Add GPT-6 Astra 1× + ICL native evaluation logs
Adds matching metadata, all 260 sanitized single-episode native Bench2Dex evaluator rows and original summaries, and upstream-generated per-task logs for the SenseNova-RoboRSI GPT-6 Astra 1× + ICL Full/unseen entry.
source_manifest.json records original and sanitized SHA-256 hashes. Only private absolute scene and asset paths were made relative; the log structures and all numeric evaluator values are unchanged. Task 61 episode 8 is explicitly labeled same_index_rerun: its original native row and summary were lost, and the first valid same-index rerun passed the pinned named-results certifier. It was successful with LSCR 1.0, matching the preserved historical result. The 260-slot aggregate is 68/260 stable successes and 48.769231% LSCR.
The paired GitHub results, README, and logo PR is https://github.com/Bench2Dex/Bench2Dex-Leaderboard/pull/2. The submission validator reports zero errors and 26 expected 10-versus-50-episode coverage warnings. SenseNova-RoboRSI authorized public release.