Fix evaluate_all_pai to pass --skip-llm-judge to task main() functions

Fix eval_dvc.py main() to support --skip-llm-judge flag

Fix DVC evaluation to compute temporal F1 when --skip-llm-judge is set

Add semantic similarity matching for Next Action evaluation

Fix CVS_acc to use raw accuracy instead of component_balanced_accuracy

Fix TAL overall metrics computation and extraction

Use STG overall metrics instead of placeholder

Fix STG evaluation to extract bbox_dict from struc_info

Add placeholder for STG metrics when evaluation returns empty

Fix STG and next_action metric printing in overall mode

Show evaluation metrics even in silent mode (--grouping overall)

Keep --skip-llm-judge flag, set caption metrics to 0 when skipped

Remove --skip-llm-judge flag for production

Add process.wait() to ensure returncode is set

Add matplotlib to requirements

Filter DVC/VS/RC from tasks list when skip-llm-judge is set

Add back --skip-llm-judge for faster testing

Remove skip-llm-judge - use semantic similarity fallback

Add --skip-llm-judge flag for faster evaluation

Remove debug messages - system working correctly

Add debug inside loop to verify entry

Add debug at line 741

Add debug before task list print

Remove DEBUG filter - allow debug messages through

Debug before/after imports

Add debug at function entry and after analyze

Add explicit flush debug

update

update

update

update

update

add merged format

update

update name

update

Add cache/ to gitignore and document test results

Implement secure ground truth with prediction-only submission format

Remove ground_truth.json

Upload ground_truth.json

Add support for pre-computed LLM judge scores

Make validation flexible: accept both 'answer'/'response' and 'gnd'/'ground_truth' field names

Add contact column back to leaderboard display

Fix leaderboard display: remove average column, show 10 metrics, fix Tasks tab

Update evaluation metrics and leaderboard display

Complete evaluator fixes for all 8 tasks

Fix syntax errors and add TAL wrapper functions

Fix evaluate_all_pai.py to use eval_caption_llm_judge

Consolidate evaluation scripts and remove hardcoded paths

Remove all captioning_metrics dependencies