- Add rewrite_v2_nemo.txt: less directive on scene closure, natural endings
- Add repair_v2_nemo.txt: same approach, avoids mechanical conclusion formulas
- PromptStore.render() now accepts prompt_profile kwarg with fallback to v1
- Config: [prompt_profiles] maps model to profile (e.g. mistral-nemo -> v2_nemo)
- Mark qwen2.5:7b as accepted baseline Ollama via Mascarade P2P
- Close 3 P0 TODOs
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Route everything through Mascarade :8100 (Mistral API, Apple CoreML, Ollama P2P)
- New local ref: qwen3-4b-instruct-2507-q4f16 (3x faster than previous)
- Add mistral:mistral-large-latest to priority_models and french_models
- Point ollama_openai_base_url to Mascarade instead of llama-server :8091
- Update auto-sync with current model status
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Updated the rewrite prompt to remove markdown elements such as titles, bullet points, and code fences, ensuring a cleaner narrative output.
- Enhanced the GenerationPipeline tests to verify that code fences and chapter titles are stripped before manuscript promotion.
- Adjusted NextLots tests to include new functionality for command runners and added tests for deduplication of headings and timeout handling.
- Introduced new documentation files detailing project context and execution plans for March 2026, outlining current project status and objectives.
- Added operational memory documents to summarize project state and decisions for resuming work on the ai-novel-engine.
L'électron rare