Science
Build. Measure. Test. Publish.
Imagine All The People builds synthetic populations under explicit constraints, measures their coherence, tests the robustness of results and documents the methods used to examine them.
The language model is involved in interactions with synthetic individuals. On its own, it does not define the statistical structure of the population.
Verify the evidence
Five questions. Five levels of evidence.
01 — Foundations
How is the population built?
- Multidimensional constraints
- Maximum Entropy Relaxation
- Calibration
02 — Validation
How do we know what is robust?
- External references
- Deviation measurement
- Auditability
03 — Benchmarks
What have we actually measured?
- Explicit protocol
- Quantitative measures
- Scope of the result
04 — Publications
What has actually been published?
- Submitted preprints
- Work in preparation
- Sources and identifiers
05 — Team
Who is behind this work?
- Identifiable contributions
- Authors
- Documented affiliations
Maximum Entropy Relaxation.
Formulation
p(x) = exp( Σₖ λₖ · fₖ(x) ) / Z(λ)
- p(x)probability of a complete individual configuration x
- fₖ(x)statistical constraint k (unary, binary or ternary)
- λₖmultiplier associated with constraint k
- Z(λ)normalization constant of the exponential family
Among the distributions compatible with the known constraints, the approach seeks the maximum-entropy one: the distribution that adds no further structure beyond what the retained data imposes.
SourceFrançois Pachet, Jean-Daniel Zucker — Maximum Entropy Relaxation of Multi-Way Cardinality Constraints for Synthetic Population Generation, ArXiv — ArXiv:2603.22558 View
Preprint · not peer reviewed.
What it constrains
- Marginals
- Cross-distributions
- Reference distributions
What it does not determine
- An exact human response
- The future
- The client's decision
One result, one protocol.
Result
K ≥ 28
A threshold observed in this protocol, not universal superiority.
Protocol
- NPORS-derived · 4 to 40 attributes
- Unary, binary and ternary constraints
- Generalized raking vs Maximum Entropy Relaxation
Multiple checks. No magic score.
Reference
Confront the model with the available external knowledge.
Deviation
Measure the differences between target and generated distributions.
Variation
Vary the structuring assumptions to identify robust and fragile results.
Auditability
Make it possible to trace a result back to the methodological conditions that produced it.
Not everything has the same status.
Published
1 preprint submitted to ArXiv (ArXiv:2603.22558).
Not peer reviewed.
In preparation
Work described, not submitted.
Editorial contribution
Contribution to structuring a field of research.
What the method does not allow us to claim.
- Not an individual prediction.
- Not a substitute for real-world data.
- Not certainty about the future.
- Not an automated decision.
Identifiable contributions.
François Pachet
Population synthesis · constrained generation
Jean-Christophe Baillie
Simulation · cognitive models · dynamic agents
Jean-Daniel Zucker
Complex systems · methodological robustness
Science → decision
Why all of this matters for a decision.
Without explicit controls
- Hidden assumptions
- Results that are difficult to challenge
With a documented method
- Explicit constraints
- Measured deviations
- Informed judgment
Your next decision
Which decision do you want to explore?
Describe your need. We can point you to the right level of support.
