olmo_tap.final_evals.elo

Tournament 1 Elo evaluation harness.

Configuration-level Elo for four entrants (base OLMo, Hydra+Security, Hydra+Security+Robustness, Hydra+PoE) with Claude as an LLM judge.

Modules

elo_engine	Permutation-averaged Elo with K-factor sensitivity sweep.
entrants	Entrant definitions for the configuration-level Elo tournament.
generate	Entrant -> response generation for the configuration-level Elo tournament.
judge	LLM-judge pipeline using the Anthropic Batch API with prompt caching.
match_builder	Pairwise match list construction for the configuration-level Elo run.
prompts	Prompt-bank construction for Tournament 1.
report	Tournament reporting helpers.
run_tournament	Tournament orchestrator: response cache → judges → Elo → artifacts.
scripts	One-off scripts for the Elo evaluation harness.
types	Shared, dependency-light data types for the Elo tournament pipeline.