yogsoth-ai Dynamic Tracking
Purpose
Maintain live rankings that update incrementally as new comparisons arrive. Suitable for ongoing evaluation, A/B testing, and systems where candidates enter/exit over time. Tracks rating uncertainty and temporal dynamics.
When to use
- Continuous stream of comparisons (not batch)
- Candidates added/removed over time
- A/B testing or tournament-style evaluation
- Need to track rating trajectories and momentum
Budget
| Resource | Allocation |
|---|---|
| Comparisons per update | 1+ (online update) |
| Rating recalculation | After each comparison or micro-batch |
| Stability window | Last 20-50 comparisons for convergence check |
| Decay handling | Increase uncertainty for inactive candidates |
State Ledger
candidates: {} # id → {mu, sigma, last_active, history: []}
comparison_log: [] # [{pair, winner, timestamp, context}]
method: "" # elo | glicko2 | trueskill2 | whr
parameters: {} # K-factor, tau, beta, etc.
iteration: 0
window_size: 30
convergence: {stable: false, score: 0.0}
Available Tactics
- adaptive-pair-selection — select next matchup for maximum information
- consistency-audit-loop — periodic transitivity audit
Available SOPs
- pair-selector
- comparison-executor
- rating-update
- convergence-check
- cycle-detection
- ranking-synthesis
Execution Guidance
- Initialize new candidates with default prior
- For each incoming comparison: a. Run comparison-executor (or accept external result) b. Run rating-update with online method c. Every window_size comparisons, run convergence-check
- Periodically run cycle-detection on recent window
- Increase sigma for candidates inactive > threshold
- On demand: run ranking-synthesis for current snapshot
Output Format
current_ranking:
- {rank: 1, candidate: "...", mu: 1842, sigma: 45, games: 23}
- {rank: 2, candidate: "...", mu: 1798, sigma: 52, games: 18}
method: glicko2
total_comparisons: 342
last_updated: "2026-05-19T14:30:00Z"
rating_period: 7
top_movers:
- {candidate: "...", delta: +120, last_5: "WWWLW"}
stability_score: 0.88
Available Tactics
Optional, no fixed order; the final leaf is always a sop.
| Tactic | When to use |
|---|---|
| adaptive-pair-selection | Iteratively select maximally informative pairs, execute comparisons, update ratings, and check convergence until ranking stabilizes. |
| consistency-audit-loop | Detect preference cycles, localize inconsistent judgments, request corrections, and recompute ratings until consistency threshold is met. |
Available SOPs
Optional, no fixed order; the final leaf is always a sop.
| SOP | When to use |
|---|---|
| ranking-synthesis | Produce the final ranking artifact from converged ratings and consistency report. |