Hinsley Documentation
Crowdsourcing Scenario Likelihoods
After building scenarios, the next step is assessing how likely each one is. Rather than relying on a single analyst's judgment, Hinsley lets you crowdsource likelihood estimates from multiple people - team members, subject-matter experts, and external guests - and automatically aggregates them into a consensus view.
The Likelihood Scale
Every likelihood estimate in Hinsley uses a standardized seven-point scale aligned with common intelligence community probability language:
| Rating | Probability Range |
|---|---|
| Almost no chance | 1 – 5% |
| Very unlikely | 5 – 20% |
| Unlikely | 20 – 45% |
| Roughly even chance | 45 – 55% |
| Likely | 55 – 80% |
| Very likely | 80 – 95% |
| Almost certain | 95 – 99% |
This scale keeps assessments comparable across forecasters, and when estimates are aggregated the math maps cleanly to probability values.
Three Sources of Likelihood
Each scenario can display up to three likelihoods side by side:
- Hinsley estimate - An AI-generated likelihood with a written justification. Hinsley draws on the analysis's source material, decomposition, and scenario details to produce this estimate. It also tracks trend direction over the past seven days. See How the Hinsley Estimate Is Produced, below.
- Your estimate - Your personal likelihood rating. You can set or update it from the scenario detail view at any time. In a set that is both mutually exclusive and exhaustive, editing one estimate opens the form for every scenario in the set.
- Aggregate estimate - The combined consensus of all individual estimates. This appears automatically once a scenario has estimates from two or more contributors. The Hinsley estimate counts as one contributor.
How the Hinsley Estimate Is Produced
The Hinsley estimate comes from an ensemble of AI models. The ensemble is the account's question-forecasting roster: the same models, sampled the same number of times, that the account uses to forecast its questions.
- Each model in the roster assesses every scenario in the set, once per configured sample.
- Each model's samples are averaged into one likelihood per scenario.
- The models' likelihoods are averaged with equal weight. In a mutually exclusive and exhaustive set, the result is then adjusted to total exactly 100%.
- A final step reads every model's reasoning and writes one justification for the combined likelihoods. This step writes the justification only. It does not change the numbers.
The result is published as one Hinsley estimate: one likelihood per scenario, and one justification covering the whole set. The individual models' assessments are not shown and do not count separately in the aggregate.
If a model fails during a run, the estimate is published from the models that succeeded. A failed model can contribute its assessment from an earlier run of the last seven days, if the scenarios, the analysis horizon, and the set's constraints have not changed since. If every model fails, the previous Hinsley estimate stays in place.
How Aggregation Works
When multiple humans submit their estimates, Hinsley computes an aggregate likelihood using the following approach:
- Each estimate is converted to a probability. An estimate entered as a percentage is used as is. An estimate entered as a rating is converted to a point near the middle of its range. For example, Likely maps to 68% and Very unlikely maps to 13%.
- The arithmetic mean of all individual probabilities is calculated.
- The mean is mapped back to the corresponding rating on the seven-point scale.
In a set that is both mutually exclusive and exhaustive, Hinsley scales the mean probabilities so that the aggregate totals exactly 100% before it maps each one to a rating.
This process runs automatically each time a new estimate is submitted or an existing one is updated. The aggregate includes every current individual estimate from logged-in users and guest forecasters, plus the Hinsley estimate as a single contributor.
Configuring Who Can Submit
Analysis owners control who can submit likelihood estimates through the submission settings. Three access levels are available:
- Collaborators only - Only users who have been added as collaborators on the analysis can submit estimates. This is the default.
- Account members - Any user in the same organization account can submit.
- Anyone with the link - Anyone who has the submission link can submit, including external guests without a Hinsley account.
Inviting External Forecasters
When accepting submissions from anyone with the link, you can send email invitations directly from the submission settings panel. Enter one or more email addresses and Hinsley will send each recipient a personalized link to the submission form. The link pre-fills their email address so they only need to enter their name and assessments.
Tracking Changes Over Time
Hinsley preserves a full history of every likelihood estimate. Each time a forecaster updates their rating, the previous value is retained in the history log. You can review this history from the scenario detail view to see how assessments have shifted as new information emerged.
The likelihood history also includes a chart that visualizes trends over time, plotting the aggregate likelihood, the Hinsley AI estimate, and individual forecaster estimates on a timeline, so divergence between forecasters or between human and AI assessments is visible on a single chart.