Evidence Library

Drop in payer policies, KOL transcripts, competitor readouts, market research. The platform extracts the structured content and breaks it into reviewable passages; you pin the ones that matter to a KBQ so every recommendation has a defensible source trail.

How this works

Two things ground the AI on this page — both are sent on every AI call across Overview, Strategy, Scenarios, and Outputs.

1
House view (below) — short team framing: asset, what "good" looks like, decision rules.
2
Uploads (further down) — parsed into reviewable passages. Only the ones you pin to a KBQ count as evidence.

Together they clear coverage-gap flags on the Issue Map and are quoted verbatim — with source — inside every AI narrative. Nothing unpinned ever reaches the model.

House view

Team framing · sent on every AI call alongside pinned chunks

Grounding context · injected into every AI call

Asset: Wegovy (semaglutide 2.4mg) — Novo Nordisk, GLP-1 RA for chronic weight management.

What "good" looks like for this launch:
- Defend BMI 30+ w/ comorbidities as the durable commercial wedge (best access, lowest abandonment).
- Convert HCP demand into filled scripts — PA burden + copay shock are the #1 leakage points.
- Hold rebate depth at the level that buys formulary parity without giving away GTN we can't recover.

Watchlist:
- Lilly tirzepatide (Zepbound) — assume aggressive contracting in Yr 2-3, especially commercial.
- Medicare Part D coverage expansion is the upside scenario, not the base case.
- Compounded GLP-1s are a near-term distortion, not a multi-year threat.

How we make decisions here:
- Quant moves first (engine). AI narrates the quant — it does not invent numbers.
- Every recommendation must name (a) the lever, (b) the quantified impact, (c) what would have to be true, (d) the KBQ it depends on.
- If evidence isn't pinned to a KBQ, treat that KBQ as open — flag it in any deck.

Tip: keep this short and specific. Asset + what "good" looks like + watchlist + decision rules. ~10–25 lines is the sweet spot.

✅ Upload these

• Payer policy bulletins, formulary PDFs, PA criteria
• PBM RFP responses, rebate term sheets
• KOL / advisory board transcripts & readouts
• Competitor launch reports, analyst notes
• Claims studies, real-world evidence, KFF / IQVIA pulls
• Internal forecast memos & assumption logs

🎯 What good looks like

• Source is dated & attributable (publisher, date on page)
• Contains a specific number or policy clause you can defend
• Pinned to the narrowest KBQ it actually answers — not the parent theme
• 2–6 pins per major KBQ; triangulate, don't pile on
• Re-pin when you update a snapshot so the audit trail tracks

⛔ Don't upload

• PHI, patient-identifiable data, or raw claims rows
• Anything under NDA you can't share with the analyst team
• Marketing decks with no underlying data (low signal)
• Files > 15MB — split or extract the relevant section first
• Scans where OCR will be unreliable — retype the key table

Workflow: Drop file → wait for parse → expand the doc → read each chunk → click Pin to KBQ on the ones that genuinely answer a question. Skip the rest. A chunk that isn't pinned is invisible to the AI — that's the feature, not a bug. It keeps narratives defensible.

Drop a document here or click to browse

PDF, DOCX, PPTX, XLSX, TXT, MD · 15MB max · enterprise-grade document understanding

No documents yet. Uploads stay in this browser session and are tied to the current model snapshots.