AIMN:VERDICT is the architecture for adversarially-validated entity intelligence across the 301 monitored biopharma entities in the AimwellBio Federated Intelligence Network. The full 4-agent verdict engine is in activation. What is live today is the proof layer underneath it: the dossier and briefing pipeline, multi-provider LLM analysis with an independent verifier model, a hallucination-detection layer, confidence calibration, and the curated corpus / ATLAS — every claim cited to a public-source URL, saved to your account, encrypted in transit and at rest.
Every AIMN:VERDICT now also carries a G1–G7 grade for use in dashboards, IC memos, and regulatory submissions. PROCEED · DELAY · KILL remains the canonical verdict; the grade adds confidence-and-impact resolution.
Citeline tracks trials. Cortellis tracks regulatory. AlphaSense indexes transcripts. Bloomberg shows financials. Expert networks charge $1,000/hour for qualitative depth. Five separate logins, five separate invoices, and at the end the analyst still has to assemble the verdict by hand.
Click a marker on /atlas. Streaming progress shows the engine reading PubMed, ClinicalTrials.gov, SEC EDGAR, FDA enforcement, and EMA EPAR in parallel. Within 90 seconds you receive a 2–3 page rendered verdict with every claim hyperlinked to its source. Re-run with deeper search at 4-iteration mode in 150–180 seconds.
Every verdict separates the entity's public narrative (what they claim in 10-K MD&A and IR materials) from the public record (what trials, filings, enforcement, and peer-reviewed literature actually contain). The variance between the two is the verdict. Methodology disclosed in full at /methodology.
After generation, the engine HEAD-checks every cited URL. If more than 10% fail to resolve, the verdict's confidence score drops to LOW with a yellow disclaimer. This is the institutional-credibility gate. It is what separates a verdict from generic AI search output. Incumbents do not ship this.
A reduced-fidelity sample of the verdict format is published for public review. Full verdicts are member-gated; this sample shows the structure, the citation discipline, and the variance-analysis layout. Marked SAMPLE — not a live production verdict.
The 4-agent adversarial verdict engine is in activation. This panel is an architecture preview of the request flow Member and Founding Contributor tiers will use once the engine is live. What is available today is the published dossier sample and the live briefing proof layer. Select an entity below to preview the flow and request verdict access.
Observer can read 5 sample verdicts (preview). Member tier ($199/mo) includes a monthly verdict allowance after activation across all 301 monitored entities. Founding Member locks the founding rate for life: $208/mo or $2,500 once during the founding window, versus the $499/mo standard after.
Request verdict access →Synthesis pipeline: Private Beta · Full production deployment Q3 2026 · Methodology →
This is exactly what a Member or Founding Contributor experiences. Steps marked LIVE are operational. PRIVATE BETA means access by application only. IN ACTIVATION / PLANNED means on the engineering roadmap — committed, not yet shipped. The 4-agent adversarial verdict engine itself is in activation; the dossier and briefing proof layer is live today.
Click any of 301 monitored biopharma markers on AIMN:ATLAS. Observer-tier users can browse the map; Member-tier users trigger verdict generation.
Live for ObserverThe engine reads PubMed, ClinicalTrials.gov, SEC EDGAR, OpenFDA, and bioRxiv in parallel. Streaming progress is visible. Generation completes in 90 seconds standard, 150–180 seconds at 4-iteration depth.
Private BetaAfter generation, every cited URL is HEAD-checked. If more than 10% fail, the verdict's confidence score drops to LOW with a yellow disclaimer. This step runs on every verdict at every tier.
In activationThe rendered verdict — including source list, confidence score, and generation timestamp — is saved to your account, encrypted in transit and at rest. 90-day cache. Re-run to refresh.
Planned · Member tierThe PDF export will include: entity name, generation timestamp (UTC), source count, citation-resolution rate, confidence score, and watermark. Structured for institutional workflows. Export endpoint is in activation and becomes available after onboarding.
In activation · Member tierMembers will be able to generate a signed URL to share a specific verdict with colleagues. The recipient sees the verdict is unmodified. Links expire at a configurable TTL. Founding Contributor tier includes team-wide vault access. Signed-URL sharing is in activation and becomes available after onboarding.
In activation · Member tierInstitutional partners do not buy black-box AI tooling. They buy methodology they can defend in a regulator's room. The full AIMN:VERDICT methodology is published openly at /methodology. Three pillars below.
The system prompt instructs the model to surface variance between public narrative and public record. This template is the brand IP. Available for member review on request — not generated content.
Post-generation, every cited URL is HEAD-checked against the public web. If >10% fail to resolve, the verdict's confidence score drops to LOW with a yellow disclaimer. Non-negotiable.
0–100 score driven by source coverage (PubMed depth, recent SEC filings, trial volume) and citation resolution rate. Visible on every verdict. Members trust the floor — they know which verdicts to act on and which to flag.
Standard intelligence tools surface what is known. Adversarial intelligence surfaces what is hidden and what is in the blind spot. The Johari Window is the classification framework that separates a sourced verdict from a summary.
Confirmed public record: published trial results, 10-K filings, approved labeling, guidance documents. AIMN:VERDICT aggregates and cites this layer. It is the baseline — the entity's declared public position.
Regulatory gaps, competitive pressure, trial-failure signals, or enforcement patterns that haven't surfaced in the entity's own communications yet. This is where early-warning value lives.
The adversarial finding: the variance between what the entity claims in MD&A, IR materials, and press releases versus what the public record — trials, filings, enforcement, peer literature — actually contains. This is the primary AIMN:VERDICT output.
Cross-source inferences visible only when PubMed, SEC EDGAR, CT.gov, OpenFDA, and bioRxiv are read in combination. No single database surfaces these. The 5-source parallel read is what makes them visible.
Full methodology published at /methodology · Adversarial Validation Standard v1 at /standard
The Founding-Contributor tier finances the AIMN:VERDICT engineering build. Founding cohort capacity is reviewed manually, and pricing rises in subsequent waves once the founding allotment closes. Reserve now.
When you receive a brief that influences a high-stakes decision, what is the step that determines whether you trust it?
Just want to monitor signals? Observer Access is free for credentialed pros — see /observer →
Plus: Member tier includes 50 verdicts/month. Founding ($208/mo or $2,500 one-time) locks the founding rate for life, versus the $499/mo standard, plus acknowledgments-page citation on the methodology paper.
Hand us anything your team is about to use. We return a source-risk review with every citation traced. No sales call. No follow-up sequence. The review itself is the conversation.
The decisions that come back to haunt you are not the ones you got wrong on purpose. They are the ones built on intelligence no one had time to check.
View Sample Verdict →Teams searched documents, reviewed guidance updates, followed trial movement, tracked competitors, and built internal briefs by hand. Slow but auditable. Every claim had a paper trail.
Speed. The same brief that took a week now takes minutes. The trade-off: unsupported claims, stale guidance, missing context, and incomplete summaries can now move into serious decisions before anyone challenges them.
AimwellBio is the verification layer that sits between AI-generated intelligence and the decision someone signs their name to. Source-trace, adversarial review, confidence labels, institutional memory.
Seven ways unsupported scientific intelligence can enter regulatory, clinical, board, and investor workflows. Open to credentialed professionals. No sales sequence, no list-trading.