update-pairwise-rating
Purpose
Update latent candidate ratings and uncertainty from a pairwise outcome using an explicitly chosen rating model.
Input contract
required: [prior_ratings, pairwise_outcome, rating_model, update_parameters]
optional: [rating_history, draw_policy, uncertainty_floor]
constraints: [model, outcome direction, confidence, and parameter values are recorded]
Procedure
- Validate the pairwise outcome and retrieve the two prior rating states.
- Apply the declared Bradley-Terry, Elo, Glicko, TrueSkill, or equivalent update equations.
- Update both ratings and uncertainty, preserving the pre-update state and the reason for the magnitude.
- Emit the updated ratings, uncertainty change, and consistency flags for downstream audit.
Output contract
produces: [updated_ratings, uncertainty_updates, update_trace, consistency_flags]
delta_fields: [findings, evidence_updates, uncertainties, decisions]
Quality gates
- The update is reproducible from prior ratings, outcome, model, and parameters.
- A weak or low-confidence outcome cannot cause a larger update than the declared model permits.
Failure and counterexamples
Do not update ratings when the pairwise record is tied to the wrong candidate IDs or model. Do not silently reset rating history.
Provenance map
- resolved: dynamic-tracking
1---2name: update-pairwise-rating3description: "Update candidate latent ratings and uncertainty from sequential pairwise outcomes using an explicitly chosen Bradley-Terry/Elo/Glicko/TrueSkill-style update model."4metadata:5 source: https://skillmd.com/skills/yogsoth-ai/update-pairwise-rating6---78# update-pairwise-rating9## Purpose10Update latent candidate ratings and uncertainty from a pairwise outcome using an explicitly chosen rating model.11## Input contract12```yaml13required: [prior_ratings, pairwise_outcome, rating_model, update_parameters]14optional: [rating_history, draw_policy, uncertainty_floor]15constraints: [model, outcome direction, confidence, and parameter values are recorded]16```17## Procedure181. Validate the pairwise outcome and retrieve the two prior rating states.192. Apply the declared Bradley-Terry, Elo, Glicko, TrueSkill, or equivalent update equations.203. Update both ratings and uncertainty, preserving the pre-update state and the reason for the magnitude.214. Emit the updated ratings, uncertainty change, and consistency flags for downstream audit.22## Output contract23```yaml24produces: [updated_ratings, uncertainty_updates, update_trace, consistency_flags]25delta_fields: [findings, evidence_updates, uncertainties, decisions]26```27## Quality gates28- The update is reproducible from prior ratings, outcome, model, and parameters.29- A weak or low-confidence outcome cannot cause a larger update than the declared model permits.30## Failure and counterexamples31Do not update ratings when the pairwise record is tied to the wrong candidate IDs or model. Do not silently reset rating history.32## Provenance map33- resolved: dynamic-tracking