From research question to decision

This is an illustrative worked example, not a real customer engagement: it follows the public sample dossier that ships with the workspace — "SGLT2 inhibition in cardiorenal injury" — so every step below matches exactly what you see in the read-only demo. The cited sources are real published papers; the engagement around them is constructed for demonstration.

The decision under assessment

A translational team needed to answer one question, framed at intake as:

"Does SGLT2 inhibition deliver cardio-renal protection — reduced interstitial fibrosis in tissue models and improved kidney and cardiac outcomes in humans — compared with standard of care?"

Intake captured the full frame in one pass: the mechanism (SGLT2 inhibition), the tissue and models in scope (kidney and cardiac interstitium, human outcome trials and mouse models), the comparator (standard of care or placebo), the endpoints (interstitial fibrosis, composite kidney outcome, cardiac outcomes), the programme stage (preclinical), and the named reviewer, who would have to approve the final dossier. A decision deadline of 2026-12-01 and an evidence cut-off of 2026-10-01 were recorded so the assessment would age visibly.

Building the evidence base

The evidence module turned the question into a structured table of five records rather than a reading list. Three are large human outcome trials — DAPA-CKD (dapagliflozin in chronic kidney disease), DAPA-HF (dapagliflozin in heart failure with reduced ejection fraction) and EMPA-REG OUTCOME (empagliflozin in type 2 diabetes) — and two are mouse models: one showing empagliflozin attenuates renal interstitial fibrosis in diabetic kidney disease, and one showing SGLT2 inhibition does not improve active lupus nephritis in MRLlpr/lpr mice.

Each record carries its design explicitly: setting (human or animal), model, exposure and dose, comparator, endpoint, measured effect with its confidence interval, and the study's own stated limitations. Every record also carries a DOI or PubMed identifier and a link to the exact source, so no claim floats free of its evidence. Two things surfaced that a narrative summary would have hidden:

• The human trials measure composite kidney and cardiac outcomes — none measures an interstitial-fibrosis endpoint directly, so the human signal and the tissue mechanism do not close the loop on their own.
• The cross-context difference came from the mouse side: in the active lupus nephritis model, SGLT2 inhibition produced no improvement. That is a negative result in a different disease model — not a direct contradiction of the diabetic-model finding, though it does rule out a generalised anti-fibrotic mechanism claim. The contradictions section flags this between the paired records instead of burying it in different paragraphs.

Claims, review and the approval gate

The synthesis step converted the table into claims, each typed as an observation, an interpretation or a proposal, and each tied to its supporting records. The executive assessment holds the observation across the human trials and the mechanistic mouse study; the contradictions section holds the interpretation that the benefit is not universal across kidney-disease contexts; the translational gaps and study options sections hold the proposals.

Nothing could be exported until the named reviewer approved the exact dossier version. The review record in the sample shows the gate working across versions: version 1 came back with changes requested (clarify that the DAPA-CKD endpoint is a composite, not an isolated fibrosis endpoint, and separate human from animal uncertainty); version 2 came back again (add the lupus-nephritis model as a cross-context difference and restrict the mechanism claim accordingly); version 3 added the cardiac outcome trials so the table matched the full cardio-renal question, and was approved by the named reviewer on 2026-10-10. Each content edit bumped the version, invalidated the earlier approval, and required re-review — so the exported dossier carries a version number that provably matches an approval.

The decision

The approved dossier gave the go/no-go meeting three things it did not have before:

• A decision brief and executive assessment separating what the evidence shows, what it means and what the team proposes, with each claim traceable to reviewed records.
• An explicit contradictions section naming the lupus-nephritis model, so the meeting argued about a stated cross-context difference rather than discovering it.
• A translational gaps section stating what is missing before any scale-up decision: human tissue validation of the anti-fibrotic endpoint, and a proposed kidney-tissue cohort study stratified by diabetic versus inflammatory aetiology.

The decision, as the sample dossier records it: the evidence supports continuing preclinical work, but only with a human-tissue validation step before any translational scale-up decision. The dossier had turned a binary judgement based on a reading list into a scoped, reviewable decision with a defined next gate.

The deliverable, inspectable

The dossier itself is the product, so it ships in public view. Below is the exact exported deliverable for this walkthrough — the same nine-section, version-locked document an approved export produces. Open it to inspect every evidence record, its confidence interval and stated limitations, or download it as PDF or markdown.

Why this is the general pattern

Every project in Corelia Intelligence follows the same arc: a framed question at intake, a structured evidence table with a full provenance trail, typed claims with human review, a version-locked approved dossier, and an export that stakeholders can trust without re-reading the sources. The assessment ages visibly, and when new evidence arrives the dossier is refreshed rather than rebuilt.

You can reproduce this exact walkthrough in the read-only demo with the public sample dossier, or bring your own question to a trial.