Key changes based on feedback:
- Reposition as "optimizer & NX configurator" not "LLM-first"
- Add Part 2: Study Characterization & Performance Learning
- Add Part 3: Protocol Evolution workflow (Research → Review → Approve)
- Add Part 4: MCP-first development approach with documentation hierarchy
- Emphasize simulation optimization over CAD/mesh concerns
- Add LAC knowledge accumulation for parameter-performance relationships
- Add privilege levels for protocol approval (user/power_user/admin)
- Update sound bites and core messaging
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Added GitHub remote URL (Anto01/Atomizer) to CLAUDE.md so it persists
across sessions. Also recorded in LAC user_preference.jsonl.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Expanded Key Directories section with full optimization_engine structure
- Added Import Migration section with new import paths
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Update feature_registry.json paths to new module locations (v0.3.0)
- Update cheatsheet with new import paths (v2.3)
- Mark migration plan as completed (v3.0)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add TrialManager (trial_manager.py) for consistent trial_NNNN naming
- Add DashboardDB (dashboard_db.py) for Optuna-compatible database schema
- Update CLAUDE.md with trial management documentation
- Update ATOMIZER_CONTEXT.md with v1.8 trial system
- Update cheatsheet v2.2 with new utilities
- Update SYS_14 protocol to v2.3 with TrialManager integration
- Add LAC learnings for trial management patterns
- Add archive/README.md for deprecated code policy
Key principles:
- Trial numbers NEVER reset (monotonic)
- Folders NEVER get overwritten
- Database always synced with filesystem
- Surrogate predictions are NOT trials (only FEA results)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Adds tests/debug_lateral_discrepancy.py to investigate differences between
Zernike OPD lateral displacement reporting and Simcenter post-processing.
Key findings documented:
- OPD reports sqrt(dx² + dy²) - combined XY magnitude
- Simcenter shows individual components (dx or dy)
- Both are correct, OPD magnitude is more meaningful for optics
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Replace Standard (Z-only) extraction with OPD method (X,Y,Z displacement)
- Add toggle buttons to switch between WFE, ΔX, ΔY, ΔZ views
- Show method comparison metrics (OPD vs Standard RMS difference)
- Display lateral displacement statistics (max/RMS)
- Fall back to Standard method if BDF geometry file not found
- Use ZernikeOPDExtractor for more accurate WFE computation (+8-11%)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Major improvements to Zernike WFE visualization:
- Add ZernikeDashboardInsight: Unified dashboard with all orientations (40°, 60°, 90°)
on one page with light theme and executive summary
- Add OPD method toggle: Switch between Standard (Z-only) and OPD (X,Y,Z) methods
in ZernikeWFEInsight with interactive buttons
- Add lateral displacement maps: Visualize X,Y displacement for each orientation
- Add displacement component views: Toggle between WFE, ΔX, ΔY, ΔZ in relative views
- Add metrics comparison table showing both methods side-by-side
New extractors:
- extract_zernike_figure.py: ZernikeOPDExtractor using BDF geometry interpolation
- extract_zernike_opd.py: Parabola-based OPD with focal length
Key finding: OPD method gives 8-11% higher WFE values than Standard method
(more conservative/accurate for surfaces with lateral displacement under gravity)
Documentation updates:
- SYS_12: Added E22 ZernikeOPD as recommended method
- SYS_16: Added ZernikeDashboard, updated ZernikeWFE with OPD features
- Cheatsheet: Added Zernike method comparison table
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Dashboard integration for Study Insights module (SYS_16):
- Backend: New /api/insights/ routes for generating and viewing insights
- Frontend: New Insights.tsx page with Plotly visualization
- Navigation: Added Insights tab between Analysis and Results
Available insight types:
- Zernike WFE (wavefront error for mirrors)
- Stress Field (Von Mises stress contours)
- Modal Analysis (natural frequencies/mode shapes)
- Thermal Field (temperature distribution)
- Design Space (parameter-objective exploration)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Remove old study folders that have been superseded or archived:
- bracket_pareto_3obj
- bracket_stiffness_optimization (V1-V3)
- bracket_stiffness_optimization_atomizerfield
- drone_gimbal_arm_optimization
- m1_mirror_adaptive_V11 through V15
- m1_mirror_zernike_optimization
- simple_beam_optimization
- training_data_export_test
- uav_arm_atomizerfield_test
These studies have been consolidated or are no longer needed.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
New Tools (tools/):
- analyze_study.py: Generate comprehensive optimization reports
- find_best_iteration.py: Find best iteration folder, optionally copy it
- archive_best_design.py: Archive best design to 3_results/best_design_archive/<timestamp>/
Protocol Updates:
- OP_02_RUN_OPTIMIZATION.md v1.1: Add mandatory archive_best_design step
in Post-Run Actions. This MUST be done after every optimization run.
V14 Updates:
- run_optimization.py: Auto-archive best design at end of optimization
- optimization_config.json: Expand bounds for V14 continuation
- lateral_outer_angle: min 13->11 deg (was at 4.7%)
- lateral_inner_pivot: min 7->5 mm (was at 8.1%)
- lateral_middle_pivot: max 23->27 mm (was at 99.4%)
- whiffle_min: max 60->72 mm (was at 96.3%)
Usage:
python tools/analyze_study.py m1_mirror_adaptive_V14
python tools/find_best_iteration.py m1_mirror_adaptive_V14
python tools/archive_best_design.py m1_mirror_adaptive_V14
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
New E11 Part Mass Extractor:
- Add nx_journals/extract_part_mass_material.py - NX journal using
NXOpen.MeasureManager.NewMassProperties() for accurate geometry-based mass
- Add optimization_engine/extractors/extract_part_mass_material.py - Python
wrapper that reads JSON output from journal
- Add E11 entry to extractors/catalog.json
Documentation Updates:
- SYS_12_EXTRACTOR_LIBRARY.md: Add mass accuracy warning noting pyNastran
get_mass_breakdown() under-reports ~7% on hex-dominant meshes with
tet/pyramid fill elements. E11 (geometry .prt) should be preferred over
E4 (BDF) unless material is overridden at FEM level.
- 01_CHEATSHEET.md: Add mass extraction tip
V14 Config:
- Expand design variable bounds (blank_backface_angle max 4.5°,
whiffle_triangle_closeness max 80mm, whiffle_min max 60mm)
Testing showed:
- E11 from .prt: 97.66 kg (accurate - matches NX GUI)
- E4 pyNastran get_mass_breakdown(): 90.73 kg (~7% under-reported)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Comprehensive guide teaching users how to interact with Atomizer so that
the learning system evolves correctly. Covers:
- The right mindset (colleague, not tool)
- Starting sessions with proper context
- Communicating goals, constraints, preferences
- Creating and running optimization studies
- Analyzing and validating results
- Reporting errors effectively
- Contributing to LAC (recording insights, outcomes, workarounds)
- Ending sessions properly to capture learnings
Includes:
- Mermaid diagrams for learning loop and flows
- Good vs bad examples for every interaction type
- Complete example session transcript
- Quick reference card for common patterns
- Golden rules summary
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Add persistent knowledge system that enables Atomizer to learn from every
session and improve over time.
## New Files
- knowledge_base/lac.py: LAC class with optimization memory, session insights,
and skill evolution tracking
- knowledge_base/__init__.py: Package initialization
- .claude/skills/modules/learning-atomizer-core.md: Full LAC skill documentation
- docs/07_DEVELOPMENT/ATOMIZER_CLAUDE_CODE_INSTRUCTIONS.md: Master instructions
## Updated Files
- CLAUDE.md: Added LAC section, communication style, AVERVS execution framework,
error classification, and "Atomizer Claude" identity
- 00_BOOTSTRAP.md: Added session startup/closing checklists with LAC integration
- 01_CHEATSHEET.md: Added LAC CLI and Python API quick reference
- 02_CONTEXT_LOADER.md: Added LAC query section and anti-pattern
## LAC Features
- Query similar past optimizations before starting new ones
- Record insights (failures, success patterns, workarounds)
- Record optimization outcomes for future reference
- Suggest protocol improvements based on discoveries
- Simple JSONL storage (no database required)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Analysis of 258 FEA trials showed best trial (#45) hitting bounds on
5 parameters. Expanded bounds to allow exploration of promising regions:
- lateral_inner_angle: max 28.5 → 30.0° (was at 99.2% of range)
- lateral_inner_pivot: min 9.0 → 7.0 mm (was at 4.6% of range)
- lateral_middle_pivot: min 18.0 → 15.0 mm (was at 7.7% of range)
- whiffle_min: min 35.0 → 30.0 mm (was at 4.0% of range)
- whiffle_outer_to_vertical: min 68.0 → 60.0° (was at 5.3% of range)
- blank_backface_angle: narrowed to 4.1-4.2° (focus on optimal region)
V14 seeds from 496 prior FEA trials (V11+V12+V13) using TPE sampler.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The Method Selector now uses relative accuracy thresholds to assess
NN suitability by comparing NN error to problem variability (CV ratio).
NNQualityAssessor features:
- Physics-based objective classification (linear, smooth, nonlinear, chaotic)
- CV ratio computation: nn_error / coefficient_of_variation
- Turbo suitability score based on relative thresholds
- Data collection from validation_report.json, turbo_report.json, and study.db
Quality thresholds by objective type:
- Linear (mass, volume): max 2% error, CV ratio < 0.5
- Smooth (frequency): max 5% error, CV ratio < 1.0
- Nonlinear (stress, stiffness): max 10% error, CV ratio < 2.0
- Chaotic (contact, buckling): max 20% error, CV ratio < 3.0
CLI output now includes:
- Per-objective NN quality table with error, CV, ratio, and quality indicator
- Turbo suitability and hybrid suitability percentages
- Warnings when NN error exceeds physics-based thresholds
Updated SYS_15_METHOD_SELECTOR.md to v2.0 with full NN Quality Assessment documentation.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Phase 2 - Structural Analysis:
- extract_principal_stress: σ1, σ2, σ3 principal stresses from OP2
- extract_strain_energy: Element and total strain energy
- extract_spc_forces: Reaction forces at boundary conditions
Phase 3 - Multi-Physics:
- extract_temperature: Nodal temperatures from thermal OP2 (SOL 153/159)
- extract_temperature_gradient: Thermal gradient approximation
- extract_heat_flux: Element heat flux from thermal analysis
- extract_modal_mass: Modal effective mass from F06 (SOL 103)
- get_first_frequency: Convenience function for first natural frequency
Documentation:
- Updated SYS_12_EXTRACTOR_LIBRARY.md with E12-E18 specifications
- Updated NX_OPEN_AUTOMATION_ROADMAP.md marking Phase 3 complete
- Added test_phase3_extractors.py for validation
All extractors follow consistent API pattern returning Dict with
success, data, and error fields for robust error handling.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Dashboard enhancements:
- Add Analysis page with tabs: Overview, Parameters, Pareto, Correlations, Constraints, Surrogate, Runs
- Add PlotlyCorrelationHeatmap for parameter-objective correlation analysis
- Add PlotlyFeasibilityChart for constraint satisfaction visualization
- Add PlotlySurrogateQuality for FEA vs NN prediction comparison
- Add PlotlyRunComparison for comparing optimization runs within a study
Real-time improvements:
- Replace watchdog file-watching with SQLite database polling for better Windows reliability
- Add DatabasePoller class with 2-second polling interval
- Enhanced WebSocket messages: trial_completed, new_best, pareto_update, progress
Desktop notifications:
- Add useNotifications hook using Web Notifications API
- Add NotificationSettings toggle component
- Notify users when new best solutions are found
Config editor:
- Add PUT /studies/{study_id}/config endpoint with auto-backup
- Add ConfigEditor modal with tabs: General, Variables, Objectives, Settings, JSON
- Prevents editing while optimization is running
Enhanced Pareto visualization:
- Add dark mode styling with transparent backgrounds
- Add stats bar showing Pareto, FEA, NN, and infeasible counts
- Add Pareto front connecting line for 2D view
- Add table showing top 10 Pareto-optimal solutions
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Phase 1 - Accurate study status detection:
- Add is_optimization_running() to check for active processes
- Add get_accurate_study_status() with proper status logic
- Status now: not_started, running, paused, completed
- Add "paused" status styling (orange) to Home page
Phase 2 - Global Claude terminal:
- Create ClaudeTerminalContext for app-level state
- Create GlobalClaudeTerminal floating component
- Terminal persists across page navigation
- Shows green indicator when connected
- Remove inline terminal from Dashboard
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Reorganize dashboard: control panel on top, charts stacked vertically
- Add Set Context button to Claude terminal for study awareness
- Add conda environment instructions to CLAUDE.md
- Fix STUDY_REPORT.md location in generate-report.md skill
- Claude terminal now sends study context with skills reminder
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Add trial limiting (300 max) and reduce polling to 15s for large studies
- Make dashboard layout wider with col-span adjustments
- Claude terminal now runs from Atomizer root for CLAUDE.md/skills access
- Add study context display in terminal on connect
- Add KaTeX math rendering styles for study reports
- Add surrogate tuner module for hyperparameter optimization
- Fix backend proxy to port 8001
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Add embedded Claude Code terminal with xterm.js for full CLI experience
- Create WebSocket PTY backend for real-time terminal communication
- Add terminal status endpoint to check CLI availability
- Update dashboard to use Claude Code terminal instead of API chat
- Add optimization control panel with start/stop/validate actions
- Add study context provider for global state management
- Update frontend with new dependencies (xterm.js addons)
- Comprehensive README documentation for all new features
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
- python launch_dashboard.py starts both backend and frontend
- Ctrl+C gracefully shuts down both servers
- Color-coded terminal output for status
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Add ConvergencePlot component with running best, statistics, gradient fill
- Add ParameterImportanceChart with Pearson correlation analysis
- Add StudyReportViewer with KaTeX math rendering and full markdown support
- Update pruning endpoint to query Optuna database directly
- Add /report endpoint for STUDY_REPORT.md files
- Fix chart data transformation for single/multi-objective studies
- Update Protocol 13 documentation with new components
- Update generate-report skill with dashboard integration
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
- environment.yml: Added PyTorch with CUDA 12.1, PyG (torch-geometric),
and TensorBoard for neural network training
- INSTALL_INSTRUCTIONS.md: Step-by-step guide for installing Miniconda
and setting up the Atomizer environment
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
- requirements.txt: Added all dependencies including PyTorch,
torch-geometric, tensorboard for neural network training
- install.bat: One-click installation script that installs all
dependencies with proper version constraints
- train_neural.bat: Training script that runs parametric neural
network training on collected FEA data
Usage:
1. Double-click install.bat to install dependencies
2. Double-click train_neural.bat to train on bracket study
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Rebuilds missing neural network components based on documentation:
- neural_models/parametric_predictor.py: Design-conditioned GNN that
predicts all 4 optimization objectives (mass, frequency, displacement,
stress) directly from design parameters. ~500K trainable parameters.
- train_parametric.py: Training script with multi-objective loss,
checkpoint saving with normalization stats, and TensorBoard logging.
- Updated __init__.py to export ParametricFieldPredictor and
create_parametric_model for use by optimization_engine/neural_surrogate.py
These files enable the neural acceleration workflow:
1. Collect FEA training data (189 trials already collected)
2. Train parametric model: python train_parametric.py --train_dir ...
3. Run neural-accelerated optimization with --enable-nn flag
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Permanently integrates the Atomizer-Field GNN surrogate system:
- neural_models/: Graph Neural Network for FEA field prediction
- batch_parser.py: Parse training data from FEA exports
- train.py: Neural network training pipeline
- predict.py: Inference engine for fast predictions
This enables 600x-2200x speedup over traditional FEA by replacing
expensive simulations with millisecond neural network predictions.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
