Running your first causal analysis

This guide walks you through a complete causal analysis session in Wu-Weism — from signing in to reviewing a counterfactual claim in the Claim Ledger. By the end, you will have touched all three rungs of Pearl’s causal ladder in a single conversation.

What you will build

A three-question session on cortisol and hippocampal volume that progressively climbs from association (Rung 1) to intervention (Rung 2) to counterfactual (Rung 3), with each response grounded in the neuroscience Structural Causal Model (SCM).

Navigate to your Wu-Weism instance and sign in. Once authenticated, go to /chat.You will see the Causal Chat interface: a message input at the bottom, the model settings panel in the sidebar, and an empty conversation thread. Before sending your first message, confirm that a provider key is configured — the model selector in the sidebar should show an active provider. If it shows a configuration warning, see Configuring your AI provider first.

Ask your first causal question

Type a precise causal question in the message input. Specificity matters — the more clearly you frame a causal relationship, the more accurately Wu-Weism can classify the domain and load the right SCM.

Example question — Rung 1

Does increased cortisol exposure during stress events causally increase hippocampal volume reduction?

Press Enter or click Send.

Frame your question around a specific mechanism rather than a general topic. “Does X causally affect Y under condition Z?” consistently produces better-grounded responses than open-ended questions like “Tell me about cortisol.”

Watch domain classification

Before the response content appears, Wu-Weism classifies your question into a research domain. You will see a brief status indicator:

Classifying domain...
Domain: neuroscience
Loading SCM truth cartridge...

Domain classification determines which Structural Causal Model (SCM) — called a Truth Cartridge — is loaded to constrain the response. For this question, the system identifies neuroscience and loads the corresponding cartridge, which encodes known causal relationships between stress hormones, the HPA axis, and hippocampal structure.If your question spans multiple domains (e.g., a pharmacological intervention studied in a neuroscience context), Wu-Weism selects the primary domain based on the dependent variable in your question.

Read the structured response

The response for a Rung 1 (association) question will include:

Causal density rung: The response header indicates which rung the answer operates on. For a purely observational question, you will see Rung 1 — Association.
SCM constraints applied: The specific causal nodes and edges from the neuroscience cartridge that are relevant to your question.
Governed claim: A structured claim with an uncertainty label (e.g., metric-bearing if empirical data was cited, structural if the answer derives from the SCM alone).
Provenance trace: The reasoning chain that produced the claim, including which SCM edges were activated.

Example response shape

[Rung 1 — Association | neuroscience SCM]

Elevated cortisol during acute stress events is associated with accelerated 
hippocampal volume reduction, mediated by glucocorticoid receptor activation 
in CA1 and CA3 subfields. This association is structurally supported by the 
HPA-axis → hippocampus pathway in the neuroscience SCM.

Claim recorded: claim_001
Uncertainty: structural | Confidence: 0.74

The claim ID (e.g., claim_001) links to a permanent entry in the Claim Ledger.

Trigger Rung 2 with an intervention question

Now escalate to an intervention by asking what would happen under a deliberate manipulation. Wu-Weism detects intervention intent from phrasing like “if we intervened,” “what would happen if we set,” or explicit do() notation.

Example question — Rung 2

What would happen if we intervened to reduce cortisol by 50% during stress events?

The response will shift to Rung 2 — Intervention:

Example response shape

[Rung 2 — Intervention | do(cortisol = 0.5 × baseline)]

Intervening to reduce cortisol by 50% during stress events is estimated to 
attenuate hippocampal volume reduction by approximately 30–45%, based on 
the adjustment set {HPA_activity, glucocorticoid_receptor_density}.

Identifiability: PASSED — required confounders available in SCM
Affected nodes: cortisol → GR_activation → CA1_volume, CA3_volume
Delta: −0.38 SD (hippocampal volume reduction, 95% CI: −0.21 to −0.55)

Claim recorded: claim_002
Uncertainty: metric-bearing | Confidence: 0.61

Notice that the system reports identifiability status. If the required confounders were not available in the SCM, the system would downgrade to association-level and explain which confounders are missing. See Running interventions for a full explanation of the identifiability gate.

Trigger Rung 3 with a counterfactual question

Finally, ask a counterfactual — a question about what would have happened in an alternate world where a past condition was different. Counterfactuals operate on Rung 3 and require both an observed outcome and a hypothetical antecedent.

Example question — Rung 3

Would the hippocampal volume reduction have occurred if cortisol levels had been normal throughout?

The response will escalate to Rung 3 — Counterfactual:

Example response shape

[Rung 3 — Counterfactual | necessity analysis]

Under the counterfactual assumption that cortisol remained at baseline (normal 
range) throughout the stress period, the model estimates that hippocampal 
volume reduction would not have reached clinical significance in 71% of 
simulated individuals.

Necessity score: 0.71 (cortisol elevation is necessary for observed outcome 
in most cases)
Sufficiency score: 0.43 (cortisol elevation alone is not sufficient — 
co-activation of the inflammatory pathway is required in ~57% of cases)

Claim recorded: claim_003
Uncertainty: metric-bearing | Confidence: 0.58

Necessity and sufficiency scores are standard counterfactual measures from the Pearl framework. A necessity score above 0.5 means the intervention (normal cortisol) would have prevented the outcome in most individuals.

Review your claims in /claims

Navigate to /claims to see the Claim Ledger for this session. All three claims produced in this conversation — claim_001, claim_002, and claim_003 — are recorded with:

The rung they were produced on
The SCM cartridge and specific edges activated
Uncertainty label and confidence score
The full provenance trace (the reasoning chain)
Timestamp and session ID

Claims are permanent and auditable. You can export them individually or in bulk from the Claim Ledger interface. See Claim governance for details on how to share, challenge, or annotate claims.

What you covered

Step	Rung	Question type	Key output
Initial question	Rung 1	Association	Causal association with SCM support
Intervention	Rung 2	`do()` operator	Effect estimate with identifiability check
Counterfactual	Rung 3	Necessity/sufficiency	Probability of necessity and sufficiency

Next steps

Learn how interventions are evaluated and what happens when identifiability fails: Running interventions
Analyze supporting literature alongside your causal questions: Analyzing research papers with PDF upload
Understand how claims are structured and governed: Claim Ledger

​What you will build

​What you covered

​Next steps

What you will build

What you covered

Next steps