update-confidence-from-evidence
Purpose
Update confidence in a concept, relation, claim, or model component from new supporting or contradicting evidence.
Input contract
required: [target_record, prior_confidence, new_evidence, update_rule]
optional: [support_weights, contradiction_weights, confidence_floor]
constraints: [each update cites evidence direction, provenance, weight rule, and prior state]
Procedure
- Verify target identity and freeze the prior confidence and evidence ledger.
- Classify each new item as supporting, contradicting, or non-informative under the declared rule.
- Apply the update rule, record the weight and direction, and cap the result to the permitted confidence range.
- Emit the updated confidence, evidence trace, and unresolved conflicts.
Output contract
produces: [updated_confidence, evidence_trace, conflict_register, update_rationale]
delta_fields: [findings, evidence_updates, hypothesis_updates, uncertainties, decisions]
Quality gates
- Every confidence change is reproducible from the prior state, evidence classification, and update rule.
- Contradicting evidence cannot be omitted from the ledger merely because the net score rises.
Failure and counterexamples
Do not treat repeated copies of one source as independent support. If evidence polarity is ambiguous, leave confidence unchanged and mark uncertainty.
Provenance map
- resolved: confidence-update
- resolved: confidence-scoring
1---2name: update-confidence-from-evidence3description: "Update confidence in a concept, relation, claim, or model component from new supporting/contradicting evidence with explicit rationale."4metadata:5 source: https://skillmd.com/skills/yogsoth-ai/update-confidence-from-evidence6---78# update-confidence-from-evidence9## Purpose10Update confidence in a concept, relation, claim, or model component from new supporting or contradicting evidence.11## Input contract12```yaml13required: [target_record, prior_confidence, new_evidence, update_rule]14optional: [support_weights, contradiction_weights, confidence_floor]15constraints: [each update cites evidence direction, provenance, weight rule, and prior state]16```17## Procedure181. Verify target identity and freeze the prior confidence and evidence ledger.192. Classify each new item as supporting, contradicting, or non-informative under the declared rule.203. Apply the update rule, record the weight and direction, and cap the result to the permitted confidence range.214. Emit the updated confidence, evidence trace, and unresolved conflicts.22## Output contract23```yaml24produces: [updated_confidence, evidence_trace, conflict_register, update_rationale]25delta_fields: [findings, evidence_updates, hypothesis_updates, uncertainties, decisions]26```27## Quality gates28- Every confidence change is reproducible from the prior state, evidence classification, and update rule.29- Contradicting evidence cannot be omitted from the ledger merely because the net score rises.30## Failure and counterexamples31Do not treat repeated copies of one source as independent support. If evidence polarity is ambiguous, leave confidence unchanged and mark uncertainty.32## Provenance map33- resolved: confidence-update34- resolved: confidence-scoring