Replace the static checking template with generator-driven staged progress before the MiniCPM call and keep the final report render unchanged.
Co-authored-by: Codex <codex@openai.com>
Render the trusted-person copy plan as a light preview block, copy from a button data attribute, and humanize additional model tactic labels.
Co-authored-by: Codex <codex@openai.com>
Increase the default MiniCPM generation budget to avoid truncated JSON, humanize model-style labels in the report UI, and keep the copy-plan preview readable.
Co-authored-by: Codex <codex@openai.com>
Make the v3 MiniCPM LoRA adapter the default ZeroGPU model path, add v2/v3 hard-case data generation, record Modal training/eval evidence, and update submission docs.
Co-authored-by: Codex <codex@openai.com>
