Add action categories counts to the row.

Add big questions results to the row.

Add results from a specific data_id to the row.

Add hallucination counts to the row.

Add .html extensions to files that don't have .html extensions but contain HTML

Aggregate results from multiple games into a single match result. Args: game_results: List of (result, white_agent, blac

Internal implementation of line counts analysis with timeout

Apply one SAN move for the side to move. Returns (new_board, from, to, new_ep).

Is square (r, c) attacked by a piece of the given color?

Build combined 'ALL' matrix with normalized scores from all games.

Build match pairings for sims_per_round simulations. Strategy: Round-robin style - pair agents and repeat as needed. For eac

Calculate time running for a job Returns: Tuple of (formatted_string, seconds)

Remove API endpoint files that were created during freezing

Clean up compute lock after use to avoid memory leak

Commit changes to the agent's codebase.

Compile an agent's code in the game container. Args: agent: The agent to compile idx: Index for naming the output dir

Compile an agent's code in the game container. Args: agent: The agent to compile idx: Index for naming the output dir

Recompile each agent's engine in the game container and return engine paths. Returns: dict mapping agent name to engine

Create a dummy agent for matrix evaluation.

Create a violin plot showing the distribution of Elo scores for each model.

Create a matrix plot showing the percentage of times each model achieves each rank.

Evaluate a single matrix cell using pre-initialized agents. Returns (i, j, result).

Extract modified file paths from a git diff and get their full content. Returns a dict mapping file path to full file content. Only in

Internal implementation of batch jobs fetching with timeout

Internal implementation of find_all_game_folders with timeout

Generate AWS console link for a job

Generate emagedoc.xyz viewer link for a job

Generate S3 console link for a job

Get AWS command for a job

Get the current commit hash.

Get or create a lock for a specific cache key

Read diff data from changes_r{round}.json file. Returns None if file doesn't exist.

Get round info for a job

Calculate score for a round. Returns (p1_score, p2_score) where each is 0.0, 0.5, or 1.0.

Pre-initialize agents for all rounds for each player.

Initialize a pool of game objects for parallel execution.

Initialize a row with basic instance metadata and NaN values for all data types.

Count added + removed lines in a unified diff string using unidiff.PatchSet. Returns {} if the patch cannot be parsed.

Count added + removed lines in a unified diff string using unidiff.PatchSet. Returns 0 if the patch cannot be parsed.

Load existing progress from matrix.json if it exists.

Load match pairings from stored JSON file. Returns: Dict mapping match_idx to (agent1_name, agent2_name) tuple

Extract the SAN token list from PGN movetext (strip headers/comments/numbers/result).

Negative log-likelihood for Bradley-Terry model with L2 regularization.

Parse all games from a PGN file. Args: pgn_content: Content of the PGN file (may contain multiple games) Return

Parse all available log files in the tournament directory (metadata only, no content)

Parse a single simulation log to determine the winner. Args: log_path: Path to the log file sim_meta: Simulation meta

Extract a display model name from a player config (mirrors the viewer).

Create line plots showing Elo vs max rounds for each game.

Create line plots showing Elo vs round for each game.

Copy agent codebases into game's container

Can `piece` at (fr,fc) pseudo-legally move/capture to (tr,tc)? (path + pattern only)

Read all match result files and parse them. Returns: List of (winner, agent1_name, agent2_name) tuples winne

Internal implementation of render_game_viewer with timeout

Validate the optional ``ladder_rules`` block and return ``(min_round_win_fraction, win_last_k)``. Defaults: win at least ``min_round_win_fraction

Save plot in both PDF and PNG formats.

Save plot in both PDF and PNG formats.

Add cache_control to the last block of the last message.

Algebraic square (e.g. 'e4') -> (row, col), row 0 = rank 8.

Command to start the submission's server on $PORT. Submissions are either the Python starter (main.py) or any other language via a run.sh laun

Remove the 'text' key we add to non-text blocks for interactive display and 'caller' (None) that model_dump() emits from newer SDK versions.

Wait for ports to be served, up to timeout seconds. Returns: List of ports that are actually served after timeout.

Write all changes to a JSON file in players/{name}/changes_r{round}.json

Aggregate win rates across multiple games using micro-averaging

Calculate file reuse ratio per player across all arenas.

Analyze line counts across all rounds for all files that appear in changed files

Check trajectory logs for commands accessing /opponent_codebases/

Analyze scores per round for each competitor from round_stats in metadata.json. Scores are calculated as wins + 0.5*ties, same as in the table

Track opponent code access by round number (early vs late tournament)

Calculate throwaway files per player, averaged across tournaments.

Analyze all tournaments in a directory for round-to-round recovery patterns. Args: log_dir: Path to directory containing tournament