CharacterBench 0.2.1-alpha

Activation patch for the 0.2 external alpha.

• Requires an explicit Ollama model selection instead of silently choosing the first installed model.
• Disables Run until a model is chosen.
• Adds a first-run hint recommending a smaller text model.
• Preserves the 0.2 character-card import, starter-check review, and local baseline regression workflow.

The release ZIP passed clean extraction, self-test, doctor, privacy scan, and a real browser run against a local Ollama model.

SHA-256: 81d4f1b386e39a372306a75cb7b9792374958c57cd26f3d1706227cd4a30d1e1

CharacterBench 0.2.0-alpha

CharacterBench 0.2.0-alpha turns the UI preview into the main external-testing workflow.

Highlights:

• Import standard Character Card JSON/PNG files
• Generate and review conservative starter checks
• Add simple custom mention/avoid checks
• Run against local Ollama models
• Save a privacy-minimal local baseline
• Compare later runs for regressions and improvements
• Apache-2.0 licensed

The UI is local-first and binds to 127.0.0.1. CharacterBench adds no telemetry. Scores remain diagnostic signals, not general model-quality rankings.