Enhance Figure 1 with systems-focused comments and fix LaTeX itemize errors

## Figure 1 Enhancements Added informative comments to TinyTorch code example (Figure 1c) to highlight systems insights students gain from implementation: **Linear layer:** - Memory calculation: `out × in_features × 4B` - Computational complexity: `O(batch × in × out) FLOPs` **Adam optimizer:** - Memory insight: "Why 2× memory vs SGD?" (momentum + variance buffers) - Algorithm clarity: "Exponential moving avg" for momentum/variance updates - Optimization detail: "Per-parameter adaptive lr" for update rule These comments make the contrast with PyTorch/TensorFlow (Figures 1a-b) more explicit: black-box API usage vs. understanding memory costs, computational complexity, and algorithmic trade-offs. ## LaTeX Fixes Fixed 4 itemize environment errors from empirical validation roadmap section: - Added blank lines before all \begin{itemize} environments - Affected sections: Quantitative Success Criteria (line 600), Phase 1-3 (lines 1036, 1045, 1054) Paper now compiles cleanly: 24 pages, 389KB PDF, no LaTeX errors. 🤖 Generated with Claude Code (https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
2026-05-21 15:31:00 -05:00 · 2025-11-18 19:06:55 -05:00
parent 313e23e6fb
commit 272ffd6409
7 changed files with 606 additions and 289 deletions
--- a/COMPREHENSIVE_MODULE_REVIEW.md
+++ b/COMPREHENSIVE_MODULE_REVIEW.md
@@ -0,0 +1,453 @@
+# Comprehensive Module Review: Educational Framework Analysis
+
+**Date**: [Current]  
+**Reviewer**: Expert PyTorch Developer + Educational Materials Specialist  
+**Scope**: All 20 modules (01-20) + archived modules
+
+---
+
+## Executive Summary
+
+Overall, the TinyTorch framework demonstrates **strong progressive building** and **educational clarity**. However, several **anti-patterns** from AI code generation ("reward hacking") have been identified that undermine the educational experience. This review identifies issues and provides actionable fixes.
+
+### Key Findings
+
+✅ **Strengths**:
+- Clear progressive building from Module 01 → 20
+- Explicit imports in most modules (no star imports)
+- Good dependency documentation in recent modules (13, 16, 18)
+- Consistent module structure and organization
+
+⚠️ **Issues Found**:
+1. **`import_previous_module` helper function** - Used in Modules 02, 04, 06 (problematic pattern)
+2. **Try/except fallbacks** - Some modules silently handle missing dependencies
+3. **Inconsistent error handling** - Some modules raise educational errors, others fail silently
+4. **Module 20 capstone** - Optional imports handled with None assignments (should be explicit)
+
+---
+
+## Detailed Module-by-Module Review
+
+### ✅ Module 01: Tensor (Foundation)
+**Status**: ✅ EXCELLENT
+- **Imports**: Clean, no dependencies (foundation module)
+- **Progressive Building**: N/A (foundation)
+- **Educational Clarity**: Excellent documentation
+- **Issues**: None
+
+**Pattern**:
+```python
+import numpy as np
+# No dependencies - this is the foundation
+```
+
+---
+
+### ⚠️ Module 02: Activations
+**Status**: ⚠️ NEEDS FIX
+- **Imports**: Uses `import_previous_module` helper function
+- **Progressive Building**: Should import Tensor directly
+- **Educational Clarity**: Good, but helper function obscures dependencies
+- **Issues**: 
+  - `import_previous_module()` function uses `sys.path.append()` (line 828)
+  - Used in test code, but pattern is problematic
+
+**Current Pattern** (PROBLEMATIC):
+```python
+def import_previous_module(module_name: str, component_name: str):
+    import sys
+    import os
+    sys.path.append(os.path.join(os.path.dirname(__file__), '..', module_name))
+    module = __import__(f"{module_name.split('_')[1]}_dev")
+    return getattr(module, component_name)
+```
+
+**Should Be**:
+```python
+from tinytorch.core.tensor import Tensor
+```
+
+**Recommendation**: Remove `import_previous_module` helper. Use direct imports from `tinytorch` package.
+
+---
+
+### ✅ Module 03: Layers
+**Status**: ✅ GOOD (with minor note)
+- **Imports**: Direct imports from `tinytorch.core.tensor` and `tinytorch.core.activations`
+- **Progressive Building**: Clear dependencies
+- **Educational Clarity**: Excellent
+- **Issues**: 
+  - Commented-out `import_previous_module` function (line 750) - should be removed entirely
+  - Uses `import_previous_module` in test code (line 781) - should use direct imports
+
+**Current Pattern**:
+```python
+from tinytorch.core.tensor import Tensor
+from tinytorch.core.activations import ReLU, Sigmoid
+```
+
+**Recommendation**: Remove commented-out `import_previous_module` code and update test code to use direct imports.
+
+---
+
+### ⚠️ Module 04: Losses
+**Status**: ⚠️ NEEDS FIX
+- **Imports**: Uses `import_previous_module` helper function
+- **Progressive Building**: Should import Tensor, Linear, ReLU directly
+- **Educational Clarity**: Good documentation, but helper obscures dependencies
+- **Issues**: 
+  - `import_previous_module()` function defined (line 84)
+  - Uses `sys.path.append()` pattern
+  - However, ALSO has direct imports (lines 92-94) - inconsistent!
+
+**Current Pattern** (INCONSISTENT):
+```python
+def import_previous_module(module_name: str, component_name: str):
+    # ... sys.path.append pattern ...
+
+# Import from tinytorch package
+from tinytorch.core.tensor import Tensor
+from tinytorch.core.layers import Linear
+from tinytorch.core.activations import ReLU
+```
+
+**Recommendation**: Remove `import_previous_module` function entirely. Keep only direct imports.
+
+---
+
+### ✅ Module 05: Autograd
+**Status**: ✅ GOOD (with acceptable exception)
+- **Imports**: Direct imports from tinytorch package
+- **Progressive Building**: Clear dependencies
+- **Educational Clarity**: Excellent
+- **Issues**: 
+  - Try/except ImportError that silently passes (line 1149) - **ACCEPTABLE** for optional dependencies during development
+  - This is fine because it's handling optional monkey-patching of loss functions
+
+**Current Pattern** (ACCEPTABLE):
+```python
+try:
+    # Monkey-patch loss functions for autograd tracking
+    BinaryCrossEntropyLoss.forward = tracked_bce_forward
+    # ...
+except ImportError:
+    # Activations/losses not yet available (happens during module development)
+    pass
+```
+
+**Recommendation**: Keep as-is. This is acceptable for optional dependencies during development.
+
+---
+
+### ⚠️ Module 06: Optimizers
+**Status**: ⚠️ NEEDS FIX
+- **Imports**: Direct imports in main code, but uses `import_previous_module` in test code
+- **Progressive Building**: Clear in main code
+- **Educational Clarity**: Good
+- **Issues**: 
+  - Uses `import_previous_module` in test code (lines 1261-1264)
+  - Commented-out function definition (line 1228)
+
+**Current Pattern** (TEST CODE ISSUE):
+```python
+# In test code:
+Tensor = import_previous_module('01_tensor', 'Tensor')
+Linear = import_previous_module('03_layers', 'Linear')
+ReLU = import_previous_module('02_activations', 'ReLU')
+MSELoss = import_previous_module('04_losses', 'MSELoss')
+```
+
+**Recommendation**: Update test code to use direct imports:
+```python
+from tinytorch.core.tensor import Tensor
+from tinytorch.core.layers import Linear
+from tinytorch.core.activations import ReLU
+from tinytorch.core.losses import MSELoss
+```
+
+---
+
+### ✅ Module 07: Training
+**Status**: ✅ EXCELLENT
+- **Imports**: Direct imports, clean
+- **Progressive Building**: Clear dependencies
+- **Educational Clarity**: Excellent
+- **Issues**: None (commented-out `import_previous_module` should be removed)
+
+**Pattern**:
+```python
+from tinytorch.core.tensor import Tensor
+from tinytorch.core.layers import Linear
+from tinytorch.core.losses import MSELoss, CrossEntropyLoss
+from tinytorch.core.optimizers import SGD, AdamW
+```
+
+---
+
+### ✅ Module 08: DataLoader
+**Status**: ✅ EXCELLENT
+- **Imports**: Direct imports, clean
+- **Progressive Building**: Clear (only needs Tensor)
+- **Educational Clarity**: Excellent
+- **Issues**: None
+
+---
+
+### ✅ Module 09: Spatial
+**Status**: ✅ EXCELLENT
+- **Imports**: Direct imports, clean
+- **Progressive Building**: Clear (only needs Tensor)
+- **Educational Clarity**: Excellent
+- **Issues**: None
+
+---
+
+### ✅ Module 10: Tokenization
+**Status**: ✅ EXCELLENT
+- **Imports**: Direct imports, clean
+- **Progressive Building**: Clear (only needs Tensor)
+- **Educational Clarity**: Excellent
+- **Issues**: None
+
+---
+
+### ✅ Module 11: Embeddings
+**Status**: ✅ EXCELLENT
+- **Imports**: Direct imports, clean
+- **Progressive Building**: Clear (only needs Tensor)
+- **Educational Clarity**: Excellent
+- **Issues**: None
+
+---
+
+### ✅ Module 12: Attention
+**Status**: ✅ EXCELLENT
+- **Imports**: Direct imports, clean
+- **Progressive Building**: Clear dependencies (Tensor, Linear)
+- **Educational Clarity**: Excellent
+- **Issues**: None
+
+---
+
+### ✅ Module 13: Transformers
+**Status**: ✅ EXCELLENT (RECENTLY FIXED)
+- **Imports**: Direct imports with educational error messages
+- **Progressive Building**: Excellent dependency documentation
+- **Educational Clarity**: Excellent
+- **Issues**: None (fixed in recent hot fixes)
+
+**Pattern** (EXEMPLARY):
+```python
+try:
+    from tinytorch.core.tensor import Tensor  # Module 01: Foundation
+except ImportError as e:
+    raise ImportError(
+        "❌ Module 13 (Transformers) requires Module 01 (Tensor) to be completed first.\n"
+        "   Please complete Module 01 first, then run 'tito module complete 01'.\n"
+        "   Original error: " + str(e)
+    ) from e
+```
+
+---
+
+### ✅ Module 14: Profiling
+**Status**: ✅ EXCELLENT
+- **Imports**: Direct imports, clean
+- **Progressive Building**: Clear dependencies
+- **Educational Clarity**: Excellent
+- **Issues**: None
+
+---
+
+### ✅ Module 15: Quantization
+**Status**: ✅ EXCELLENT
+- **Imports**: Direct imports, clean
+- **Progressive Building**: Clear dependencies
+- **Educational Clarity**: Excellent
+- **Issues**: None
+
+---
+
+### ✅ Module 16: Compression
+**Status**: ✅ EXCELLENT (RECENTLY FIXED)
+- **Imports**: Direct imports with educational error messages
+- **Progressive Building**: Excellent dependency documentation
+- **Educational Clarity**: Excellent
+- **Issues**: None (fixed in recent hot fixes)
+- **Note**: Sequential class defined locally as testing utility (correct)
+
+---
+
+### ✅ Module 17: Memoization
+**Status**: ✅ EXCELLENT
+- **Imports**: Direct imports, clean
+- **Progressive Building**: Clear dependencies
+- **Educational Clarity**: Excellent
+- **Issues**: None
+
+---
+
+### ✅ Module 18: Acceleration
+**Status**: ✅ EXCELLENT (RECENTLY FIXED)
+- **Imports**: Direct imports with educational error messages
+- **Progressive Building**: Excellent dependency documentation
+- **Educational Clarity**: Excellent
+- **Issues**: None (fixed in recent hot fixes)
+
+---
+
+### ⚠️ Module 19: Benchmarking
+**Status**: ⚠️ ACCEPTABLE (with note)
+- **Imports**: Direct imports + optional external dependencies
+- **Progressive Building**: Clear
+- **Educational Clarity**: Good
+- **Issues**: 
+  - Try/except for optional external dependencies (pandas, matplotlib, psutil) - **ACCEPTABLE**
+  - These are external libraries, not TinyTorch modules
+  - Creates fallback classes when pandas not available (line 177) - acceptable for optional dependency
+
+**Current Pattern** (ACCEPTABLE):
+```python
+try:
+    import pandas as pd
+    PANDAS_AVAILABLE = True
+except ImportError:
+    PANDAS_AVAILABLE = False
+    # Create a simple DataFrame-like class for when pandas is not available
+    class pd:
+        class DataFrame:
+            # ... minimal fallback
+```
+
+**Recommendation**: Keep as-is. This is acceptable for optional external dependencies.
+
+---
+
+### ⚠️ Module 20: Capstone
+**Status**: ⚠️ NEEDS IMPROVEMENT
+- **Imports**: Direct imports + try/except for optional features
+- **Progressive Building**: Clear
+- **Educational Clarity**: Good, but could be more explicit
+- **Issues**: 
+  - Try/except that sets things to None (lines 254-266)
+  - Should raise educational errors instead of silently setting to None
+
+**Current Pattern** (COULD BE BETTER):
+```python
+try:
+    from tinytorch.optimization.quantization import QuantizedLinear
+except ImportError:
+    QuantizedLinear = None  # Not available
+
+try:
+    from tinytorch.optimization.compression import magnitude_prune, structured_prune
+except ImportError:
+    magnitude_prune = None
+    structured_prune = None
+```
+
+**Recommendation**: Either:
+1. Make these required dependencies (raise educational errors)
+2. Or document clearly that these are optional features
+
+---
+
+## Critical Issues Summary
+
+### 🔴 HIGH PRIORITY
+
+1. **Remove `import_previous_module` helper function**
+   - **Modules**: 02, 04, 06
+   - **Impact**: Obscures dependencies, uses non-portable sys.path manipulation
+   - **Fix**: Replace with direct imports from `tinytorch` package
+
+2. **Update test code to use direct imports**
+   - **Modules**: 02, 03, 06
+   - **Impact**: Test code should follow same patterns as main code
+   - **Fix**: Replace `import_previous_module` calls with direct imports
+
+### 🟡 MEDIUM PRIORITY
+
+3. **Module 20: Make optional dependencies explicit**
+   - **Impact**: Students might not understand why features are None
+   - **Fix**: Either make required or document clearly as optional
+
+4. **Remove commented-out code**
+   - **Modules**: 01, 03, 06, 07
+   - **Impact**: Clutters codebase
+   - **Fix**: Remove commented-out `import_previous_module` functions
+
+---
+
+## Recommended Fixes
+
+### Fix 1: Remove `import_previous_module` from Module 02
+**File**: `modules/02_activations/activations_dev.py`
+- Remove function definition (line 828)
+- Update test code to use direct imports
+
+### Fix 2: Remove `import_previous_module` from Module 04
+**File**: `modules/04_losses/losses_dev.py`
+- Remove function definition (line 84)
+- Keep only direct imports (already present)
+
+### Fix 3: Update Module 06 test code
+**File**: `modules/06_optimizers/optimizers_dev.py`
+- Remove commented-out function (line 1228)
+- Update test code (lines 1261-1264) to use direct imports
+
+### Fix 4: Improve Module 20 optional dependencies
+**File**: `modules/20_capstone/capstone.py`
+- Either make dependencies required (raise educational errors)
+- Or document clearly as optional features with clear messaging
+
+### Fix 5: Clean up commented code
+**Files**: Multiple modules
+- Remove all commented-out `import_previous_module` functions
+
+---
+
+## Educational Principles Validation
+
+### ✅ Progressive Building
+- **Status**: EXCELLENT
+- **Evidence**: Clear dependency chains, modules build on each other logically
+- **Note**: Some modules use helper functions that obscure this
+
+### ✅ Explicit Imports
+- **Status**: EXCELLENT (after fixes)
+- **Evidence**: No star imports found, direct imports used throughout
+- **Note**: Helper functions need to be removed
+
+### ✅ Educational Clarity
+- **Status**: EXCELLENT
+- **Evidence**: Good documentation, clear learning objectives
+- **Note**: Recent modules (13, 16, 18) have exemplary dependency documentation
+
+### ✅ No Reward Hacking
+- **Status**: GOOD (after recent fixes)
+- **Evidence**: Recent fixes removed duplicate classes, fallbacks, hardcoded paths
+- **Note**: `import_previous_module` pattern is a form of reward hacking
+
+---
+
+## Conclusion
+
+The TinyTorch framework is **well-structured and educational**, with recent fixes addressing major issues. The remaining issues are primarily:
+
+1. **Legacy helper functions** (`import_previous_module`) that should be removed
+2. **Test code** that should follow same patterns as main code
+3. **Optional dependencies** that should be more explicitly documented
+
+**Overall Grade**: A- (Excellent, with minor improvements needed)
+
+**Priority Actions**:
+1. Remove `import_previous_module` from Modules 02, 04
+2. Update test code in Modules 02, 03, 06
+3. Improve Module 20 optional dependency handling
+4. Clean up commented code
+
+---
+
+**Next Steps**: Implement fixes for high-priority issues.
+
--- a/paper/paper.aux
+++ b/paper/paper.aux
--- a/paper/paper.bbl
+++ b/paper/paper.bbl
@@ -1,4 +1,4 @@
-\begin{thebibliography}{37}
+\begin{thebibliography}{39}
 \providecommand{\natexlab}[1]{#1}
 \providecommand{\url}[1]{\texttt{#1}}
 \expandafter\ifx\csname urlstyle\endcsname\relax
@@ -21,173 +21,11 @@ Colby~R. Banbury, Vijay~Janapa Reddi, Max Lam, William Fu, Amin Fazel, Jeremy
 \newblock \emph{arXiv preprint arXiv:2003.04821}.
 \newblock URL \url{http://arxiv.org/abs/2003.04821v4}.

-\bibitem[Burstein et~al.(1992)Burstein, Henry, Collison, Marczak, Sligar,
-  Watson, Marquez, Abbasalizad-Farhangi, Abbasi, Abd-Allah, Abdoli, Abdollahi,
-  Abdollahpour, Abdulkader, Abrigo, Acharya, Adebayo, Adekanmbi, Adham,
-  Afshari, Aghaali, Ahmadi, Ahmadi, Ahmadpour, Ahmed, Akal, Akinyemi, Alahdab,
-  Alam, Alamene, Alene, Alijanzadeh, Alinia, Alipour, Aljunid, Almalki,
-  Al-Mekhlafi, Altirkawi, Alvis-Guzman, Amegah, Amini, Amit, Anbari, Androudi,
-  Anjomshoa, Ansari, Antonio, Arabloo, Arefi, Aremu, Armoon, Arora, Artaman,
-  Asadi, Asadi-Aliabadi, Ashraf-Ganjouei, Assadi, Ataeinia, Atre, Quintanilla,
-  Ayanore, Azari, Babaee, Babazadeh, Badawi, Bagheri, Bagherzadeh, Baheiraei,
-  Balouchi, Barac, Bassat, Baune, Bayati, Bedi, Beghi, Behzadifar, Behzadifar,
-  Belay, Bell, Bell, Berbada, Bernstein, Bhattacharjee, Bhattarai, Bhutta,
-  Bijani, Bohlouli, Breitborde, Britton, Browne, Nagaraja, Busse, Butt, Car,
-  Cárdenas, Castañeda-Orjuela, Cerin, Chanie, Chatterjee, Chu, Cooper, Costa,
-  Dalal, Dandona, Dandona, Daoud, Daryani, Das~Gupta, Davis, Davis~Weaver,
-  Davitoiu, De~Neve, Demeke, Demoz, Deribe, Desai, Deshpande, Desyibelew, Dey,
-  Dharmaratne, Dhimal, Diaz, Doshmangir, Duraes, Dwyer-Lindgren, Earl,
-  Ebrahimi, Ebrahimpour, Effiong, Eftekhari, Ehsani-Chimeh, El~Sayed,
-  El~Sayed~Zaki, El~Tantawi, El-Khatib, Emamian, Enany, Eskandarieh, Eyawo,
-  Ezalarab, Faramarzi, Fareed, Faridnia, Faro, Fazaeli, Fazlzadeh, Fentahun,
-  Fereshtehnejad, Fernandes, Filip, Fischer, Foigt, Foroutan, Francis,
-  Fukumoto, Fullman, Gallus, Gebre, Gebrehiwot, Gebremeskel, Gessner, Geta,
-  Gething, Ghadimi, Ghadiri, Ghajarzadeh, Ghashghaee, Gill, Gill, Golding,
-  Gomes, Gona, Gopalani, Gorini, Goulart, Graetz, Greaves, Green, Guo,
-  Haj-Mirzaian, Haj-Mirzaian, Hall, Hamidi, Haririan, Haro, Hasankhani,
-  Hasanpoor, Hasanzadeh, Hassankhani, Hassen, Hegazy, Hendrie, Heydarpour,
-  Hird, Hoang, Hollerich, Rad, Hoseini-Ghahfarokhi, Hossain, Hosseini,
-  Hosseinzadeh, Hostiuc, Hostiuc, Househ, Hsairi, Ilesanmi, Imani-Nasab, Iqbal,
-  Irvani, Islam, Islam, Jürisson, Balalami, Jalali, Javidnia, Jayatilleke,
-  Jenabi, Ji, Jobanputra, Johnson, Jonas, Shushtari, Jozwiak, Kabir, Kahsay,
-  Kalani, Kalhor, Karami, Karki, Kasaeian, Kassebaum, Keiyoro, Kemp, Khabiri,
-  Khader, Khafaie, Khan, Khan, Khan, Khang, Khatab, Khater, Khater, Khatony,
-  Khazaei, Khazaei, Khazaei-Pool, Khubchandani, Kianipour, Kim, Kimokoti,
-  Kinyoki, Kisa, Kisa, Kolola, Kosen, Koul, Koyanagi, Kraemer, Krishan, Krohn,
-  Kugbey, Kumar, Kumar, Kumar, Kuupiel, Lacey, Lad, Lami, Larsson, Lee, Leili,
-  Levine, Li, Lim, Listl, Longbottom, Lopez, Lorkowski, Magdeldin, Abd
-  El~Razek, Abd El~Razek, Majeed, Maleki, Malekzadeh, Malta, Mamun, Manafi,
-  Manda, Mansourian, Martins-Melo, Masaka, Massenburg, Maulik, Mayala, Mazidi,
-  McKee, Mehrotra, Mehta, Meles, Mendoza, Menezes, Meretoja, Meretoja,
-  Mestrovic, Miller, Miller-Petrie, Mills, Milne, Mini, Mir, Mirjalali,
-  Mirrakhimov, Mohamadi, Mohammad, Darwesh, Mezerji, Mohammed, Mohammed,
-  Mokdad, Molokhia, Monasta, Moodley, Moosazadeh, Moradi, Moradi, Moradi,
-  Moradi-Lakeh, Moradinazar, Moraga, Morawska, Mosapour, Mousavi, Mueller,
-  Muluneh, Mustafa, Nabavizadeh, Naderi, Nagarajan, Nahvijou, Najafi, Nangia,
-  Ndwandwe, Neamati, Negoi, Negoi, Ngunjiri, Thi~Nguyen, Nguyen, Nguyen,
-  Nielsen, Ningrum, Nirayo, Nixon, Nnaji, Nojomi, Noroozi, Nosratnejad,
-  Noubiap, Motlagh, Ofori-Asenso, Ogbo, Oladimeji, Olagunju, Olfatifar, Olum,
-  Olusanya, Oluwasanu, Onwujekwe, Oren, Ortega-Altamirano, Ortiz, Osarenotor,
-  Osei, Osgood-Zimmerman, Otstavnov, Owolabi, ~, Pagheh, Pakhale, Panda-Jonas,
-  Pandey, Park, Parsian, Pashaei, Patel, Pepito, Pereira, Perkins, Pickering,
-  Pilgrim, Pirestani, Piroozi, Pirsaheb, Plana-Ripoll, Pourjafar, Puri,
-  Qorbani, Quintana, Rabiee, Rabiee, Radfar, Rafiei, Rahim, Rahimi,
-  Rahimi-Movaghar, Rahimzadeh, Rajati, Raju, Ramezankhani, Ranabhat, Rasella,
-  Rashedi, Rawal, Reiner, Renzaho, Rezaei, Rezapour, Riahi, Ribeiro, Roever,
-  Roro, Roser, Roshandel, Roshani, Rostami, Rubagotti, Rubino, Sabour, Sadat,
-  Sadeghi, Saeedi, Safari, Safari-Faramani, Safdarian, Sahebkar, Salahshoor,
-  Salam, Salamati, Salehi, Zahabi, Salimi, Salimzadeh, Salomon, Sambala, Samy,
-  Santric~Milicevic, Jose, Saraswathy, Sarmiento-Suárez, Sartorius, Sathian,
-  Saxena, Sbarra, Schaeffer, Schwebel, Sepanlou, Seyedmousavi, Shaahmadi,
-  Shaikh, Shams-Beyranvand, Shamshirian, Shamsizadeh, Sharafi, Sharif,
-  Sharif-Alhoseini, Sharifi, Sharma, Sharma, Sheikh, Shields, Shigematsu,
-  Shiri, Shiue, Shuval, Siddiqi, Silva, Singh, Sinha, Sisay, Sisay, Sliwa,
-  Smith, Somayaji, Soofi, Soriano, Sreeramareddy, Sudaryanto, Sufiyan, Sykes,
-  Sylaja, Tabarés-Seisdedos, Tabb, Tabuchi, Taveira, Temsah, Terkawi, Tessema,
-  Thankappan, Thirunavukkarasu, To, Tovani-Palone, Tran, Tran, Ullah, Usman,
-  Uthman, Vahedian-Azimi, Valdez, van Boven, Vasankari, Vasseghian, Veisani,
-  Venketasubramanian, Violante, Vladimirov, Vlassov, Vos, Vu, Vujcic, Waheed,
-  Wakefield, Wang, Wang, Wang, Ward, Weintraub, Weldegwergs, Weldesamuel,
-  Westerman, Wiysonge, Wondafrash, Woyczynski, Wu, Xu, Yadegar, Yamada,
-  Yazdi-Feyzabadi, Yilgwan, Yip, Yonemoto, Lebni, Younis, Yousefifard, Yousof,
-  Yu, Yusefzadeh, Zabeh, Moghadam, Bin~Zaman, Zamani, Zandian, Zangeneh, Zerfu,
-  Zhang, Ziapour, Zodpey, Murray, and Hay]{perkins1992transfer}
-R~Burstein, NJ~Henry, ML~Collison, LB~Marczak, A~Sligar, S~Watson, N~Marquez,
-  M~Abbasalizad-Farhangi, M~Abbasi, F~Abd-Allah, A~Abdoli, M~Abdollahi,
-  I~Abdollahpour, RS~Abdulkader, MRM Abrigo, D~Acharya, OM~Adebayo,
-  V~Adekanmbi, D~Adham, M~Afshari, M~Aghaali, K~Ahmadi, M~Ahmadi, E~Ahmadpour,
-  R~Ahmed, CG~Akal, JO~Akinyemi, F~Alahdab, N~Alam, GM~Alamene, KA~Alene,
-  M~Alijanzadeh, C~Alinia, V~Alipour, SM~Aljunid, MJ~Almalki, HM~Al-Mekhlafi,
-  K~Altirkawi, N~Alvis-Guzman, AK~Amegah, S~Amini, AML Amit, Z~Anbari,
-  S~Androudi, M~Anjomshoa, F~Ansari, CAT Antonio, J~Arabloo, Z~Arefi, O~Aremu,
-  B~Armoon, A~Arora, A~Artaman, A~Asadi, M~Asadi-Aliabadi, A~Ashraf-Ganjouei,
-  R~Assadi, B~Ataeinia, SR~Atre, BPA Quintanilla, MA~Ayanore, S~Azari,
-  E~Babaee, A~Babazadeh, A~Badawi, S~Bagheri, M~Bagherzadeh, N~Baheiraei,
-  A~Balouchi, A~Barac, Q~Bassat, BT~Baune, M~Bayati, N~Bedi, E~Beghi,
-  M~Behzadifar, M~Behzadifar, YB~Belay, B~Bell, ML~Bell, DA~Berbada,
-  RS~Bernstein, NV~Bhattacharjee, S~Bhattarai, ZA~Bhutta, A~Bijani, S~Bohlouli,
-  NJK Breitborde, G~Britton, AJ~Browne, SB~Nagaraja, R~Busse, ZA~Butt, J~Car,
-  R~Cárdenas, CA~Castañeda-Orjuela, E~Cerin, WF~Chanie, P~Chatterjee, DT~Chu,
-  C~Cooper, VM~Costa, K~Dalal, L~Dandona, R~Dandona, F~Daoud, A~Daryani,
-  R~Das~Gupta, I~Davis, N~Davis~Weaver, DV~Davitoiu, JW~De~Neve, FM~Demeke,
-  GT~Demoz, K~Deribe, R~Desai, A~Deshpande, HD~Desyibelew, S~Dey,
-  SD~Dharmaratne, M~Dhimal, D~Diaz, L~Doshmangir, AR~Duraes, L~Dwyer-Lindgren,
-  L~Earl, R~Ebrahimi, S~Ebrahimpour, A~Effiong, A~Eftekhari, E~Ehsani-Chimeh,
-  I~El~Sayed, M~El~Sayed~Zaki, M~El~Tantawi, Z~El-Khatib, MH~Emamian, S~Enany,
-  S~Eskandarieh, O~Eyawo, M~Ezalarab, M~Faramarzi, M~Fareed, R~Faridnia,
-  A~Faro, AA~Fazaeli, M~Fazlzadeh, N~Fentahun, SM~Fereshtehnejad, JC~Fernandes,
-  I~Filip, F~Fischer, NA~Foigt, M~Foroutan, JM~Francis, T~Fukumoto, N~Fullman,
-  S~Gallus, DG~Gebre, TT~Gebrehiwot, GG~Gebremeskel, BD~Gessner, B~Geta,
-  PW~Gething, R~Ghadimi, K~Ghadiri, M~Ghajarzadeh, A~Ghashghaee, PS~Gill,
-  TK~Gill, N~Golding, NGM Gomes, PN~Gona, SV~Gopalani, G~Gorini, BNG Goulart,
-  N~Graetz, F~Greaves, MS~Green, Y~Guo, A~Haj-Mirzaian, A~Haj-Mirzaian,
-  BJ~Hall, S~Hamidi, H~Haririan, JM~Haro, M~Hasankhani, E~Hasanpoor,
-  A~Hasanzadeh, H~Hassankhani, HY~Hassen, MI~Hegazy, D~Hendrie, F~Heydarpour,
-  TR~Hird, CL~Hoang, G~Hollerich, EH~Rad, M~Hoseini-Ghahfarokhi, N~Hossain,
-  M~Hosseini, M~Hosseinzadeh, M~Hostiuc, S~Hostiuc, M~Househ, M~Hsairi,
-  OS~Ilesanmi, MH~Imani-Nasab, U~Iqbal, SSN Irvani, N~Islam, SMS Islam,
-  M~Jürisson, NJ~Balalami, A~Jalali, J~Javidnia, AU~Jayatilleke, E~Jenabi,
-  JS~Ji, YB~Jobanputra, K~Johnson, JB~Jonas, ZJ~Shushtari, JJ~Jozwiak, A~Kabir,
-  A~Kahsay, H~Kalani, R~Kalhor, M~Karami, S~Karki, A~Kasaeian, NJ~Kassebaum,
-  PN~Keiyoro, GR~Kemp, R~Khabiri, YS~Khader, MA~Khafaie, EA~Khan, J~Khan,
-  MS~Khan, YH~Khang, K~Khatab, A~Khater, MM~Khater, A~Khatony, M~Khazaei,
-  S~Khazaei, M~Khazaei-Pool, J~Khubchandani, N~Kianipour, YJ~Kim, RW~Kimokoti,
-  DK~Kinyoki, A~Kisa, S~Kisa, T~Kolola, S~Kosen, PA~Koul, A~Koyanagi, MUG
-  Kraemer, K~Krishan, KJ~Krohn, N~Kugbey, GA~Kumar, M~Kumar, P~Kumar,
-  D~Kuupiel, B~Lacey, SD~Lad, FH~Lami, AO~Larsson, PH~Lee, M~Leili, AJ~Levine,
-  S~Li, LL~Lim, S~Listl, J~Longbottom, JCF Lopez, S~Lorkowski, S~Magdeldin,
-  HM~Abd El~Razek, MM~Abd El~Razek, A~Majeed, A~Maleki, R~Malekzadeh, DC~Malta,
-  AA~Mamun, N~Manafi, AL~Manda, M~Mansourian, FR~Martins-Melo, A~Masaka,
-  BB~Massenburg, PK~Maulik, BK~Mayala, M~Mazidi, M~McKee, R~Mehrotra, KM~Mehta,
-  GG~Meles, W~Mendoza, RG~Menezes, A~Meretoja, TJ~Meretoja, T~Mestrovic,
-  TR~Miller, MK~Miller-Petrie, EJ~Mills, GJ~Milne, GK~Mini, SM~Mir,
-  H~Mirjalali, EM~Mirrakhimov, E~Mohamadi, DK~Mohammad, AM~Darwesh, NMG
-  Mezerji, AS~Mohammed, S~Mohammed, AH~Mokdad, M~Molokhia, L~Monasta,
-  Y~Moodley, M~Moosazadeh, G~Moradi, M~Moradi, Y~Moradi, M~Moradi-Lakeh,
-  M~Moradinazar, P~Moraga, L~Morawska, A~Mosapour, SM~Mousavi, UO~Mueller,
-  AG~Muluneh, G~Mustafa, B~Nabavizadeh, M~Naderi, AJ~Nagarajan, A~Nahvijou,
-  F~Najafi, V~Nangia, DE~Ndwandwe, N~Neamati, I~Negoi, RI~Negoi, JW~Ngunjiri,
-  HL~Thi~Nguyen, LH~Nguyen, SH~Nguyen, KR~Nielsen, DNA Ningrum, YL~Nirayo,
-  MR~Nixon, CA~Nnaji, M~Nojomi, M~Noroozi, S~Nosratnejad, JJ~Noubiap,
-  SN~Motlagh, R~Ofori-Asenso, FA~Ogbo, KE~Oladimeji, AT~Olagunju, M~Olfatifar,
-  S~Olum, BO~Olusanya, MM~Oluwasanu, OE~Onwujekwe, E~Oren, DDV
-  Ortega-Altamirano, A~Ortiz, O~Osarenotor, FB~Osei, AE~Osgood-Zimmerman,
-  SS~Otstavnov, MO~Owolabi, A~M ~, P, AS~Pagheh, S~Pakhale, S~Panda-Jonas,
-  A~Pandey, EK~Park, H~Parsian, T~Pashaei, SK~Patel, VCF Pepito, A~Pereira,
-  S~Perkins, BV~Pickering, T~Pilgrim, M~Pirestani, B~Piroozi, M~Pirsaheb,
-  O~Plana-Ripoll, H~Pourjafar, P~Puri, M~Qorbani, H~Quintana, M~Rabiee,
-  N~Rabiee, A~Radfar, A~Rafiei, F~Rahim, Z~Rahimi, V~Rahimi-Movaghar,
-  S~Rahimzadeh, F~Rajati, SB~Raju, A~Ramezankhani, CL~Ranabhat, D~Rasella,
-  V~Rashedi, L~Rawal, Jr~Reiner, RC, AMN Renzaho, S~Rezaei, A~Rezapour,
-  SM~Riahi, AI~Ribeiro, L~Roever, EM~Roro, M~Roser, G~Roshandel, D~Roshani,
-  A~Rostami, E~Rubagotti, S~Rubino, S~Sabour, N~Sadat, E~Sadeghi, R~Saeedi,
-  Y~Safari, R~Safari-Faramani, M~Safdarian, A~Sahebkar, MR~Salahshoor, N~Salam,
-  P~Salamati, F~Salehi, SS~Zahabi, Y~Salimi, H~Salimzadeh, JA~Salomon,
-  EZ~Sambala, AM~Samy, MM~Santric~Milicevic, BPS Jose, SYI Saraswathy,
-  R~Sarmiento-Suárez, B~Sartorius, B~Sathian, S~Saxena, AN~Sbarra,
-  LE~Schaeffer, DC~Schwebel, SG~Sepanlou, S~Seyedmousavi, F~Shaahmadi,
-  MA~Shaikh, M~Shams-Beyranvand, A~Shamshirian, M~Shamsizadeh, K~Sharafi,
-  M~Sharif, M~Sharif-Alhoseini, H~Sharifi, J~Sharma, R~Sharma, A~Sheikh,
-  C~Shields, M~Shigematsu, R~Shiri, I~Shiue, K~Shuval, TJ~Siddiqi, JP~Silva,
-  JA~Singh, DN~Sinha, MM~Sisay, S~Sisay, K~Sliwa, DL~Smith, R~Somayaji,
-  M~Soofi, JB~Soriano, CT~Sreeramareddy, A~Sudaryanto, MB~Sufiyan, BL~Sykes,
-  PN~Sylaja, R~Tabarés-Seisdedos, KM~Tabb, T~Tabuchi, N~Taveira, MH~Temsah,
-  AS~Terkawi, ZT~Tessema, KR~Thankappan, S~Thirunavukkarasu, QG~To,
-  MR~Tovani-Palone, BX~Tran, KB~Tran, I~Ullah, MS~Usman, OA~Uthman,
-  A~Vahedian-Azimi, PR~Valdez, JFM van Boven, TJ~Vasankari, Y~Vasseghian,
-  Y~Veisani, N~Venketasubramanian, FS~Violante, SK~Vladimirov, V~Vlassov,
-  T~Vos, GT~Vu, IS~Vujcic, Y~Waheed, J~Wakefield, H~Wang, Y~Wang, YP~Wang,
-  JL~Ward, RG~Weintraub, KG~Weldegwergs, GT~Weldesamuel, R~Westerman,
-  CS~Wiysonge, DZ~Wondafrash, L~Woyczynski, AM~Wu, G~Xu, A~Yadegar, T~Yamada,
-  V~Yazdi-Feyzabadi, CS~Yilgwan, P~Yip, N~Yonemoto, JY~Lebni, MZ~Younis,
-  M~Yousefifard, HSA Yousof, C~Yu, H~Yusefzadeh, E~Zabeh, TZ~Moghadam,
-  S~Bin~Zaman, M~Zamani, H~Zandian, A~Zangeneh, TA~Zerfu, Y~Zhang, A~Ziapour,
-  S~Zodpey, CJL Murray, and SI~Hay.
-\newblock Mapping 123 million neonatal, infant and child deaths between 2000
-  and 2017.
-\newblock \emph{Nature}, 574\penalty0 (7778):\penalty0 353--358, 1992.
-\newblock ISSN 0028-0836.
-\newblock \doi{10.1038/s41586-019-1545-0}.
+\bibitem[Bruner(1960)]{bruner1960process}
+Jerome~S. Bruner.
+\newblock \emph{The Process of Education}.
+\newblock Harvard University Press, Cambridge, MA, 1960.
+\newblock ISBN 978-0-674-71001-6.

 \bibitem[Chen and Zheng(2022)]{chen2022dlsyscourse}
 Tianqi Chen and Zico Zheng.
@@ -204,15 +42,6 @@ Allan Collins, John~Seely Brown, and Susan~E. Newman.
 \newblock \doi{10.4324/9781315044408-14}.
 \newblock URL \url{https://doi.org/10.4324/9781315044408-14}.

-\bibitem[Frolli et~al.(1960)Frolli, Cerciello, Ciotola, Ricci, Esposito, and
-  Sica]{bruner1960process}
-A~Frolli, F~Cerciello, S~Ciotola, MC~Ricci, C~Esposito, and LS~Sica.
-\newblock Narrative approach and mentalization.
-\newblock \emph{Behavioral sciences (Basel, Switzerland)}, 13\penalty0 (12),
-  1960.
-\newblock ISSN 2076-328X.
-\newblock \doi{10.3390/bs13120994}.
-
 \bibitem[Heffernan()]{roberthalf2024talent}
 Margaret Heffernan.
 \newblock Earning half the minimum wage: new report reveals pressures on
@@ -310,6 +139,14 @@ Paulius Micikevicius, Sharan Narang, Jonah Alben, Gregory Diamos, Erich Elsen,
 \newblock Mixed precision training.
 \newblock URL \url{http://arxiv.org/abs/1710.03740v3}.

+\bibitem[Paas(1992)]{paas1992training}
+Fred G. W.~C. Paas.
+\newblock Training strategies for attaining transfer of problem-solving skill
+  in statistics: A cognitive-load approach.
+\newblock \emph{Journal of Educational Psychology}, 84\penalty0 (4):\penalty0
+  429--434, 1992.
+\newblock \doi{10.1037/0022-0663.84.4.429}.
+
 \bibitem[Patterson et~al.()Patterson, Gonzalez, Le, Liang, Munguia, Rothchild,
  So, Texier, and Dean]{patterson2021carbon}
 David Patterson, Joseph Gonzalez, Quoc Le, Chen Liang, Lluis-Miquel Munguia,
@@ -318,6 +155,11 @@ David Patterson, Joseph Gonzalez, Quoc Le, Chen Liang, Lluis-Miquel Munguia,
 \newblock \emph{arXiv preprint arXiv:2104.10350}.
 \newblock URL \url{http://arxiv.org/abs/2104.10350v3}.

+\bibitem[Perkins and Salomon(1992)]{perkins1992transfer}
+D.~N. Perkins and Gavriel Salomon.
+\newblock Transfer of learning.
+\newblock \emph{International Encyclopedia of Education}, 1992.
+
 \bibitem[Rashidi et~al.()Rashidi, Sridharan, Srinivasan, and
  Krishna]{chakkaravarthy2023astrasim}
 Saeed Rashidi, Srinivas Sridharan, Sudarshan Srinivasan, and Tushar Krishna.
@@ -412,6 +254,12 @@ Keller~Executive Search.
 \newblock URL
  \url{https://www.kellerexecutivesearch.com/intelligence/ai-machine-learning-talent-gap-2025/}.

+\bibitem[Sorva(2012)]{sorva2012visual}
+Juha Sorva.
+\newblock \emph{Visual program simulation in introductory programming
+  education}.
+\newblock Doctoral dissertation, Aalto University, Espoo, Finland, 2012.
+
 \bibitem[Strubell et~al.()Strubell, Ganesh, and McCallum]{strubell2019energy}
 Emma Strubell, Ananya Ganesh, and Andrew McCallum.
 \newblock Energy and policy considerations for deep learning in nlp.
--- a/paper/paper.blg
+++ b/paper/paper.blg
@@ -49,45 +49,45 @@ Warning--empty year in papert1980mindstorms
 Warning--empty year in papert1980mindstorms
 Warning--empty year in zhang2021dive
 Warning--empty year in zhang2021dive
-You've used 37 entries,
+You've used 39 entries,
            2773 wiz_defined-function locations,
-            874 strings with 23762 characters,
-and the built_in function-call counts, 50306 in all, are:
-= -- 3560
-> -- 6924
-< -- 25
-+ -- 2316
- -- 2278
-* -- 5403
-:= -- 8608
-add.period$ -- 165
-call.type$ -- 37
-change.case$ -- 869
-chr.to.int$ -- 36
-cite$ -- 120
-duplicate$ -- 762
-empty$ -- 1359
-format.name$ -- 2329
-if$ -- 9551
+            879 strings with 15131 characters,
+and the built_in function-call counts, 22733 in all, are:
+= -- 1917
+> -- 1884
+< -- 29
+ -- 638
+- -- 598
+* -- 2045
+:= -- 3643
+add.period$ -- 169
+call.type$ -- 39
+change.case$ -- 314
+chr.to.int$ -- 38
+cite$ -- 124
+duplicate$ -- 796
+empty$ -- 1410
+format.name$ -- 653
+if$ -- 4599
 int.to.chr$ -- 2
 int.to.str$ -- 1
-missing$ -- 29
-newline$ -- 242
-num.names$ -- 150
-pop$ -- 1821
+missing$ -- 31
+newline$ -- 250
+num.names$ -- 158
+pop$ -- 719
 preamble$ -- 1
-purify$ -- 833
+purify$ -- 277
 quote$ -- 0
-skip$ -- 1205
+skip$ -- 667
 stack$ -- 0
-substring$ -- 441
-swap$ -- 90
+substring$ -- 445
+swap$ -- 92
 text.length$ -- 7
 text.prefix$ -- 0
 top$ -- 0
-type$ -- 392
+type$ -- 409
 warning$ -- 46
-while$ -- 147
+while$ -- 153
 width$ -- 0
-write$ -- 557
+write$ -- 579
 (There were 46 warnings)
--- a/paper/paper.out
+++ b/paper/paper.out
@@ -32,7 +32,7 @@
 \BOOKMARK [2][-]{subsection.7.2}{\376\377\000L\000i\000m\000i\000t\000a\000t\000i\000o\000n\000s\000:\000\040\000U\000n\000d\000e\000r\000s\000t\000a\000n\000d\000i\000n\000g\000\040\000S\000c\000o\000p\000e}{section.7}% 32
 \BOOKMARK [1][-]{section.8}{\376\377\000F\000u\000t\000u\000r\000e\000\040\000W\000o\000r\000k}{}% 33
 \BOOKMARK [2][-]{subsection.8.1}{\376\377\000S\000y\000s\000t\000e\000m\000s\000\040\000E\000x\000t\000e\000n\000s\000i\000o\000n\000s\000:\000\040\000A\000n\000a\000l\000y\000t\000i\000c\000a\000l\000\040\000M\000o\000d\000e\000l\000s\000\040\000a\000n\000d\000\040\000S\000i\000m\000u\000l\000a\000t\000o\000r\000s}{section.8}% 34
-\BOOKMARK [2][-]{subsection.8.2}{\376\377\000E\000m\000p\000i\000r\000i\000c\000a\000l\000\040\000V\000a\000l\000i\000d\000a\000t\000i\000o\000n}{section.8}% 35
+\BOOKMARK [2][-]{subsection.8.2}{\376\377\000E\000m\000p\000i\000r\000i\000c\000a\000l\000\040\000V\000a\000l\000i\000d\000a\000t\000i\000o\000n\000\040\000R\000o\000a\000d\000m\000a\000p}{section.8}% 35
 \BOOKMARK [2][-]{subsection.8.3}{\376\377\000C\000u\000r\000r\000i\000c\000u\000l\000u\000m\000\040\000E\000x\000t\000e\000n\000s\000i\000o\000n\000s\000:\000\040\000F\000u\000n\000d\000a\000m\000e\000n\000t\000a\000l\000s\000\040\000v\000s\000.\000\040\000P\000r\000o\000d\000u\000c\000t\000i\000o\000n\000\040\000S\000c\000o\000p\000e}{section.8}% 36
 \BOOKMARK [2][-]{subsection.8.4}{\376\377\000C\000o\000m\000m\000u\000n\000i\000t\000y\000\040\000B\000u\000i\000l\000d\000i\000n\000g\000\040\000a\000n\000d\000\040\000A\000d\000o\000p\000t\000i\000o\000n}{section.8}% 37
 \BOOKMARK [1][-]{section.9}{\376\377\000C\000o\000n\000c\000l\000u\000s\000i\000o\000n}{}% 38
--- a/paper/paper.pdf
+++ b/paper/paper.pdf
--- a/paper/paper.tex
+++ b/paper/paper.tex
@@ -234,11 +234,13 @@ model.fit(dataloader, epochs=10)
 class Linear:
    def __init__(self, in_features,
                 out):
+        # Memory: out × in_features × 4B
        self.weight = Tensor.randn(
            out, in_features)
        self.bias = Tensor.zeros(out)

    def forward(self, x):
+        # O(batch × in × out) FLOPs
        return (x @ self.weight.T +
                self.bias)

@@ -249,6 +251,7 @@ class Adam:
        self.lr = lr
        # 2× optimizer state:
        # momentum + variance
+        # Why 2× memory vs SGD?
        self.m = [Tensor.zeros_like(p)
                  for p in params]
        self.v = [Tensor.zeros_like(p)
@@ -257,10 +260,12 @@ class Adam:
    def step(self):
        for i, p in enumerate(
                self.params):
+            # Exponential moving avg
            self.m[i] = (0.9*self.m[i] +
                        0.1*p.grad)
            self.v[i] = (0.999*self.v[i] +
                        0.001*p.grad**2)
+            # Per-parameter adaptive lr
            p.data -= (self.lr *
                self.m[i] /
                (self.v[i].sqrt() + 1e-8))
@@ -591,6 +596,7 @@ Each milestone: (1) recreates actual breakthroughs using exclusively student cod

 \noindent\textbf{Quantitative Success Criteria:}
 To ensure reproducibility and clear learning objectives, each milestone defines explicit success thresholds balancing historical plausibility (recognizing pure-Python performance limits) with pedagogical validation (ensuring implementations are functionally correct, not just syntactically complete):
+
 \begin{itemize}[leftmargin=*, itemsep=1pt, parsep=0pt]
    \item \textbf{M03 (1958 Perceptron)}: 100\% accuracy on 4-point XOR-like linearly separable problem, demonstrating basic gradient descent convergence.
    \item \textbf{M06 (1969 XOR Solution)}: 100\% accuracy on XOR classification within 20 epochs, proving multi-layer networks solve non-linear problems.
@@ -1026,6 +1032,7 @@ TinyTorch's CPU-only design prioritizes pedagogical transparency, but students b
 While TinyTorch's design is grounded in established learning theory (cognitive load~\citep{sweller1988cognitive}, progressive disclosure, cognitive apprenticeship~\citep{collins1989cognitive}), its pedagogical effectiveness requires empirical validation through controlled classroom studies. We commit to the following validation roadmap:

 \noindent\textbf{Phase 1: Pilot Deployment (Fall 2025, $n=30$--50 students)}
+
 \begin{itemize}[leftmargin=*, itemsep=1pt, parsep=0pt]
    \item \textbf{Institutions}: Deploy at 2--3 universities in introductory ML systems courses as primary hands-on framework alongside theory lectures.
    \item \textbf{Cognitive load measurement}: Paas Mental Effort Rating Scale~\citep{paas1992training} administered after Modules 05 (autograd) and 09 (CNNs) to test progressive disclosure hypothesis---does dormant feature activation reduce cognitive load compared to introducing autograd as separate framework?
@@ -1034,6 +1041,7 @@ While TinyTorch's design is grounded in established learning theory (cognitive l
 \end{itemize}

 \noindent\textbf{Phase 2: Comparative Study (Spring 2026, $n=100$--150 students)}
+
 \begin{itemize}[leftmargin=*, itemsep=1pt, parsep=0pt]
    \item \textbf{Design}: Randomized controlled trial comparing TinyTorch (systems-first, build-from-scratch) versus PyTorch-only (application-first, use-existing-frameworks) versus lecture-only (control) across 3 sections of same ML course with identical theory content.
    \item \textbf{Conceptual understanding}: ML systems concept inventory (adapted from program visualization assessment~\citep{sorva2012visual} for systems thinking) administered pre-course and post-course, measuring understanding of autograd mechanics, memory profiling, computational complexity, and optimization tradeoffs.
@@ -1042,6 +1050,7 @@ While TinyTorch's design is grounded in established learning theory (cognitive l
 \end{itemize}

 \noindent\textbf{Phase 3: Longitudinal Tracking (2026--2027, $n=200$+ students)}
+
 \begin{itemize}[leftmargin=*, itemsep=1pt, parsep=0pt]
    \item \textbf{Retention}: Re-administer concept inventory at 6 months and 12 months post-course to measure long-term retention of systems thinking---does implementation-based learning persist better than lecture-based learning?
    \item \textbf{Advanced course performance}: Track TinyTorch cohort versus control group performance in subsequent ML systems courses (e.g., CMU Deep Learning Systems~\citep{chen2022dlsyscourse}, distributed training courses) to measure preparation effectiveness.