When MODEL_X references (MODEL_0, MODEL_1, etc.) are used, the code now
properly looks up actual model names from the dynamic registry (MODEL_CLIENT_MAP)
instead of hardcoding "model" or requiring MODEL_NAME_X environment variables.
Changes:
- app.py: Look up models from MODEL_CLIENT_MAP for the specified endpoint
- guarded_ai.py: Query endpoints for actual model names at initialization
- guarded_ai.py: Use dynamic registry for MODEL_X lookups
This fixes the "model not found" error when using activities with MODEL_X
references like activity37.
Respects existing stable configuration where:
- MODEL_0 = Hermes (default for classification and feedback)
- MODEL_1 = Qwen (for code generation)
- MODEL_2 = GPT
Updated:
- All function defaults in activity.py: MODEL_1 -> MODEL_0
- activity37: Uses MODEL_0 for classification, MODEL_1 for code feedback
This works with the existing environment variable setup without requiring changes to vars.sh.
Critical fix for guarded_ai.py:
- Add MODEL_NAME_{n} environment variable support
- Fixes hard-coded "model" string that breaks Azure OpenAI and other endpoints
- Falls back to "model" if MODEL_NAME_{n} not specified
- Some endpoints require actual deployment name in model parameter
Validator improvement:
- Allow feedback_prompts as alternative to feedback_tokens_for_ai
- Prevents false warning when using metadata_feedback_filter with new prompt system
Documentation:
- Added MODEL_NAME_{n} examples to CLAUDE.md
- Documented that Azure and similar endpoints need this variable
All 8 activities validated: 0 errors, 0 warnings
Transformed planning document into comprehensive completion report:
- Status: 8 activities completed (30-37), 6,112 lines of YAML
- Documented new classifier_model and feedback_model feature
- Added model setup guide for Qwen3-Coder-30B
- Detailed activity summaries with special features
- Technical architecture and implementation decisions
- Usage examples and future enhancements
Key highlights:
- All activities validated with 0 errors
- Dual-model architecture explained
- Activity 37 flagship feature: universal programming language support
- Hermes excellence in role-playing scenarios
Added detailed comments showing how to use the recommended model:
- hf.co/unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF:Q4_K_M
- Setup instructions for llama.cpp (with GPU offloading)
- Alternative setup with ollama
- Environment variable configuration examples
This 30B parameter model is specifically optimized for code generation
across all programming languages, making it perfect for the universal
programming activity.
Changes:
- Enhanced get_openai_client_and_model() to support MODEL_X references
- Added model parameter (default "MODEL_1") to all AI functions:
- categorize_response()
- generate_ai_feedback()
- provide_feedback()
- provide_feedback_prompts()
- translate_text()
- Updated simulate_activity() to:
- Read classifier_model and feedback_model from YAML
- Support step-level model overrides
- Pass appropriate models to classifier vs feedback functions
This ensures the CLI simulation tool matches the production activity.py behavior.
Allow activities to specify separate models for classification and feedback:
- classifier_model: Used for categorizing user responses into buckets
- feedback_model: Used for generating AI feedback and translations
Both fields can be set at activity level (defaults) and overridden at step level.
Updated activity37 to use:
- MODEL_1 (Hermes) for classification
- MODEL_3 (Qwen 3 Coder) for feedback
This allows using specialized models for different tasks, e.g., fast classification
with accurate feedback generation from domain-specific models.
Created 3 comprehensive educational activities without embedded Python:
1. activity35-american-history.yaml - Advanced American History for gifted students
- Founding principles and Constitutional design
- Civil War causes and Reconstruction failure
- Civil Rights Movement strategies
- Primary source analysis and critical historical thinking
- Connects past to present issues
2. activity36-biblical-history.yaml - Biblical History & Ancient Near East
- Ancient Near Eastern context (Mesopotamia, Egypt, Canaan)
- Archaeological evidence and historical reconstruction
- Israelite history (Exodus, Monarchy, Exile)
- Roman period and early Christianity
- Foundation myths vs historical facts
- Cultural adaptation and religious transformation
3. activity37-programming-languages.yaml - Universal Programming Concepts
- Student chooses ANY programming language (Python, C++, COBOL, anything)
- AI adapts all examples/feedback to chosen language via metadata
- Covers: stdout/output, variables, data types, control flow, loops, functions
- All examples use stdout to display messages
- Concepts applicable to every language
- Language-specific syntax provided by AI
All activities:
- Use only YAML features (no embedded Python)
- Validate successfully with 0 errors
- Provide sophisticated educational content
- Use AI feedback for personalization
- Include critical thinking and reflection
- Track progress via metadata
Total: 8 new educational activities across 2 commits (5 from previous commit + 3 now)
Created 5 comprehensive educational activities that use only YAML features
(buckets, transitions, metadata operations, AI feedback) without Python scripts:
- activity30-logic-puzzles.yaml: Critical thinking through deductive reasoning,
contrapositives, syllogisms, and knights/knaves puzzles
- activity31-scientific-method.yaml: Learn scientific method through historical
case studies (Semmelweis, Newton) and experimental design principles
- activity32-world-geography.yaml: Choose-your-own-adventure journey exploring
continents, countries, capitals, and cultural facts
- activity33-environmental-science.yaml: Role-playing as environmental consultant
making sustainability decisions on transportation, energy, land use, waste, and food
- activity34-media-literacy.yaml: Develop critical media consumption skills,
evaluate sources, recognize bias, fact-check claims, and spot manipulation
All activities:
- Follow existing YAML schema and validate successfully
- Use Socratic buckets for educational feedback
- Include set_language support
- Track progress via metadata
- Provide AI-generated personalized feedback
- Are educational, engaging, and progressively challenging
- Include final reflection steps
Also added NEW_ACTIVITIES_PLAN.md documenting the planning process and
design rationale for each activity.
- Remove duplicate messages from section transitions like activity14
- Fix coin categorization issue - 'use coin' was being misclassified as 'use_key_and_password'
- Add section_4:step_2 for post-safe-opening state with proper coin slot options
- Update tokens_for_ai to properly distinguish between different user actions
- Now players can properly access the secret compartment using the coin
- Activity validated and passes all checks
Remove prescriptive ship destruction descriptions and let the AI be creative.
Since skip_condition ensures these prompts only run when ships are actually
destroyed, we can make the prompts more concise and focused on the outcome.
Add skip_condition logic to feedback prompts to prevent AI from generating
false ship destruction messages when no ships were actually destroyed.
Changes:
- Add skip_condition parameter support in provide_feedback_prompts()
- Support all_null, all_false, and all_true condition types
- Apply skip_condition to battleship Ship Status and Game Over prompts
- Add comprehensive unit tests covering all skip condition scenarios
- Test real battleship scenario that was causing hallucinations
This prevents the AI from creating false positive ship destruction messages
when metadata indicates no ships were actually sunk (all null values).
- Removed feedback_tokens_for_ai from step 3 Game Over
- Exit transition already has appropriate content_blocks
- Eliminates duplicate farewell messages when exiting
Major improvements to battleship game feedback accuracy and user experience:
## New Multi-Prompt Feedback System
- Replaced single feedback with 3 specialized prompts: Shot Report, Ship Status, Game Over
- Each prompt has individual metadata filtering to see only relevant data
- Shot Report only sees hit/miss data, Ship Status only sees ship destruction data
- Added STFU token system to suppress empty messages (filtered out automatically)
## Technical Implementation
- Added per-prompt metadata_filter support in YAML structure
- Updated app.py and guarded_ai.py to handle prompt-specific filtering
- Legacy single-prompt system still works with transition-level filtering
- Added comprehensive test suite for feedback system validation
## User Experience Fixes
- Fixed TTS queue blocking JavaScript execution (async promises instead of await)
- Ship Status now correctly reports who destroyed which ship (role confusion fixed)
- Game Over only appears when game actually ends (no more random messages)
- Maintained dramatic storytelling while ensuring factual accuracy
## Battleship-Specific Improvements
- Ship destruction messages only appear when ships actually sink
- Clear separation of concerns: hits/misses vs ship destruction vs game over
- Eliminated false positive ship destruction reports
- Fixed role reversal where wrong player got credit for destruction
The battleship narrator now provides accurate, contextual feedback while preserving the dramatic naval warfare atmosphere.
- Fix battleship feedback perspective confusion with better Hermes prompting
- Add auto-play TTS button with localStorage persistence and queueing system
- Move activity controls below model/voice selectors in sidebar
- Add activity controls to mobile hamburger menu
- Fix model/activity dropdowns to stay within container bounds
- Filter activities API to only show .yaml/.yml files
- Clean up system message labels by moving to usernames (System (Feedback), System (Question))
- Apply black formatting to app.py
- Updated validator terminal step detection to only flag truly terminal steps
- Fixed validator to accept integers and booleans in buckets (as supported by app.py)
- Fixed metadata_remove format in activity17 from dictionary to list of strings
- Added proper terminal section to activity3.yaml without questions/buckets
- Fixed missing restart transition and bucket in activity28
- Removed unused game_end transitions from battleship files
- Updated exit transitions to go directly to step_4 (goodbye step)
- Applied black formatting to validator code
All 30 activity YAML files now validate successfully with 0 errors and 0 warnings.
- Add matplotlib.use("Agg") backend configuration to prevent runtime errors in headless environments
- Add error handling guards for script results that might return None
- Fix AI targeting logic to exclude already-fired cells in super hunter and hunter modes
- Update CLAUDE.md with matplotlib best practices
- Add pass statements to empty else blocks that only contained commented prints
- Ensures Python syntax remains valid after commenting out debug statements
- Add user_response to pre-script metadata for better game state management
- Implement metadata_feedback_filter to control feedback data exposure
- Improve ship destruction announcements and game over messaging
- Add debug logging for ship sinking events
- Include test ship configuration file
- Add new Hermes Reasoner AI mode that combines probability analysis with LLM reasoning
- Implement pre-script and post-script architecture in app.py for flexible YAML processing
- Fix game ending detection by adding transition override mechanism
- Add probability matrix visualization and strategic move analysis
- Support both legacy processing_script and new pre_script/post_script naming
- Restore full ship complement for complete battleship gameplay
- Replace unsafe eval() with sympy for secure expression parsing
- Add YAML anchors to eliminate code duplication in processing scripts
- Implement multiple function plotting with comma-separated syntax
- Add dynamic plot ranges based on function characteristics
- Include automatic function type detection and analysis
- Streamline activity flow: intro → demo plot → open sandbox
- Add comprehensive error handling with visual error messages
- Support enhanced mathematical notation (arcsin, ln, implied multiplication)
modified: research/activity24-math-plot.yaml
In this mode the AI will switch from random to hunting all the positions
around the latest hit. It's still not as smart as a human but you will
start to feel hunted as the game progresses versus the other game mode.
modified: research/activity29-battleship.yaml
The app seems to queue the category, feedback/content_blocks question
The strange part is the set_background happens in the middle of the
script after category but it comes first and FAST! In about a second
while the other messages take about 4 secs to finally arrive.
modified: app.py
modified: research/activity29-battleship.yaml
modified: templates/chat.html
