Bayesian-Style Update Session¶
Facilitated estimation session — instantiates Belief Revision Workflow
A working session that weighs new evidence against the prior belief — its diagnosticity, its source, and the base rate — to decide how far, and in which direction, confidence should actually move.
Most arguments about contrary evidence are really arguments about how much it should count. A Bayesian-Style Update Session is the working meeting that answers exactly that: it holds up the prior belief, asks how likely the new evidence would be if the belief were true versus if it were false, factors in the base rate and the reliability of where the evidence came from, and lands on a revised confidence. Its defining move is to treat the update as a weighing, not a debate — the size of the belief shift is a function of how diagnostic the evidence is, not of who argues hardest. It borrows the logic of Bayes' rule (prior, likelihood, posterior) but not necessarily its arithmetic: the confidence can be a probability, an ordinal band, or a plain "this should move us a lot / a little," so long as the reasoning about weight is explicit.[n1]
Example¶
A hospital tumor board holds the working belief that a patient's lung nodule is benign — it has been stable on two prior scans. A new PET scan lights up mildly. Rather than let the loudest voice win, the board runs the update explicitly. The prior: benign, roughly 80% (illustrative). The likelihood: how often does a truly benign nodule show this degree of PET uptake, versus a malignant one? The radiologist notes mild uptake is only weakly diagnostic — common in inflammation too. The base rate for malignancy in a nodule with this stability history is low. And the source: a single scan on a machine known to run hot. Weighed together, the evidence nudges confidence down — from ~80% benign to maybe ~65% — enough to warrant a short-interval follow-up, not enough to justify an invasive biopsy. The session's product is not a decision; it is a defensible new confidence with the weighing shown, so the next team can see why the belief moved as little as it did.
How it works¶
- Fix the prior first. State the current belief and its confidence before the new evidence is discussed, so the prior isn't quietly rewritten to fit the update.
- Elicit the likelihood, both ways. Ask how expected the evidence is under the belief and under its main rival. Evidence that is equally likely either way is not diagnostic, however dramatic it feels.
- Fold in the base rate. Anchor on how common the belief's truth is at baseline, guarding against base-rate neglect[1].
- Discount by source. Weight the evidence by the reliability and independence of where it came from — one shaky instrument, or three independent reads.
- Combine into a movement. Produce a directional, sized confidence change — up, down, or "unmoved" — with the weighing legible.
Tuning parameters¶
- Quantification level — full probabilities, ordinal bands, or qualitative "a lot/a little." Numbers sharpen reasoning but invite false precision; match the granularity to how much the evidence can actually bear.
- Prior-anchoring discipline — how firmly the prior is pinned before evidence enters. Strict pinning fights hindsight but can feel rigid.
- Likelihood elicitation — one estimator or an independent panel reconciled afterward. Panels catch blind spots but cost time and can anchor on the first number spoken.
- Base-rate insistence — whether a base rate is a mandatory input or an optional check. Mandatory is more honest; it can stall when no base rate exists.
- Update conservatism — a global tendency to move confidence boldly or cautiously per unit of evidence, tuned to how noisy the domain is.
When it helps, and when it misleads¶
Its strength is that it makes the weight of evidence the explicit object of discussion, which is where most revision failures actually hide — a dramatic but non-diagnostic anomaly gets over-weighted, or a quiet-but-decisive result gets ignored. Forcing the "how likely under each belief?" question exposes both.
Its failure mode is the theater of numbers: a session that dresses guesses as probabilities produces false precision, and a wrong or self-serving prior contaminates everything downstream ("garbage prior, garbage posterior"). The classic misuse is running the arithmetic to ratify a conclusion already chosen — tuning the likelihoods until the posterior lands where someone wanted it. The guarding discipline is to keep the update ordinal when the evidence is thin, to keep the prior and the likelihoods visible and separate from the result so they can be challenged, and to keep the weighing separate from the decision it feeds.
How it implements the components¶
evidence_weighting_frame— its core: a structured comparison of prior belief, likelihood-under-each-hypothesis, base rate, and alternatives, so evidence is weighed rather than rhetorically contested.source_credibility_check— the evidence is explicitly discounted by the reliability and independence of its source before it moves the belief.confidence_update— its deliverable is the revised confidence itself, expressed as a probability, band, or qualitative movement.
It does not make revision socially safe (threat_reduction — that's Dissonance-Safe Dialogue), file the written entry (revision_record — Belief Update Log), or supply the standing scope-and-monitoring bands that label the result (belief_scope_boundary — Confidence Scale).
Related¶
- Instantiates: Belief Revision Workflow — it is the workflow's weighing core, turning conflicting evidence into a sized confidence movement.
- Consumes: Confidence Scale supplies the graded vocabulary the resulting confidence is expressed in.
- Sibling mechanisms: Belief Update Log · Confidence Scale · Decision Log Update · Dissonance-Safe Dialogue · Learning Retrospective · Prediction Error Review
Editorial Notes¶
Form Classification¶
Form family: Communication, Facilitation & Learning
Rationale: A working session has participants state priors, compare evidence under competing beliefs, incorporate base rates and source quality, and agree how confidence should move, so its operative form is facilitated epistemic learning.
Nearest alternative: Assessment, Review & Assurance — The session yields an evidence judgment, but structured shared reasoning and belief revision among participants are the defining enactment.
Review outcome: Adjudicated after independent review; medium confidence.
Origin Attribution¶
Primary origin: Philosophy
Origin pattern: Cross-disciplinary synthesis
Present-day reach: Multi-domain
Rationale: Bayesian epistemology treats degrees of belief as credences and conditionalization as the norm governing how confidence should change when evidence arrives.
Related originating lineages:
- Psychology — Judgment and decision research supplies the base-rate-neglect and evidence-weighting debiasing problem.
- Statistics & Experimental Design — Bayesian inference supplies the prior, likelihood, evidence, and posterior structure borrowed by the session.
Review resolution: The Stanford Encyclopedia of Philosophy identifies conditionalization as a core Bayesian norm for changing degrees of belief in light of evidence, which more directly matches a qualitative confidence-update session than formal parameter estimation does. NIST supplies the prior-likelihood-posterior statistical form and Tversky and Kahneman supply the judgment-bias lineage, so philosophy is primary within a synthesized facilitation method.
Attribution caveat: The facilitated, qualitative session is an Encyclopedia operationalization rather than a named philosophical or statistical procedure.
Encyclopedia synthesis: The exact catalogued form synthesizes established practice rather than reproducing a single standard historical label.
Review outcome: Researched adjudication after independent review; high confidence.
Sources consulted:
- Stanford Encyclopedia of Philosophy — Bayesian Epistemology
- NIST — Bayesian Methods Class, Day 1
- Science — Judgment under Uncertainty: Heuristics and Biases
Notes¶
[n1] Bayes' theorem relates a posterior belief to the prior times the likelihood of the evidence under that belief. The "-style" in the name is deliberate: the session keeps the structure (prior → weighted evidence → posterior) while allowing qualitative confidence, so it avoids demanding precision the evidence cannot support. ↩
References¶
[1] Kahneman, D., & Tversky, A. "On the Psychology of Prediction". Psychological Review 80(4), 237–251 (1973). Documents base-rate neglect as intuitive prediction that is insensitive to prior probabilities. registry ↩