dsh-evidence-arena: A Multi-Model Programming Comparison Workbench for DeepSeek Harness
`dsh-evidence-arena` is a community plugin for DeepSeek Harness that addresses the lack of evidence and cumbersome switching in multi-model parallel programming comparisons. It supports running two or three Builders in parallel within an isolated environment to solve the same task, uses an independent Reviewer to perform logic and security checks, and provides reproducible checkpoints and per-file diffs. The plugin emphasizes an evidence-first approach, writing back only human-verified code to the original workspace, and operates in a zero-tool sealed environment to ensure safety. It is compatible with DSH version 2026-08-19 and is suitable for model performance evaluation and code generation quality validation.
Read More