Rtisan PR43 tidy-r benchmark share bundle

This is a sanitized, filtered package for sharing. It contains only the cleaned logical benchmark attempts: success and verification_failure. Original raw failure / provider_error attempts, debug runs, pid files, and duplicate rerun attempts are excluded.

Key files:

• manifest.json: bundle metadata and task list
• results/attempts.csv: one cleaned row per logical attempt
• results/clean_counts.csv: cleaned status counts by model slot
• results/pass_rates_by_model_skill.csv: pass rates grouped by model and skill
• results/pass_rates_model_skill_comparison.csv: tidy-r vs no-skill comparison by model
• results/dbs/*.db: SQLite DBs filtered to selected cleaned attempts only
• trials/: selected trial directories for those attempts
• tasks/r/: copied core task definitions used in the benchmark
• skills/tidy-r/: copied skill folder used in the benchmark

Pass rate is success / total; verification_failure means the agent produced an answer but did not pass verification.