How to Tell If a Causal Vendor's Probability Model Is Auditable
On this page
A data science leader vetting a causal behavioral platform for a paid pilot needs one question answered before budget moves: can this vendor show the exact path from its modeling assumptions to a testable probability estimate, or does the output arrive as an unexplained number? A pilot spent on an opaque model cannot be defended internally when someone asks how the number was produced.
Why isn't "AI insights" a good enough answer?
Generic "AI insights" framing treats a probability estimate as a finished product, hiding the steps between the assumptions a model encodes and the number it reports. For a buyer, that gap is the actual risk: if a vendor cannot show how assumptions became a checkable estimate, there is no way to audit the estimate later, reproduce it, or defend it under internal scrutiny.
In PyMC, declaring random variables with distributions already specifies probabilistic assumptions. The generative graph can then be converted into an evaluable joint log-density graph. Inspecting those assumptions and the derived density helps audit this implementation; a computational graph alone does not prove causal identification or predictive validity.
From assumption graph to testable probability
One PyMC Bayesian inference path has three parts:
- Define the random variable graph. Each unknown quantity in the model, such as a mean, a variance, or a probability of belonging to a segment, is declared as a random variable with a distribution and its relationships to other variables.
- Convert the graph into log-probabilities. A dedicated conversion step walks the random variable graph and produces a joint log-probability graph: the mathematical object that says how likely any given set of parameter values is, given the model's assumptions and the observed data.
- Choose an inference method. Gradient-based samplers differentiate the log density; other valid methods use non-gradient MCMC, exact calculations or approximations. Record the method, diagnostics and reproducible output rather than requiring gradients for every model.
PyMC's 4.0 release rebuilt this pipeline around a graph-based computational backend, replacing the discontinued Theano project with a new fork purpose-built for this workflow (PyMC 4.0 Release Announcement). That backend has since been renamed and continued as PyTensor, reflecting further consolidation of the graph-computation layer rather than a change in the underlying pattern (PyMC's backend fork to PyTensor). The library names in this ecosystem have changed twice since 2022; the architecture pattern has not: declare assumptions as a graph, convert that graph into a checkable probability object, then sample from it.
What does this buy a buyer?
The practical advantage is inspectability. A model built this way can be interrogated at each stage: what assumptions were encoded, how they were converted into a probability estimate, and whether that estimate can be reproduced from the same graph and the same data. A model that skips straight to a final number, with no equivalent intermediate object, cannot answer any of those questions.
\log p(\theta \mid y) = \log p(y \mid \theta) + \log p(\theta) - \log p(y)Where theta is the set of model parameters, y is the observed data, p(y | theta) is the likelihood implied by the random variable graph, and p(theta) is the prior. The conversion step is what turns the right-hand side from an abstract statement into a graph that can be evaluated, differentiated, and sampled.
How does Subconscious apply the same audit standard?
Apply the same standard to a simulated choice study: identify the assignment design, audience inputs, choice model, estimand, uncertainty procedure and matched validation. Subconscious’s public working paper provides aggregate parameter-rank evidence and limitations. It does not validate every buyer study or reveal all private implementation details. Request the artifacts needed to reproduce the particular result.
Limitations
A matched human run can agree with, contradict or fail to resolve a simulated estimate. Define agreement criteria before comparing results, and distinguish stated choice from actual market behavior. The graph architecture here illustrates the open-source Bayesian ecosystem; it does not assert which internal library Subconscious or another vendor uses.
What to ask before shortlisting a vendor
Before shortlisting, ask for inspectable assumptions, data provenance, assignment or identification design, inference details, diagnostics and a reproducible comparison. A library-specific graph conversion is one way to supply those artifacts. Discuss the study when a concrete business decision needs this evidence.