Skip to main content

Documentation Index

Fetch the complete documentation index at: https://wuweism.com/llms.txt

Use this file to discover all available pages before exploring further.

Wu-Weism will not record a claim that cannot be tested. Before any hypothesis or counterfactual finding reaches the Claim Ledger, it passes through a set of gates — automated checks that enforce the logical conditions required for a claim to be scientifically meaningful. The most important of these is the Falsifiability Gate.

​
What the Falsifiability Gate does

The Falsifiability Gate evaluates whether a claim specifies — even implicitly — what evidence would disconfirm it. A claim that cannot, in principle, be proven wrong is not a scientific claim. It is an assertion. Wu-Weism enforces this as a hard gate: claims that fail the falsifiability check are not recorded in the Claim Ledger with a pass status. Instead, the response is flagged and the gate decision is written as fail with a rationale explaining what is missing.
A failed gate does not mean your research question is invalid — it usually means the question needs to be made more specific. The rationale in the gate decision tells you exactly what is required.

​
Why it matters

Unfalsifiable claims are unscientific in a precise sense: if no possible observation could contradict the claim, then the claim carries no predictive content. It cannot guide decisions, cannot be updated by new evidence, and cannot be used as input to a causal model. By enforcing falsifiability before recording claims, Wu-Weism ensures that your Claim Ledger contains only findings that are:
  • Testable — there exists some observation that would, if true, count as evidence against the claim.
  • Specific — the claim refers to a defined population, intervention, time horizon, or measurable outcome.
  • Actionable — the claim can be used as a premise in further causal reasoning.

​
How the gate works

1

Claim is produced

A response is generated by the causal reasoning engine — whether from Causal Workbench dialogue, Hybrid Synthesis, or Legal Causation analysis.
2

Falsifiability is evaluated

Before the claim is committed to the Claim Ledger, the Falsifiability Gate analyzes the claim text for:
  • Whether a measurable outcome is specified.
  • Whether a defined condition or intervention is referenced.
  • Whether a disconfirming observation is logically possible given the claim’s structure. The gate assigns a score from 0.0 to 1.0 and produces a pass, warn, or fail decision.
3

Gate decision is recorded

The decision — including the gate name, outcome, rationale, and score — is written to the claim’s gate_decisions field.
4

Claim is recorded or blocked

If the gate passes (pass or warn), the claim is recorded in the Claim Ledger. If it fails, the claim is not recorded with active status, and the response surfaces the failure rationale so you can refine your question.

​
Gate decision structure

{
  "gateName": "falsifiability",
  "decision": "fail",
  "rationale": "The claim asserts a general tendency without specifying a measurable outcome, a defined population, or conditions under which it would be false. To pass, specify what observation would contradict this claim.",
  "score": 0.21
}
FieldDescription
gateNameThe gate that evaluated this claim: falsifiability, identifiability, or novelty.
decisionpass — claim meets the criterion. warn — claim meets the threshold but with caveats noted in the rationale. fail — claim is blocked.
rationaleA plain-language explanation of why the claim passed, was flagged, or was blocked.
scoreThe gate’s numeric assessment, 0.0–1.0. Scores below approximately 0.4 result in fail.
To write claims that pass the Falsifiability Gate, make your claim testable by specifying what would disconfirm it. For example, instead of “higher marketing spend improves outcomes”, write: “Increasing marketing spend by 10% will increase customer acquisition by at least 5% over 90 days in the EMEA segment — a result below 2% would disconfirm this.” The more precisely you define the outcome, the population, and the threshold, the higher the falsifiability score.

​
What happens to blocked claims

When a claim fails the Falsifiability Gate:
  • The claim is not written to the Claim Ledger with status: active.
  • The workbench response includes the gate decision and rationale, so you can see why the claim was blocked.
  • You can refine your question and resubmit — the revised claim will be evaluated independently.
Blocked claims are not silently discarded. The failed gate decision is logged for your session so you have a record of what was attempted and why it was rejected. This is part of Wu-Weism’s commitment to a complete audit trail.

​
The Common Sense Governor

Alongside the scientific validity gates, Wu-Weism applies a Common Sense Governor: a policy layer that declines requests enabling threats, coercion, deception, or targeted harm — regardless of how they are framed scientifically. If your request is declined by the Common Sense Governor, you will see the following response:
“I can’t help with drafting or strategy that enables threats, coercion, deception, or targeted harm.”
The Common Sense Governor operates before the causal reasoning engine and is not a scientific gate — it is an ethical constraint on the platform’s use. No trace record or claim is created for requests declined at this layer.

​
The Identifiability Gate

For interventional analyses (do-calculus), Wu-Weism applies an additional gate before running the computation: The Identifiability Gate checks whether the causal effect you are asking about can actually be identified — that is, estimated without bias — given the set of confounders available in your SCM. If the causal effect is not identifiable with the available adjustment set, the gate fails and the intervention is not run. The rationale explains which confounders are missing or why identification fails under the current graph structure. 
{
  "gateName": "identifiability",
  "decision": "fail",
  "rationale": "The causal effect of advertising_spend on revenue is not identifiable. An unmeasured common cause (consumer_sentiment) lies on a back-door path that cannot be blocked by the current adjustment set. Add consumer_sentiment as an observed variable or use a front-door criterion.",
  "score": 0.18
}
This gate protects you from running analyses that would produce numerically plausible but causally invalid results.

​
The Novelty Gate (Hybrid Synthesis)

In Hybrid Synthesis, a third gate applies to synthesized hypotheses: The Novelty Gate evaluates whether a synthesized hypothesis is genuinely novel relative to prior art in your uploaded document corpus and Wu-Weism’s scientific knowledge base.
DecisionMeaning
passThe hypothesis is not substantively represented in prior literature and represents a candidate novel finding.
warnThe hypothesis has partial overlap with prior art; the rationale identifies the closest existing work.
failThe hypothesis is substantially anticipated by existing literature. It is recorded but tagged as a replication rather than a novel finding.
The Novelty Gate does not block claims — it classifies them. A fail on the Novelty Gate means your hypothesis is corroborated by prior art, which is scientifically valuable, but should not be presented as an original contribution.

​
All gates at a glance

Evaluates whether the claim specifies a testable, potentially disconfirmable condition. Hard gate: fail blocks the claim from active status in the Claim Ledger.
Checks whether the causal effect can be identified with the available confounder adjustment set before running do-calculus. Hard gate: fail prevents the intervention from being computed.
Evaluates whether synthesized hypotheses are novel relative to prior art. Soft gate: fail classifies the claim as a replication rather than blocking it.
Policy layer that declines requests enabling threats, coercion, deception, or targeted harm. Not a scientific gate — operates before the reasoning engine. No trace or claim is created for declined requests.

​
Related

Claim governance

Gate decisions are recorded in every claim record — learn how to read and export them.

Provenance and audit trails

Understand the full reasoning trail behind every claim that passes the gates.