completeness, directness. Weaker evidence is down-weighted and labelled, not silently mixed in.

Risk-of-bias · GRADE-style certainty
6

Quantitative synthesis

Where outcomes are comparable, effects are pooled with random-effects meta-analysis (restricted maximum likelihood, REML), reporting heterogeneity (I², τ², Q), prediction intervals and forest plots. Where pooling isn’t appropriate, the synthesis stays narrative — and says so.

Random-effects meta-analysis
The signature step

“Every claim, cited” — what that actually means

A general-purpose model will summarise a literature and sound confident; it won’t show you where each number came from. Evidensity inverts that. Every figure in a report resolves to the exact words it was extracted from — here is the shape of a single citation (illustrative):

Edoxaban reduced the primary endpoint — stroke or systemic embolism — versus warfarin (hazard ratio 0.79).

“The annualized rate of the primary end point during treatment was 1.50% with warfarin … as compared with 1.18% with high-dose edoxaban (hazard ratio, 0.79; 97.5% CI, 0.63 to 0.99; P<0.001 for noninferiority).”
— ENGAGE AF-TIMI 48 · Abstract

See it live: hover any citation in the GLP-1 or DOAC evidence bundle and the source’s own sentence appears.

Proof

Validated by reproduction

The strongest test of a method is whether it reproduces results an independent team has already published. Run from a single plain-text question, the pipeline reproduces the headline findings of landmark meta-analyses:

Both are on the Life Sciences page, with forest plots and the full concordance.

What you receive

One self-contained, auditable deliverable

Every engagement arrives as a single interactive evidence bundle: browse the whole corpus, trace each claim to its source quote, and inspect the credibility scorecard, forest plots and the full reference list. It is one HTML file — readable by your team without dependencies, indexable in your own systems, and stable years from now.

Start with a question

Free scoping assessment, returned within 24 hours. No commitment.

Commission a brief

or email directly: evidensity.research@gmail.com