Claim
The exact assertion, including population, comparator, endpoint, qualifier, audience, and use context.
Talent
Most AI systems help people find answers. We help institutions know when an answer should stop feeling settled.
NextConsensus evaluates a defined portfolio of evidence-sensitive claims against an agreed evidence record. We are building the data model, evaluation framework, and control system for maintaining evidence-sensitive institutional judgment over time.
Judgment under moving uncertainty.
A new trial is published. A guideline shifts. A competitor changes the relevant standard of care. A safety signal emerges. A subgroup becomes clinically important.
Yet the claims, policies, and evidence-backed positions inside healthcare organizations often remain active until someone manually notices that the world around them has changed. NextConsensus exists to make continued reliance under changing evidence observable, reviewable, and governable.
Does this specific change in external evidence alter the supportability of this specific institutional claim, in this specific context, enough to justify renewed review?
That problem is not reducible to search, RAG, summarization, or workflow tooling. Those are useful ingredients. The category is the maintained relationship among claims, evidence dependencies, evidence movement, institutional context, and expert disposition.
The data asset is a relationship, not a pile of papers.
The long-term learning loop is: claim x evidence dependency x evidence change x institutional context x expert disposition.
Evidence dependency
The sources and assumptions that made the claim supportable at a dated review point.
Evidence change
The later movement that strengthens, weakens, narrows, broadens, contradicts, or destabilizes the claim.
Institutional context
Where the claim is used, who owns review, and what threshold would justify renewed attention.
Expert disposition
What qualified reviewers decided and what should cause the judgment to be revisited.
What makes this hard.
Literature search and summarization are becoming commodities. The hard problem is maintaining the relationship among a claim, its evidence dependencies, later evidence movement, and expert decisions over time.
-
Claim identity across sources
How do you know that "Drug X improves outcomes in population Y" in a 2019 trial report is the same claim as "Drug X demonstrates benefit in subgroup Y" in a 2024 meta-analysis? Without canonical claim identity, evidence movement is just literature surveillance.
-
Temporal integrity
Every assessment is pinned to a review date. Evidence published after that date does not enter the assessment. Hindsight does not leak in. This constraint makes the output auditable — and it is computationally non-trivial at scale.
-
Evidence-state representation
The output is not a summary or a score. It is a structured state: has the evidence behind this claim strengthened, weakened, narrowed, broadened, or become more contested since the last review? Representing that state comparably across claims, sources, and time is the core representational problem.
structured claim representationevidence-dependency graphslongitudinal evidence-state comparisonscientific entity, population, comparator, and endpoint matchingreview-threshold assessment under uncertaintyexplicit provenance and temporal replayhuman-in-the-loop adjudicationworkflow integration without owning workflowlearning from institutional decisions over time
What we believe about the work.
The hard problem is relational
The question is not whether a model can summarize a paper. It is whether a specific evidence change alters the supportability of a specific institutional claim, in a specific use context, enough to justify renewed review.
We build maintained support-state records
The core object is not a document or prompt. It is a maintained state: what the claim means, why it is supportable, what evidence it depends on, what changed, what experts decided, and when that judgment should be revisited.
We earn the long term
The long-term asset is the linked evidence-change and disposition history that accumulates with each assessment. That history makes future re-review faster and more defensible — and eventually enables support-state intelligence across a portfolio.
We build for expert judgment
NextConsensus does not replace medical, regulatory, legal, or compliance judgment. It makes that judgment more timely, structured, traceable, and consistently applied.
We separate change from authority
Evidence movement is not the same as final action. The system makes continued reliance observable and reviewable while preserving the customer's authority over meaning and action.
Who should see a serious problem here.
This should attract people who are skeptical of thin copilots but interested in durable systems for evidence, time, provenance, institutional context, and accountable human judgment.
Applied AI and ML engineers
Provenance-aware retrieval, scientific document understanding, calibrated classification, temporal reasoning, graph retrieval, uncertainty estimation, and evaluation under expert disagreement.
Knowledge-graph and ontology engineers
Claims, populations, interventions, comparators, endpoints, evidence, qualifications, use contexts, dispositions, and reopening conditions need durable representation.
Clinical informaticists and evidence scientists
Systematic review, HTA, pharmacoepidemiology, medical information, guideline methodology, clinical evidence synthesis, and regulatory science.
Regulated product and design leads
Expert workflow, audit trails, uncertainty communication, decision-state design, trust calibration, and interfaces that make reasoning inspectable.
Medical affairs and regulatory operators
People who have lived stale substantiation, duplicated review, claim tracing, fragmented ownership, and repeated reconstruction of prior rationale.
Infrastructure and systems engineers
Source observation, event pipelines, versioned state, tenant-specific logic, provenance preservation, auditability, permissions, and security.
Depth over coverage. Precision over generic summaries.
What we choose to optimize — and what we leave to others.
- We are not building "chat with PubMed." Search and summarization are enabling layers, not the category.
- We are not automating scientific truth or replacing expert review.
- We are not building claims workflow, approval, or document-control infrastructure. Customers own governance and action.
- We are not claiming long-term transition modeling as a present-tense product. That capability must be earned from accumulated review data.
- We choose depth over coverage: claim identity, evidence dependencies, temporal comparison, provenance, and human adjudication before broad generic monitoring.
We release infrastructure, not just records.
Sourced through Refract, our open-source developer SDK. It turns raw public revision histories into structured timelines. Anyone can inspect it, run it, or build on it.
Buyers get the healthcare re-review layer: sourced records with caveats, source references, and review context. Developers and researchers get the observation layer that demonstrates how public knowledge change can be structured from revision histories.
This is not for everyone.
You might be a fit if:
- You think the hard problem in medical AI is maintaining prior answers as evidence changes, not producing one more answer.
- You care about structured claim representation, evidence-dependency graphs, temporal evidence comparison, and provenance.
- You are comfortable building systems where human adjudication is a feature, not a temporary crutch.
- You can distinguish supportability, materiality, authorization, and action without collapsing them.
- You can write clearly about uncertainty without hedging into meaninglessness.
- You want to help build the dataset that does not exist yet: how evidence movement becomes institutional review, decision, and action.
Method basis
Research threads behind the work.
The literature defines the problem space, not the finished system. The engineering work is structured evidence records, claim-level provenance, living evidence, and human-controlled AI.
- The FAIR Guiding Principles for scientific data management and stewardship
Supports the need for durable, inspectable, reusable evidence and provenance records.
- How citation distortions create unfounded authority: analysis of a citation network
Shows how citation patterns can make a claim appear more settled than the source record supports.
- Guidelines for Human-AI Interaction
Supports interface patterns that keep uncertainty, user control, correction, and human authority visible in AI-assisted systems.
- Living systematic review: 1. Introduction-the why, what, when, and how
Frames evidence synthesis as an updating problem, not a one-time search problem.
Help build the architecture for maintained medical judgment.
If building systems for claim identity, evidence-state comparison, provenance, human judgment, and institutional accountability is more interesting than optimizing existing tooling, get in touch.