Diagnostic Safeguards
Diagnostic safeguards prevent the model from turning every gap into an accusation.
Raw Gap vs Warranted Diagnosis
The model should distinguish a raw Perceived Evidence to Assigned Credence gap (EP-CA) from a warranted diagnosis of Core Irrationality. A raw gap is evidence that something deserves attention. It is not yet proof of bad faith.
The improved interactive model therefore tracks Excess Core Irrationality: the portion of the EP-CA gap that remains after allowing for warranted uncertainty. This keeps the model from treating every small deviation, noisy estimate, or high-uncertainty case as excess misalignment.
The Responsibility Filter
Before moving from diagnosis to blame, ask four questions. Did the agent have access to better evidence? Did they have the skill and time to process it? Were there emotional, institutional, or social pressures distorting the update? Did they resist repair once the problem became visible?
This filter preserves the project’s central distinction. It still holds people accountable for motivated override, but it avoids treating ignorance, trauma, manipulation, fatigue, or poor evidence access as if they were all the same thing.
Action Thresholds Are Separate
Credence is not the same as action. A person may rationally act before certainty when stakes are high, delay action when costs are high, or keep investigating when the evidence is not yet decision-ready.
The model is strongest when it separates three questions: what confidence is warranted, what confidence the agent assigns, and what action threshold the situation requires.
Use With Diagnostic Humility
Because Objective Evidence (E0), Perceived Evidence (EP), Assigned Credence (CA), and Deep Rationality (SD) are often estimates, the model should report diagnostic confidence as well as diagnostic type.
The safest use is comparative: which gap appears largest, which repair would reduce it, and what observation would change the diagnosis?