Fixed Docker build issues to complete UV migration.
## Changes
### README.md
- Created backend README.md (required by hatchling build system)
- Simple quick start guide referencing CLAUDE.md
### Dockerfile
- Added `build-essential` for compiling native extensions (pendulum Rust code)
- Updated copy steps to include README.md in both dev and prod stages
- Dockerfile now successfully builds both development and production images
### .dockerignore
- Added exception `!README.md` to allow README.md through
- Keeps other *.md files excluded as intended
## Testing
- ✅ Development image builds successfully (paper-dynasty-backend:dev-uv)
- ✅ Production image builds successfully (paper-dynasty-backend:prod-uv)
- ✅ Container starts and UV installs dependencies correctly
- ✅ Application attempts to start (fails only on missing .env, as expected)
## Build Results
- Dev image: 73 packages installed (with dev deps)
- Prod image: 57 packages installed (no dev deps)
- Both stages use `uv sync --frozen` for reproducible builds
- Build time: ~1-2 minutes per stage
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Migrated from pip to UV (v0.9.7) for faster, more reliable package management.
## Changes
### Package Management
- Created `pyproject.toml` with project metadata and dependencies
- Generated `uv.lock` for reproducible builds (198KB)
- Migrated all dependencies from requirements.txt:
- Production: 20 packages (FastAPI, SQLAlchemy, Pydantic, etc.)
- Development: 6 packages (pytest, black, mypy, flake8)
- Virtual environment moved from `venv/` to `.venv/` (UV default)
### Dockerfile Updates
- Updated base image: python:3.11-slim → python:3.13-slim
- Added UV installation via official image copy
- Development stage: Uses `uv sync --frozen` for all deps
- Production stage: Uses `uv sync --frozen --no-dev` for prod only
- Commands now use `uv run` prefix for auto-activation
- Set UV environment variables:
- UV_LINK_MODE=copy (for Docker compatibility)
- UV_COMPILE_BYTECODE=1 (performance)
- UV_PYTHON_DOWNLOADS=never (use system Python)
### Code Fixes
- Fixed Pydantic model circular import in `app/models/game_models.py`
- Added runtime import of PositionRating at end of file
- Called `model_rebuild()` on LineupPlayerState and GameState
- Resolves "not fully defined" errors in tests
### Documentation
- Updated `backend/CLAUDE.md` with UV usage:
- Package management section with UV commands
- Updated all code examples to use `uv run`
- Daily development workflow now uses UV
- Added dependency management guide (add/remove/update)
- Updated virtual environment location (.venv/)
### Project Files
- Added `.python-version` (3.13) for UV Python version tracking
- Removed UV template files (main.py, README.md)
- Kept existing .gitignore (already includes venv/ and .venv/)
## Testing
- ✅ All imports work correctly
- ✅ 55/55 game model tests passing
- ✅ 500+ unit tests passing (same as before migration)
- ✅ Test failures are pre-existing (not UV-related)
## Benefits
- 10-100x faster dependency resolution
- 2-3x faster installation
- Better dependency conflict resolution
- Single tool for everything (replaces pip, pip-tools, virtualenv)
- Reproducible builds with uv.lock
## Migration Path for Team
```bash
# One-time: Install UV
curl -LsSf https://astral.sh/uv/install.sh | sh
# In project
cd backend
uv sync # Creates .venv and installs everything
# Daily usage
uv run python -m app.main
uv run pytest tests/ -v
```
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Updated NEXT_SESSION.md to reflect Phase 3E-Main completion:
- Marked Phase 3E-Main as 100% complete (position ratings integration)
- Overall Phase 3E progress now 90%
- Updated Phase 3 overall progress to ~95%
- Documented all accomplishments and live test results
- Outlined remaining Phase 3E-Final tasks (10%):
* WebSocket event handlers for X-Check UI
* Redis caching upgrade from in-memory
* Full defensive lineup evaluation
* Manual vs Auto mode workflows
Included comprehensive session handoff:
- Live API test results with player 8807 (7 positions)
- Performance metrics (16,601x cache speedup)
- All files created/modified
- Quick start guide for next session
- Technical architecture notes
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Complete integration of position ratings system enabling X-Check defensive plays
to use actual player ratings from PD API with intelligent fallbacks for SBA.
**Live API Testing Verified**: ✅
- Endpoint: GET https://pd.manticorum.com/api/v2/cardpositions?player_id=8807
- Response: 200 OK, 7 positions retrieved successfully
- Cache performance: 16,601x faster (API: 0.214s, Cache: 0.000s)
- Data quality: Real defensive ratings (range 1-5, error 0-88)
**Architecture Overview**:
- League-aware: PD league fetches ratings from API, SBA uses defaults
- StateManager integration: Defenders retrieved from lineup cache
- Self-contained GameState: All data needed for X-Check in memory
- Graceful degradation: Falls back to league averages if ratings unavailable
**Files Created**:
1. app/services/pd_api_client.py (NEW)
- PdApiClient class for PD API integration
- Endpoint: GET /api/v2/cardpositions?player_id={id}&position={pos}
- Async HTTP client using httpx (already in requirements.txt)
- Optional position filtering: get_position_ratings(8807, ['SS', '2B'])
- Returns List[PositionRating] for all positions player can play
- Handles both list and dict response formats
- Comprehensive error handling with logging
2. app/services/position_rating_service.py (NEW)
- PositionRatingService with in-memory caching
- get_ratings_for_card(card_id, league_id) - All positions
- get_rating_for_position(card_id, position, league_id) - Specific position
- Cache performance: >16,000x faster on hits
- Singleton pattern: position_rating_service instance
- TODO Phase 3E-Final: Upgrade to Redis
3. app/services/__init__.py (NEW)
- Package exports for clean imports
4. test_pd_api_live.py (NEW)
- Live API integration test script
- Tests with real PD player 8807 (7 positions)
- Verifies caching, filtering, GameState integration
- Run: `python test_pd_api_live.py`
5. test_pd_api_mock.py (NEW)
- Mock integration test for CI/CD
- Demonstrates flow without API dependency
6. tests/integration/test_position_ratings_api.py (NEW)
- Pytest integration test suite
- Real API tests with player 8807
- Cache verification, SBA skip logic
- Full end-to-end GameState flow
**Files Modified**:
1. app/models/game_models.py
- LineupPlayerState: Added position_rating field (Optional[PositionRating])
- GameState: Added get_defender_for_position(position, state_manager)
- Uses StateManager's lineup cache to find active defender by position
- Iterates through lineup.players to match position + is_active
2. app/config/league_configs.py
- SbaConfig: Added supports_position_ratings() → False
- PdConfig: Added supports_position_ratings() → True
- Enables league-specific behavior without hardcoded conditionals
3. app/core/play_resolver.py
- __init__: Added state_manager parameter for X-Check defender lookup
- _resolve_x_check(): Replaced placeholder defender ratings with actual lookup
- Uses league config to check if ratings supported
- Fetches defender via state.get_defender_for_position()
- Falls back to defaults (range=3, error=15) if ratings unavailable
- Detailed logging for debugging rating lookups
4. app/core/game_engine.py
- Added _load_position_ratings_for_lineup() method
- Loads all position ratings at game start for PD league
- Skips loading for SBA (league config check)
- start_game(): Calls rating loader for both teams before marking active
- PlayResolver instantiation: Now passes state_manager parameter
- Logs: "Loaded X/9 position ratings for team Y"
**X-Check Resolution Flow**:
1. League check: config.supports_position_ratings()?
2. Get defender: state.get_defender_for_position(pos, state_manager)
3. If PD + defender.position_rating exists: Use actual range/error
4. Else if defender found: Use defaults (range=3, error=15)
5. Else: Log warning, use defaults
**Position Rating Loading (Game Start)**:
1. Check if league supports ratings (PD only)
2. Get lineup from StateManager cache
3. For each player:
- Fetch rating from position_rating_service (with caching)
- Set player.position_rating field
4. Cache API responses (16,000x faster on subsequent access)
5. Log success: "Loaded X/9 position ratings for team Y"
**Live Test Results (Player 8807)**:
```
Position Range Error Innings
CF 3 2 372
2B 3 8 212
SS 4 12 159
RF 2 2 74
LF 3 2 62
1B 4 0 46
3B 3 65 34
```
**Testing**:
- ✅ Live API: Player 8807 → 7 positions retrieved successfully
- ✅ Caching: 16,601x performance improvement
- ✅ League config: SBA=False, PD=True
- ✅ GameState integration: Defender lookup working
- ✅ Existing tests: 27/28 config tests passing (1 pre-existing URL failure)
- ✅ Syntax validation: All files compile successfully
**Benefits**:
- ✅ X-Check now uses real defensive ratings in PD league
- ✅ SBA league continues working with manual entry (uses defaults)
- ✅ No breaking changes to existing functionality
- ✅ Graceful degradation if API unavailable
- ✅ In-memory caching reduces API calls by >99%
- ✅ League-agnostic design via config system
- ✅ Production-ready with live API verification
**Phase 3E Status**: Main complete (85% → 90%)
**Next**: Phase 3E-Final (WebSocket events, Redis upgrade, full defensive lineup)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
**Architectural Improvement**: Unified player references in GameState
**Changed**: Make all player references consistent
- BEFORE: current_batter/pitcher/catcher were IDs (int)
- AFTER: current_batter/pitcher/catcher are full LineupPlayerState objects
- Matches pattern of on_first/on_second/on_third (already objects)
**Benefits**:
1. Consistent API - all player references use same type
2. Self-contained GameState - everything needed for resolution
3. No lookups needed - direct access to player data
4. Sets foundation for Phase 3E-Main (adding position ratings)
**Files Modified**:
- app/models/game_models.py: Changed current_batter/pitcher/catcher to objects
- app/core/game_engine.py: Updated _prepare_next_play() to populate full objects
- app/core/state_manager.py: Create placeholder batter on game creation
- tests/unit/models/test_game_models.py: Updated all 27 GameState tests
**Database Operations**:
- No schema changes needed
- Play table still stores IDs (for referential integrity)
- IDs extracted from objects when saving: state.current_batter.lineup_id
**Testing**:
- All 27 GameState tests passing
- No regressions in existing functionality
- Type checking passes
**Next**: Phase 3E-Main - Add PositionRating dataclass and load ratings at game start
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Updated X-Check runner advancement functions to properly delegate to
existing result handlers for non-error cases.
Changes:
- Updated x_check_g1/g2/g3 signatures to accept GameState, hit_location,
and defensive_decision parameters
- Updated x_check_f1/f2/f3 signatures to accept GameState and hit_location
- Implemented delegation logic: error cases use simple tables, non-error
cases delegate to existing tested result handlers (_execute_result,
_fb_result_*)
- Updated PlayResolver._get_x_check_advancement() to pass new parameters
- Updated all tests to provide required GameState fixtures
Benefits:
- Reuses 13 existing groundball + 4 flyball result handlers (DRY)
- No DP probability needed - X-Check d20 already tested defender
- Full game context: real lineup IDs, outs count, conditional logic
- Error cases remain simple and efficient
Test Results: 264/265 core tests passing (1 pre-existing dice failure)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Fixed incorrect double play logic that was rolling for probability
twice - once for the chart result and again for execution.
Changes:
- Removed _calculate_double_play_probability() method entirely
- Updated _gb_result_2() to execute DP deterministically
- Updated _gb_result_10() to execute DP deterministically
- Updated _gb_result_13() to execute DP deterministically
- Removed TestDoublePlayProbability test class (5 tests)
- Updated DP tests to reflect deterministic behavior
Logic: Chart already determines outcome via dice roll. When chart
says "Result 2: Double Play", the DP happens (if <2 outs and runner
on 1st exists). No additional probability roll needed.
Tests: 55/55 runner advancement tests passing
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Fixed two critical bugs in Phase 3D X-Check implementation plus
improved dice audit trail for better tracking.
BUG #1: on_base_code Mapping Error (Sequential vs Bit Field)
============================================================
The implementation incorrectly treated on_base_code as a bit field
when it is actually a sequential lookup mapping.
WRONG (bit field):
Code 3 (0b011) → R1 + R2
Code 4 (0b100) → R3 only
CORRECT (sequential):
Code 3 → R3 only
Code 4 → R1 + R2
Fixed:
- build_advancement_from_code() decoder (sequential mapping)
- build_flyball_advancement_with_error() decoder (sequential mapping)
- 13 test on_base_code values (3↔4 corrections)
- Updated documentation to clarify NOT a bit field
BUG #2: Table Data Not Matching Official Charts
================================================
7 table entries in G1_ADVANCEMENT_TABLE and G2_ADVANCEMENT_TABLE
did not match the official rulebook charts provided by user.
Fixed table entries:
- G1 Code 1, Infield In: Changed Result 3 → 2
- G1 Code 3, Normal: Changed Result 13 → 3
- G1 Code 3, Infield In: Changed Result 3 → 1
- G1 Code 4, Normal: Changed Result 3 → 13
- G1 Code 4, Infield In: Changed Result 4 → 2
- G2 Code 3, Infield In: Changed Result 3 → 1
- G2 Code 4, Normal: Changed Result 5 → 4
Also fixed 7 test expectations to match corrected tables.
IMPROVEMENT: Better Dice Audit Trail
=====================================
Updated _resolve_x_check() in PlayResolver to use proper
dice_system.roll_fielding() instead of manual die rolling.
Benefits:
- All dice tracked in audit trail (roll_id, timestamp, position)
- Automatic error_total calculation (no manual 3d6 addition)
- Consistent with codebase patterns
- Position recorded for historical analysis
Testing:
- All 59 X-Check advancement tests passing (100%)
- All 9 PlayResolver tests passing (100%)
- All table entries validated against official charts
- Complete codebase scan: no bit field operations found
Files modified:
- backend/app/core/x_check_advancement_tables.py
- backend/tests/unit/core/test_x_check_advancement_tables.py
- backend/app/core/play_resolver.py
- .claude/implementation/PHASE_3D_CRITICAL_FIX.md (documentation)
- .claude/implementation/GROUNDBALL_CHART_REFERENCE.md (new)
- .claude/implementation/XCHECK_TEST_VALIDATION.md (new)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Added comprehensive Phase 3C documentation to CLAUDE.md:
- X-Check resolution logic implementation details
- Helper methods and integration points
- Key features and algorithms
- Placeholders for future phases
- Test coverage and verification results
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Implemented complete X-Check resolution system in PlayResolver with
defense range and error table lookups.
Changes:
- Added X_CHECK case to resolve_outcome() method
- Implemented _resolve_x_check() main resolution method
- Added _adjust_range_for_defensive_position() for playing in logic
- Added _lookup_defense_table() for defense range table lookups
- Added _apply_hash_conversion() for G2#/G3# to SI2 conversion
- Added _lookup_error_chart() for error determination
- Added _determine_final_x_check_outcome() for final outcome mapping
- Added XCheckResult to PlayResult dataclass
- Integrated all Phase 3B tables (defense, error, holding runners)
Features:
- Full defense table lookup (infield, outfield, catcher)
- Error chart lookup with priority ordering (RP > E3 > E2 > E1 > NO)
- Range adjustment for playing in (+1, max 5)
- Hash conversion based on playing in OR holding runner
- Error overrides outs to ERROR outcome
- Rare play handling
- Detailed X-Check audit trail in XCheckResult
Placeholders (to be completed in later phases):
- Defender retrieval from lineup (currently uses placeholder ratings)
- SPD test implementation (currently defaults to G3)
- Batter handedness from player model
- Runner advancement tables (Phase 3D)
Testing:
- All 9 PlayResolver tests passing
- All 36 X-Check table tests passing
- All 51 runner advancement tests passing
- 325/327 total tests passing (2 pre-existing failures unrelated)
- play_resolver.py compiles successfully
Phase 3C Status: 100% COMPLETE ✅
Ready for Phase 3D (Runner Advancement Tables)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Add foundational data structures for X-Check play resolution system:
Models Added:
- PositionRating: Defensive ratings (range 1-5, error 0-88) for X-Check resolution
- XCheckResult: Dataclass tracking complete X-Check resolution flow with dice rolls,
conversions (SPD test, G2#/G3#→SI2), error results, and final outcomes
- BasePlayer.active_position_rating: Optional field for current defensive position
Enums Extended:
- PlayOutcome.X_CHECK: New outcome type requiring special resolution
- PlayOutcome.is_x_check(): Helper method for type checking
Documentation Enhanced:
- Play.check_pos: Documented as X-Check position identifier
- Play.hit_type: Documented with examples (single_2_plus_error_1, etc.)
Utilities Added:
- app/core/cache.py: Redis cache key helpers for player positions and game state
Implementation Planning:
- Complete 6-phase implementation plan (3A-3F) documented in .claude/implementation/
- Phase 3A complete with all acceptance criteria met
- Zero breaking changes, all existing tests passing
Next: Phase 3B will add defense tables, error charts, and advancement logic
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Major Phase 2 refactoring to consolidate runner advancement logic:
**Flyball System Enhancement**:
- Add FLYOUT_BQ variant (medium-shallow depth)
- 4 flyball types with clear semantics: A (deep), B (medium), BQ (medium-shallow), C (shallow)
- Updated helper methods to include FLYOUT_BQ
**RunnerAdvancement Integration**:
- Extend runner_advancement.py to handle both groundballs AND flyballs
- advance_runners() routes to _advance_runners_groundball() or _advance_runners_flyball()
- Comprehensive flyball logic with proper DECIDE mechanics per flyball type
- No-op movements recorded for state recovery consistency
**PlayResolver Refactoring**:
- Consolidate all 4 flyball outcomes to delegate to RunnerAdvancement (DRY)
- Eliminate duplicate flyball resolution code
- Rename helpers for clarity: _advance_on_single_1/_advance_on_single_2 (was _advance_on_single)
- Fix single/double advancement logic for different hit types
**State Recovery Fix**:
- Fix state_manager.py game recovery to build LineupPlayerState objects properly
- Use get_lineup_player() helper to construct from lineup data
- Correctly track runners in on_first/on_second/on_third fields (matches Phase 2 model)
**Database Support**:
- Add runner tracking fields to play data for accurate recovery
- Include batter_id, on_first_id, on_second_id, on_third_id, and *_final fields
**Type Safety Improvements**:
- Fix lineup_id access throughout runner_advancement.py (was accessing on_first directly, now on_first.lineup_id)
- Make current_batter_lineup_id non-optional (always set by _prepare_next_play)
- Add type: ignore for known SQLAlchemy false positives
**Documentation**:
- Update CLAUDE.md with comprehensive flyball documentation
- Add flyball types table, usage examples, and test coverage notes
- Document differences between groundball and flyball mechanics
**Testing**:
- Add test_flyball_advancement.py with 21 flyball tests
- Coverage: all 4 types, DECIDE scenarios, no-op movements, edge cases
🚀 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Update test fixtures to provide required current_batter_lineup_id field
- Update validator tests for new infield depth values (infield_in, normal, corners_in)
- Update runner advancement tests to match refactored runner management
- Update game model tests to work with direct base references
- Update play resolver tests for enhanced logic
- Add missing imports in test files
All changes ensure tests align with recent Phase 2 implementation updates
including the transition from RunnerState list to direct base references
(on_first, on_second, on_third) in GameState model.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Major Refactor: Outcome-First Architecture
- PlayResolver now accepts league_id and auto_mode in constructor
- Added core resolve_outcome() method - all resolution logic in one place
- Added resolve_manual_play() wrapper for manual submissions (primary)
- Added resolve_auto_play() wrapper for PD auto mode (rare)
- Removed SimplifiedResultChart (obsolete with new architecture)
- Removed play_resolver singleton
RunnerAdvancement Integration:
- All groundball outcomes (GROUNDBALL_A/B/C) now use RunnerAdvancement
- Proper DP probability calculation with positioning modifiers
- Hit location tracked for all relevant outcomes
- 13 result types fully integrated from advancement charts
Game State Updates:
- Added auto_mode field to GameState (stored per-game)
- Updated state_manager.create_game() to accept auto_mode parameter
- GameEngine now uses state.auto_mode to create appropriate resolver
League Configuration:
- Added supports_auto_mode() to BaseGameConfig
- SbaConfig: returns False (no digitized cards)
- PdConfig: returns True (has digitized ratings)
- PlayResolver validates auto mode support and raises error for SBA
Play Results:
- Added hit_location field to PlayResult
- Groundballs include location from RunnerAdvancement
- Flyouts track hit_location for tag-up logic (future)
- Other outcomes have hit_location=None
Testing:
- Completely rewrote test_play_resolver.py for new architecture
- 9 new tests covering initialization, strikeouts, walks, groundballs, home runs
- All 9 tests passing
- All 180 core tests still passing (1 pre-existing failure unrelated)
Terminal Client:
- No changes needed - defaults to manual mode (auto_mode=False)
- Perfect for human testing of manual submissions
This completes Week 7 Task 6 - the final task of Week 7!
Week 7 is now 100% complete with all 8 tasks done.
🎯 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Created runner_advancement.py with complete groundball advancement system
- Implemented GroundballResultType IntEnum with 13 rulebook-aligned results
- Built RunnerAdvancement class with chart lookup logic (Infield Back/In)
- Implemented all 13 result handlers (gb_result_1 through gb_result_13)
- Added DECIDE mechanic support for interactive runner advancement decisions
- Implemented double play probability calculation with positioning modifiers
- Created 30 comprehensive unit tests covering all scenarios (100% passing)
Key Features:
- Supports Infield Back and Infield In defensive positioning
- Handles Corners In hybrid positioning (applies In rules to corner fielders)
- Conditional results based on hit location (middle IF, right side, corners)
- Force play detection and advancement logic
- Double play mechanics with probability-based success (45% base rate)
- Result types match official rulebook exactly (1-13)
Architecture:
- IntEnum for result types (type-safe, self-documenting)
- Comprehensive hit location tracking (1B, 2B, SS, 3B, P, C, LF, CF, RF)
- Dataclasses for movements (RunnerMovement, AdvancementResult)
- Probability modifiers: Infield In (-15%), hit location (±10%)
Testing:
- 30 unit tests covering chart lookup, all result types, and edge cases
- Double play probability validation
- All on-base codes (0-7) tested
- All groundball types (A, B, C) verified
Status: Week 7 Tasks 4 & 5 complete (~87% of Week 7 finished)
Next: Task 6 (PlayResolver Integration)
Related: #task4 #task5 #runner-advancement #double-play #week7
Complete manual outcome workflow for SBA and PD manual mode gameplay:
**WebSocket Event Handlers** (app/websocket/handlers.py):
- roll_dice: Server rolls dice, stores in state, broadcasts to players
- submit_manual_outcome: Validates and processes player submissions
- Events: dice_rolled, outcome_accepted, outcome_rejected, play_resolved
**Game Engine Integration** (app/core/game_engine.py):
- resolve_manual_play(): Processes manual outcomes with server dice
- Uses ab_roll for audit trail, player outcome for resolution
- Same orchestration as resolve_play() (save, update, advance inning)
**Data Model** (app/models/game_models.py):
- pending_manual_roll: Stores server dice between roll and submission
**Terminal Client** (terminal_client/):
- roll_dice command: Roll dice and display results
- manual_outcome command: Submit outcomes from physical cards
- Both integrated into REPL for testing
**Tests** (tests/unit/websocket/test_manual_outcome_handlers.py):
- 12 comprehensive tests covering all validation paths
- All tests passing (roll_dice: 4, submit_manual_outcome: 8)
**Key Decisions**:
- Server rolls dice for fairness (not players!)
- One-time roll usage (cleared after submission)
- Early validation (check pending roll before accepting)
- Field-level error messages for clear feedback
**Impact**:
- Complete manual mode workflow ready
- Frontend WebSocket integration supported
- Terminal testing commands available
- Audit trail with server-rolled dice maintained
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Add ability to roll back the last N plays, useful for correcting mistakes
or recovering from corrupted plays. Deletes plays from database and
reconstructs game state by replaying remaining plays.
Database Operations (app/database/operations.py):
- delete_plays_after(): Delete plays with play_number > target
- delete_substitutions_after(): Delete lineup entries with after_play >= target
- delete_rolls_after(): Delete dice rolls (kept for reference, not used)
Game Engine (app/core/game_engine.py):
- rollback_plays(): Main rollback orchestration
- Validates: num_plays > 0, enough plays exist, game not completed
- Deletes plays and substitutions from database
- Clears in-memory roll tracking
- Calls state_manager.recover_game() to rebuild state
- Returns updated GameState
Terminal Client (terminal_client/commands.py, terminal_client/repl.py):
- rollback_plays(): Command wrapper with user-friendly output
- do_rollback(): REPL command with argument parsing
Usage:
⚾ > rollback 3
Validations:
- Cannot roll back more plays than exist
- Cannot roll back completed games
- Rolling back across innings is allowed
- Substitutions after rolled-back plays are undone
Testing:
- ✅ Successfully rolls back 2 plays from 5-play game
- ✅ Correctly validates rollback of 10 plays when only 2 exist
- ✅ Game state properly reconstructed via replay
Note: Dice rolls kept in database for auditing (don't affect state).
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Add ability to force specific play outcomes instead of random dice rolls,
enabling targeted testing of specific game scenarios.
Changes:
- play_resolver.resolve_play(): Add forced_outcome parameter, bypass dice
rolls when provided, create dummy AbRoll with placeholder values
- game_engine.resolve_play(): Accept and pass through forced_outcome param
- terminal_client/commands.py: Pass forced_outcome to game engine
Testing:
- Verified TRIPLE, HOMERUN, and STRIKEOUT outcomes work correctly
- Dummy AbRoll properly constructed with all required fields
- Game state updates correctly with forced outcomes
Example usage in REPL:
resolve_with triple
resolve_with homerun
Fixes terminal client testing workflow to allow controlled scenarios.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Remove unused PlayResult creation code that had incorrect import path
and missing required fields. The forced outcome feature is experimental
and not yet implemented - the code was just showing warnings anyway.
Fixes ImportError when running 'resolve_with <outcome>' command.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
New Commands:
- manual_outcome <outcome> [location] - Validates ManualOutcomeSubmission
- Tests outcome and location validation
- Shows which outcomes require hit location
- Displays clear validation errors
- test_location <outcome> [handedness] [count] - Tests hit location distribution
- Generates sample hit locations for an outcome
- Shows distribution table with percentages
- Validates pull rates (45% pull, 35% center, 20% opposite)
- Supports both LHB and RHB
Implementation:
- Added validate_manual_outcome() to GameCommands class
- Added test_hit_location() to GameCommands class
- Added do_manual_outcome() to REPL
- Added do_test_location() to REPL
- Uses ManualOutcomeSubmission model from Task 3
- Uses calculate_hit_location() helper from Task 3
Testing:
- Tested manual_outcome with valid outcomes (groundball_c SS, strikeout)
- Tested manual_outcome with invalid outcome (proper error display)
- Tested test_location with groundball_c for RHB (shows distribution)
- All validation and display working correctly
Note: Full play resolution integration deferred to Week 7 Task 6 (WebSocket handlers).
These commands validate and test the new models but don't resolve plays yet.
Files Modified:
- terminal_client/commands.py (+117 lines)
- terminal_client/repl.py (+65 lines)
- Enhanced validate_defensive_decision() with comprehensive validation:
- Validate all alignments (normal, shifted_left, shifted_right, extreme_shift)
- Validate all infield depths (in, normal, back, double_play)
- Validate all outfield depths (in, normal, back)
- Validate hold_runners require actual runners on specified bases
- Validate hold_runners only on bases 1, 2, or 3
- Validate double_play depth requires runner on first
- Validate double_play depth not allowed with 2 outs
- Enhanced validate_offensive_decision() with comprehensive validation:
- Validate all approaches (normal, contact, power, patient)
- Validate steal_attempts only to bases 2, 3, or 4
- Validate steal_attempts require runner on base-1
- Validate bunt_attempt not allowed with 2 outs
- Validate bunt_attempt and hit_and_run cannot be simultaneous
- Validate hit_and_run requires at least one runner on base
- Added 24+ comprehensive test cases covering all edge cases:
- 13 new defensive decision validation tests
- 16 new offensive decision validation tests
- All tests pass (54/54 passing)
Clear error messages for all validation failures.
Follows 'Raise or Return' pattern with ValidationError exceptions.
Terminal Client Enhancements:
- Added list_outcomes command to display all PlayOutcome values
- Added resolve_with <outcome> command for testing specific scenarios
- TAB completion for all outcome names
- Full help documentation and examples
- Infrastructure ready for Week 7 integration
Files Modified:
- terminal_client/commands.py - list_outcomes() and forced outcome support
- terminal_client/repl.py - do_list_outcomes() and do_resolve_with() commands
- terminal_client/completions.py - VALID_OUTCOMES and complete_resolve_with()
- terminal_client/help_text.py - Help entries for new commands
Phase 3 Planning:
- Created comprehensive Week 7 implementation plan (25 pages)
- 6 major tasks covering strategic decisions and result charts
- Updated 00-index.md to mark Week 6 as 100% complete
- Documented manual outcome testing feature
Week 6: 100% Complete ✅
Phase 3 Week 7: Ready to begin
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Renamed check_d20 → chaos_d20 throughout dice system
- Expanded PlayOutcome enum with granular variants (SINGLE_1/2, DOUBLE_2/3, GROUNDBALL_A/B/C, etc.)
- Integrated PlayOutcome from app.config into PlayResolver
- Added play_metadata support for uncapped hit tracking
- Updated all tests (139/140 passing)
Week 6: 100% Complete - Ready for Phase 3
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
This commit includes cleanup from model refactoring and terminal client
modularization for better code organization and maintainability.
## Game Models Refactor
**Removed RunnerState class:**
- Eliminated separate RunnerState model (was redundant)
- Replaced runners: List[RunnerState] with direct base references:
- on_first: Optional[LineupPlayerState]
- on_second: Optional[LineupPlayerState]
- on_third: Optional[LineupPlayerState]
- Updated helper methods:
- get_runner_at_base() now returns LineupPlayerState directly
- get_all_runners() returns List[Tuple[int, LineupPlayerState]]
- is_runner_on_X() simplified to direct None checks
**Benefits:**
- Matches database structure (plays table has on_first_id, etc.)
- Simpler state management (direct references vs list management)
- Better type safety (LineupPlayerState vs generic runner)
- Easier to work with in game engine logic
**Updated files:**
- app/models/game_models.py - Removed RunnerState, updated GameState
- app/core/play_resolver.py - Use get_all_runners() instead of state.runners
- app/core/validators.py - Updated runner access patterns
- tests/unit/models/test_game_models.py - Updated test assertions
- tests/unit/core/test_play_resolver.py - Updated test data
- tests/unit/core/test_validators.py - Updated test data
## Terminal Client Refactor
**Modularization (DRY principle):**
Created separate modules for better code organization:
1. **terminal_client/commands.py** (10,243 bytes)
- Shared command functions for game operations
- Used by both CLI (main.py) and REPL (repl.py)
- Functions: submit_defensive_decision, submit_offensive_decision,
resolve_play, quick_play_sequence
- Single source of truth for command logic
2. **terminal_client/arg_parser.py** (7,280 bytes)
- Centralized argument parsing and validation
- Handles defensive/offensive decision arguments
- Validates formats (alignment, depths, hold runners, steal attempts)
3. **terminal_client/completions.py** (10,357 bytes)
- TAB completion support for REPL mode
- Command completions, option completions, dynamic completions
- Game ID completions, defensive/offensive option suggestions
4. **terminal_client/help_text.py** (10,839 bytes)
- Centralized help text and command documentation
- Detailed command descriptions
- Usage examples for all commands
**Updated main modules:**
- terminal_client/main.py - Simplified by using shared commands module
- terminal_client/repl.py - Cleaner with shared functions and completions
**Benefits:**
- DRY: Behavior consistent between CLI and REPL modes
- Maintainability: Changes in one place affect both interfaces
- Testability: Can test commands module independently
- Organization: Clear separation of concerns
## Documentation
**New files:**
- app/models/visual_model_relationships.md
- Visual documentation of model relationships
- Helps understand data flow between models
- terminal_client/update_docs/ (6 phase documentation files)
- Phased documentation for terminal client evolution
- Historical context for implementation decisions
## Tests
**New test files:**
- tests/unit/terminal_client/__init__.py
- tests/unit/terminal_client/test_arg_parser.py
- tests/unit/terminal_client/test_commands.py
- tests/unit/terminal_client/test_completions.py
- tests/unit/terminal_client/test_help_text.py
**Updated tests:**
- Integration tests updated for new runner model
- Unit tests updated for model changes
- All tests passing with new structure
## Summary
- ✅ Simplified game state model (removed RunnerState)
- ✅ Better alignment with database structure
- ✅ Modularized terminal client (DRY principle)
- ✅ Shared command logic between CLI and REPL
- ✅ Comprehensive test coverage
- ✅ Improved documentation
Total changes: 26 files modified/created
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
This commit includes Week 6 player models implementation and critical
performance optimizations discovered during testing.
## Player Models (Week 6 - 50% Complete)
**New Files:**
- app/models/player_models.py (516 lines)
- BasePlayer abstract class with polymorphic interface
- SbaPlayer with API parsing factory method
- PdPlayer with batting/pitching scouting data support
- Supporting models: PdCardset, PdRarity, PdBattingCard, PdPitchingCard
- tests/unit/models/test_player_models.py (692 lines)
- 32 comprehensive unit tests, all passing
- Tests for BasePlayer, SbaPlayer, PdPlayer, polymorphism
**Architecture:**
- Simplified single-layer approach vs planned two-layer
- Factory methods handle API → Game transformation directly
- SbaPlayer.from_api_response(data) - parses SBA API inline
- PdPlayer.from_api_response(player_data, batting_data, pitching_data)
- Full Pydantic validation, type safety, and polymorphism
## Performance Optimizations
**Database Query Reduction (60% fewer queries per play):**
- Before: 5 queries per play (INSERT play, SELECT play with JOINs,
SELECT games, 2x SELECT lineups)
- After: 2 queries per play (INSERT play, UPDATE games conditionally)
Changes:
1. Lineup caching (game_engine.py:384-425)
- Check state_manager.get_lineup() cache before DB fetch
- Eliminates 2 SELECT queries per play
2. Remove unnecessary refresh (operations.py:281-302)
- Removed session.refresh(play) after INSERT
- Eliminates 1 SELECT with 3 expensive LEFT JOINs
3. Direct UPDATE statement (operations.py:109-165)
- Changed update_game_state() to use direct UPDATE
- No longer does SELECT + modify + commit
4. Conditional game state updates (game_engine.py:200-217)
- Only UPDATE games table when score/inning/status changes
- Captures state before/after and compares
- ~40-60% fewer updates (many plays don't score)
## Bug Fixes
1. Fixed outs_before tracking (game_engine.py:551)
- Was incorrectly calculating: state.outs - result.outs_recorded
- Now correctly captures: state.outs (before applying result)
- All play records now have accurate out counts
2. Fixed game recovery (state_manager.py:312-314)
- AttributeError when recovering: 'GameState' has no attribute 'runners'
- Changed to use state.get_all_runners() method
- Games can now be properly recovered from database
## Enhanced Terminal Client
**Status Display Improvements (terminal_client/display.py:75-97):**
- Added "⚠️ WAITING FOR ACTION" section when play is pending
- Shows specific guidance:
- "The defense needs to submit their decision" → Run defensive [OPTIONS]
- "The offense needs to submit their decision" → Run offensive [OPTIONS]
- "Ready to resolve play" → Run resolve
- Color-coded command hints for better UX
## Documentation Updates
**backend/CLAUDE.md:**
- Added comprehensive Player Models section (204 lines)
- Updated Current Phase status to Week 6 (~50% complete)
- Documented all optimizations and bug fixes
- Added integration examples and usage patterns
**New Files:**
- .claude/implementation/week6-status-assessment.md
- Comprehensive Week 6 progress review
- Architecture decision rationale (single-layer vs two-layer)
- Completion status and next priorities
- Updated roadmap for remaining Week 6 work
## Test Results
- Player models: 32/32 tests passing
- All existing tests continue to pass
- Performance improvements verified with terminal client
## Next Steps (Week 6 Remaining)
1. Configuration system (BaseConfig, SbaConfig, PdConfig)
2. Result charts & PD play resolution with ratings
3. API client for live roster data (deferred)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Fixed critical bugs in game recovery and play persistence:
1. Terminal REPL Auto-Recovery:
- Added _ensure_game_loaded() helper to auto-recover games from database
- Calls state_manager.recover_game() when game not in memory
- Calls _prepare_next_play() after recovery to populate snapshot fields
- Enables seamless continuation of games across REPL sessions
2. Play Validation:
- Added verification in _save_play_to_db() for required fields
- Ensures batter_id, pitcher_id, catcher_id are never NULL
- Raises ValueError with clear error message if fields missing
- Prevents database constraint violations
3. Updated Commands:
- All REPL commands now call _ensure_game_loaded()
- Commands: defensive, offensive, resolve, status, quick_play, box_score
- Fixes "Game state not found" errors on recovered games
Root Cause:
- state_manager.recover_game() rebuilds GameState from database
- But didn't populate snapshot fields (current_batter_lineup_id, etc.)
- _save_play_to_db() requires these fields to save plays
- Solution: Call _prepare_next_play() after recovery
Files Modified:
- app/core/game_engine.py - Added verification in _save_play_to_db()
- terminal_client/repl.py - Added _ensure_game_loaded() and integrated
Testing: Successfully recovered game, submitted decisions, and resolved plays
🚀 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Split player model architecture into dedicated documentation files for clarity
and maintainability. Added Phase 1 status tracking and comprehensive player
model specs covering API models, game models, mappers, and testing strategy.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Add comprehensive unit and integration tests for Week 5 deliverables:
- test_play_resolver.py: 18 tests covering outcome resolution and runner advancement
- test_validators.py: 36 tests covering game state, decisions, lineups, and flow
- test_game_engine.py: 7 test classes for complete game flow integration
Update implementation documentation to reflect completed status:
- 00-index.md: Mark Phase 2 Weeks 4-5 complete with test coverage
- 02-week5-game-logic.md: Comprehensive test details and completion status
- 02-game-engine.md: Forward-looking snapshot pattern documentation
Week 5 now fully complete with 54 unit tests + 7 integration test classes passing.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Replaced awkward "lookback" pattern with clean "prepare → execute → save"
orchestration that captures state snapshots BEFORE each play.
Key improvements:
- Added per-team batter indices (away_team_batter_idx, home_team_batter_idx)
- Added play snapshot fields (current_batter/pitcher/catcher_lineup_id)
- Added on_base_code bit field for efficient base situation queries
- Created _prepare_next_play() method for snapshot preparation
- Refactored start_game() with hard lineup validation requirement
- Refactored resolve_play() with explicit 6-step orchestration
- Updated _save_play_to_db() to use snapshots (no DB lookbacks)
- Enhanced state recovery to rebuild from last play (single query)
- Added defensive lineup position validator
Benefits:
- No special cases for first play
- Single source of truth in GameState
- Saves 18+ database queries per game
- Fast state recovery without replay
- Complete runner tracking (before/after positions)
- Explicit orchestration (easy to debug)
Testing:
- Added 3 new test functions (lineup validation, snapshot tracking, batting order)
- All 5 test suites passing (100%)
- Type checking cleaned up with targeted suppressions for SQLAlchemy
Documentation:
- Added comprehensive "Type Checking & Common False Positives" section to CLAUDE.md
- Created type-checking-guide.md and type-checking-summary.md
- Added mypy.ini configuration for SQLAlchemy/Pydantic
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Fixes:
✅ Updated GameEngine._save_play_to_db() to fetch real lineup IDs
- Gets active batting/fielding lineups from database
- Extracts batter, pitcher, catcher IDs by position
- No more hardcoded placeholder IDs
✅ Shortened AbRoll.__str__() to fit VARCHAR(50)
- "WP 1/10" instead of "AB Roll: Wild Pitch Check..."
- "AB 6,9(4+5) d20=12/10" for normal rolls
- Prevents database truncation errors
✅ Created comprehensive test script (scripts/test_game_flow.py)
- Tests single at-bat flow
- Tests full half-inning (50+ plays)
- Creates dummy lineups for both teams
- Verifies complete game lifecycle
Test Results:
✅ Successfully ran 50 at-bats across 6 innings
✅ Score tracking: Away 5 - Home 2
✅ Inning advancement working
✅ Play persistence to database
✅ Roll batch saving at inning boundaries
✅ State synchronization (memory + DB)
GameEngine Verified Working:
✅ Game lifecycle management (create → start → play → complete)
✅ Decision submission (defensive + offensive)
✅ Play resolution with AbRoll system
✅ State management and persistence
✅ Inning advancement logic
✅ Score tracking
✅ Lineup integration
✅ Database persistence
Ready for:
- WebSocket integration
- Frontend connectivity
- Full game simulations
- AI opponent integration
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Core Components:
✅ GameValidator (validators.py)
- Validates game state and decisions
- Rule enforcement for baseball gameplay
- Game-over and inning continuation logic
✅ PlayResolver (play_resolver.py)
- Resolves play outcomes using AbRoll system
- Simplified result charts for MVP
- Handles wild pitch/passed ball checks
- Runner advancement logic for all hit types
- PlayOutcome enum with 12 outcome types
✅ GameEngine (game_engine.py)
- Orchestrates complete game flow
- Start game, submit decisions, resolve plays
- Integrates DiceSystem with roll context
- Batch saves rolls at end of each half-inning
- Persists plays and game state to database
- Manages inning advancement and game completion
Integration Features:
- Uses advanced AbRoll system (not simplified d20)
- Roll context tracking per inning
- Batch persistence at inning boundaries
- Full audit trail with roll history
- State synchronization between memory and database
Architecture:
GameEngine → PlayResolver → DiceSystem
↓ ↓
GameValidator StateManager
↓ ↓
Database In-Memory Cache
Ready For:
✅ End-to-end at-bat testing
✅ WebSocket integration
✅ Result chart configuration
✅ Advanced decision logic (Phase 3)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Attempted Fix:
- Created test-specific engine with NullPool
- Monkeypatched DatabaseOperations to use test engine
- Reference: https://github.com/MagicStack/asyncpg/issues/863
Result:
❌ NullPool did NOT resolve the issue
- Tests still fail after #4 with "another operation is in progress"
- Error occurs during fixture setup, not in test bodies
- Timestamps show pytest setting up multiple fixtures concurrently
Root Cause Analysis:
The issue isn't connection pooling - it's async fixture dependency chains.
When pytest-asyncio sets up `sample_game` fixture (which uses `db_ops`),
it creates overlapping async contexts that asyncpg can't handle.
Evidence:
- Individual tests: ✅ PASS
- First 4 tests together: ✅ PASS
- Tests 5-16: ❌ FAIL with concurrent operation errors
- Unit tests: ✅ 87/88 PASS (core logic proven correct)
Conclusion:
This is a complex pytest-asyncio + SQLAlchemy + asyncpg interaction
requiring architectural test changes (separate test DB, sync fixtures, etc).
Not worth solving pre-MVP given tests work individually and code is proven.
Workaround:
Run test classes separately - each class passes fine:
pytest tests/integration/database/test_roll_persistence.py::TestRollPersistenceBatch -v
pytest tests/integration/database/test_roll_persistence.py::TestRollRetrieval -v
pytest tests/integration/database/test_roll_persistence.py::TestRollDataIntegrity -v
pytest tests/integration/database/test_roll_persistence.py::TestRollEdgeCases -v
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Changes:
- Created tests/integration/conftest.py with shared fixtures
- Added README.md documenting asyncpg connection pool issue
- Fixed uuid4 import in test_roll_persistence.py
Issue Analysis:
- Integration tests work individually but fail when run together (12+ tests)
- AsyncPG error: "cannot perform operation: another operation is in progress"
- Root cause: pytest-asyncio + asyncpg connection reuse across rapid fixtures
- Tests #1-4 pass, then connection pool enters bad state
Test Status:
✅ 87/88 unit tests pass (1 pre-existing timing issue)
✅ Integration tests PASS individually
⚠️ Integration tests FAIL when run together (fixture issue, not code bug)
Workarounds:
- Run test classes separately
- Run individual tests
- Use pytest-xdist for isolation
The tests themselves are well-designed and use production code paths.
This is purely a test infrastructure limitation to be resolved post-MVP.
Core dice and roll persistence logic is proven correct by unit tests.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Core Implementation:
- Created roll_types.py with AbRoll, JumpRoll, FieldingRoll, D20Roll dataclasses
- Implemented DiceSystem singleton with cryptographically secure random generation
- Added Roll model to db_models.py with JSONB storage for roll history
- Implemented save_rolls_batch() and get_rolls_for_game() in database operations
Testing:
- 27 unit tests for roll type dataclasses (100% passing)
- 35 unit tests for dice system (34/35 passing, 1 timing issue)
- 16 integration tests for database persistence (uses production DiceSystem)
Features:
- Unique roll IDs using secrets.token_hex()
- League-specific logic (SBA d100 rare plays, PD error-based rare plays)
- Automatic derived value calculation (d6_two_total, jump_total, error_total)
- Full audit trail with context metadata
- Support for batch saving rolls per inning
Technical Details:
- Fixed dataclass inheritance with kw_only=True for Python 3.13
- Roll data stored as JSONB for flexible querying
- Indexed on game_id, roll_type, league_id, team_id for efficient retrieval
- Supports filtering by roll type, team, and timestamp ordering
Note: Integration tests have async connection pool issue when run together
(tests work individually, fixture cleanup needed in follow-up branch)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
