Add Mora V0 evaluation logs

Mora V0 is a vision-language-action (VLA) foundation model developed by 墨奇智能 (Morphi Robot). We perform task-specific post-training on the Mora V0 base model using Bench2Dex robot demonstrations. We evaluate the resulting policies across all 26 official Bench2Dex task and robot settings.

Coverage is 26 official tasks, four generalization channels per task and 50 episodes per channel. The one selected Mora V0 checkpoint per task is disclosed with its SHA-256. The local consistency audit recomputes 33.2308% SR, 52.3083% LSCR and 0.973682 efficiency. Efficiency is defined on 25 tasks with successes. Task 27 has 0 successes and is included in SR and LSCR.

The channel JSONL files concatenate finalized native episode rows in index order. Each original worker shard and its summary are retained separately. The Bench2Dex native writer regenerated summaries for split channels.

Seven episode indices in tasks 34 and 60 initially produced an angle-decoding ValueError. The archived evaluator retried those indices with the same seed. The finalized record contains three successes and four failures. We report the finalized result and retain the seven initial error records in native_shards for review. Counting the initial errors as failures gives 1,725/5,200 (33.1731% SR). We request maintainer review of this recovery convention.

Companion policy PR: https://github.com/Bench2Dex/Bench2Dex-Leaderboard/pull/4.

This replaces #3 with a revised model introduction. Scores, evaluation protocol and evidence are unchanged. The payload upload is being completed before publishing this draft.

Upload complete: all 674 files at commit 1de49926c48c865c8a6dd16662ba6ecc584ac43b match the reviewed submission package by file hash and size. The revised model introduction is included. Companion GitHub PR: https://github.com/Bench2Dex/Bench2Dex-Leaderboard/pull/4. This PR replaces #3.