Professor Sherman Kent - founding father of modern day intelligence analysis.
"Estimating is what you do when
you do not know"
The Problem
Anyone who either needs or produces informative answers is affected by a disinformation and misinformation world - where synthetic media, anonymous sources, and AI-generated content increasingly drown out the truth. Fact-checking and verification services help, but only a fraction of claims can be feasibly verified in sufficient time to meet fast-paced decision-making environments.
Our Vision
If we want to tackle disinformation and misinformation at scale, we need to apply this same structured form of estimating to claims themselves - in a way that is far faster than traditional analysis. This is why ENSEMBLE matters - it uses a structured, Bayesian-reasoning approach to estimate the probability of a claim being true, delivering fast, high-fidelity credibility assessments that consistently update as new data comes to light
The Need to Estimate
When intelligence analysts – who often deal with problems that cannot be easily verified – make assessments, they focus on likelihoods rather than binary truth. This is why they use Structured Analytic Techniques (SATs) – structured and methodical approaches to reasoning that reduce cognitive biases and lead to estimations that can support high-level decision-making.
FACT-CHECKING
Requires verifiable evidence and cannot account for uncertainty.
Relies on manual investigation and difficult to scale.
Not suited to fast-moving events and real-time decision-making.
.png)
.png)
TRUE
FALSE
ENSEMBLE
Uses multiple data points and is designed to mitigate uncertainty.
Uses automated investigation and is highly scalable.
Well suited to fast-moving events and real-time decision-making.
.png)
88%
Answers: is it true?
Answers: is it credible?
Requires verifiable evidence and cannot account for uncertainty
Uses multiple datapoints and designed to embrace uncertainty
Relies on manual investigation and difficult to scale
Uses automated investigation and is highly scalable
Not suited to fast-moving events and real-time decision-making
Well suited to fast-moving events and real-time decision-making
ENSEMBLE
.png)
FACT-CHECKING
Operational Flexibility
ENSEMBLE assesses claims made within a closed or open dataset of news articles, reports, or other qualitative materials. It operates as a standalone tool that can interact with external systems and be integrated into existing analytical or document management workflows.
Beyond Generative AI
ENSEMBLE cautiously incorporates AI tools for language processing and information extraction, but its assessments are driven by probabilistic models rather than LLM generation. This allows credibility to be evaluated systematically and transparently.
Three Key Stages
1 Extract Claims
The system starts by extracting and filtering relevant claims and related source information from any given qualitative dataset. This is done to understand relationships between claims, sources, and the claim being assessed.
2 Measure Agreements
The system then measures levels of agreement and disagreement across the totality of evidence. This process is applied both to the claim being assessed as well as agreeing and disagreeing claims - mapping out a multi-layered ecosystem of consensus.
3 Weigh Source Trustworthiness
Finally, ENSEMBLE assigns trustworthiness weightings to the sources behind each claim in the consensus map. These are calculated from hierarchical baseline features such as type / quality of evidence for untested sources, as well as the credibility track record of known sources.
Claim credibility
By combining levels of agreement with trustworthiness weightings, ENSEMBLE produces a dynamic claim credibility score - a probability that the claim is true.
Ingest Data
Assess Claims
Insights
Audit (optional)
"Estimating is what you do when you do not know"
Professor Sherman Kent - founding father of modern day intelligence analysis.
The Problem
Anyone who either needs or produces informative answers is affected by a disinformation and misinformation world - where synthetic media, anonymous sources, and AI-generated content increasingly drown out the truth. Fact-checking and verification services help, but only a fraction of claims can be feasibly verified in sufficient time to meet fast-paced decision-making environments.
The Need to Estimate
When intelligence analysts – who often deal with problems that cannot be easily verified – make assessments, they focus on likelihoods rather than binary truth. This is why they use Structured Analytic Techniques (SATs) – structured and methodical approaches to reasoning that reduce cognitive biases and lead to estimations that can support high-level decision-making.
Our Vision
If we want to tackle disinformation and misinformation at scale, we need to apply this same structured form of estimating to claims themselves - in a way that is far faster than traditional analysis. This is why ENSEMBLE matters - it uses a structured, Bayesian-reasoning approach to estimate the probability of a claim being true, delivering fast, high-fidelity credibility assessments that consistently update as new data comes to light
ENSEMBLE
FACT-CHECKING
Requires verifiable evidence and cannot account for uncertainty
Uses multiple datapoints and built to mitigate uncertainty
Relies on manual investigation and difficult to scale
Uses automated investigation and is highly scalable
Not suited to fast-moving events and real-time decision-making
Well suited to fast-moving events and real-time decision-making
Operational Flexibility
ENSEMBLE assesses claims made within a closed or open dataset of news articles, reports, or other qualitative materials. It operates as a standalone tool that can interact with external systems and be integrated into existing analytical or document management workflows.
Beyond Generative AI
ENSEMBLE cautiously incorporates AI tools for language processing and information extraction, but its assessments are driven by probabilistic models rather than LLM generation. This allows credibility to be evaluated systematically and transparently.
Three Key Stages
1 Extract Claims
The system starts by extracting and filtering relevant claims and related source information from any given qualitative dataset. This is done to understand relationships between claims, sources, and the claim being assessed.
2 Measure Agreements
The system then measures levels of agreement and disagreement across the totality of evidence. This process is applied both to the claim being assessed as well as agreeing and disagreeing claims - mapping out a multi-layered ecosystem of consensus.
3 Weigh Source Trustworthiness
Finally, ENSEMBLE assigns trustworthiness weightings to the sources behind each claim in the consensus map. These are calculated from hierarchical baseline features such as type / quality of evidence for untested sources, as well as the credibility track record of known sources.
.png)
Claim Credibility
By combining levels of agreement with trustworthiness weightings, ENSEMBLE produces a dynamic claim credibility score - a probability that the claim is true.