Fix broken paper references to deleted files and paths

- Replace PROPER_BASELINE_COMMANDS.sh → EXPERIMENTS_REFERENCE.sh
- Update experiment count from 135 → 153 (includes Phase 6 TTQ)
- Fix all paper/tmlr/ paths → paper/ (flattened directory structure)
- Update build instructions: cd paper && make (not cd paper/tmlr)
- Fix references in methods.tex, appendix, and results.tex
- Recompile paper successfully (28 pages, 552 KB)

Author: dariocazzani

paper/main.pdf

@@ -24,14 +24,14 @@ \subsection{Reproducible Research Pipeline}
 
     \item \textbf{Aggregation:} \texttt{analysis/aggregate\_results.py} recursively scans \texttt{results/raw/} and \texttt{results/raw\_kd/} directories, loading all \texttt{results.json} files into a unified DataFrame saved as \texttt{results/processed/aggregated.csv}. No filtering or manual selection—all completed experiments included.
 
-    \item \textbf{Table generation:} \texttt{analysis/generate\_tables.py} reads the CSV and produces LaTeX tables (\texttt{paper/tmlr/tables/*.tex}) with:
+    \item \textbf{Table generation:} \texttt{analysis/generate\_tables.py} reads the CSV and produces LaTeX tables (\texttt{paper/tables/*.tex}) with:
     \begin{itemize}
         \item Mean ± standard deviation across seeds
         \item Statistical tests (t-test, Cohen's d)
         \item Gap calculations (FP32 - BitNet, FP32+KD - BitNet+Recipe)
     \end{itemize}
 
-    \item \textbf{Figure generation:} \texttt{analysis/generate\_figures.py} creates publication-quality matplotlib figures (\texttt{paper/tmlr/figures/*.pdf}) directly from CSV data.
+    \item \textbf{Figure generation:} \texttt{analysis/generate\_figures.py} creates publication-quality matplotlib figures (\texttt{paper/figures/*.pdf}) directly from CSV data.
 
     \item \textbf{Paper integration:} LaTeX \texttt{\\input\{tables/*.tex\}} and \texttt{\\includegraphics\{figures/*.png\}} commands directly reference generated artifacts. No copy-paste or manual number entry.
 \end{enumerate}
@@ -258,10 +258,10 @@ \subsection{Code and Data Availability}
     \item \textbf{Training code:} \texttt{experiments/train.py}, \texttt{experiments/train\_kd.py}
     \item \textbf{Model implementations:} \texttt{bitnet/}, \texttt{experiments/models/}
     \item \textbf{Analysis scripts:} \texttt{analysis/*.py}
-    \item \textbf{Experiment documentation:} \texttt{PROPER\_BASELINE\_COMMANDS.sh} (all 135 commands)
+    \item \textbf{Experiment documentation:} \texttt{EXPERIMENTS\_REFERENCE.sh} (all 153 commands)
     \item \textbf{Raw results:} \texttt{results/raw/}, \texttt{results/raw\_kd/} (JSON files)
     \item \textbf{Processed data:} \texttt{results/processed/aggregated.csv}
-    \item \textbf{Generated artifacts:} \texttt{paper/tmlr/tables/*.tex}, \texttt{paper/tmlr/figures/*.pdf}
+    \item \textbf{Generated artifacts:} \texttt{paper/tables/*.tex}, \texttt{paper/figures/*.pdf}
 \end{itemize}
 
 Dependencies managed with \texttt{uv} (Python package manager), locked versions in \texttt{uv.lock} for exact reproducibility.

paper/sections/03_methods.tex

@@ -1,7 +1,7 @@
 % Section 5: Results
-% Generated from 133 completed experiments (results/processed/aggregated.csv)
-% Tables: paper/tmlr/tables/*.tex
-% Figures: paper/tmlr/figures/*.png
+% Generated from 153 completed experiments (results/processed/aggregated.csv)
+% Tables: paper/tables/*.tex
+% Figures: paper/figures/*.png
 
 \section{Results}
 \label{sec:results}

paper/sections/05_results.tex

@@ -23,7 +23,7 @@ \subsection{Repository Structure}
     aggregate_results.py      # CSV aggregation
     generate_tables.py        # LaTeX tables
     generate_paper_figures.py # Publication figures
-  paper/tmlr/          # Paper source
+  paper/                # Paper source
     main.tex           # Main document
     sections/          # Modular sections
     tables/            # Auto-generated LaTeX tables
@@ -61,12 +61,12 @@ \subsection{Complete Pipeline Execution}
 uv run python -m analysis.generate_tables
 uv run python -m analysis.generate_paper_figures
 uv run python -m analysis.generate_model_size_table
-# Output: paper/tmlr/tables/*.tex, paper/tmlr/figures/*.pdf
+# Output: paper/tables/*.tex, paper/figures/*.pdf
 \end{verbatim}
 
 \textbf{Step 4: Compile paper}
 \begin{verbatim}
-cd paper/tmlr && make
+cd paper && make
 # Output: main.pdf (this document)
 \end{verbatim}
 
@@ -82,7 +82,7 @@ \subsection{Verification Without Retraining}
 uv run python -m analysis.aggregate_results
 uv run python -m analysis.generate_tables
 uv run python -m analysis.generate_paper_figures
-cd paper/tmlr && make
+cd paper && make
 \end{verbatim}
 
 \textbf{Option 2: Spot-check individual experiments}
@@ -208,11 +208,11 @@ \subsection{Reproducibility Checklist}
 \begin{itemize}
     \item[$\boxtimes$] \textbf{Code:} Complete source code provided
     \item[$\boxtimes$] \textbf{Data:} Public datasets with download links
-    \item[$\boxtimes$] \textbf{Model checkpoints:} All 135 experiment checkpoints archived
+    \item[$\boxtimes$] \textbf{Model checkpoints:} All 153 experiment checkpoints archived
     \item[$\boxtimes$] \textbf{Hyperparameters:} Documented in \texttt{config.json} per experiment
     \item[$\boxtimes$] \textbf{Seeds:} Fixed seeds (42, 123, 456) for reproducibility
     \item[$\boxtimes$] \textbf{Dependencies:} Exact versions in \texttt{uv.lock}
-    \item[$\boxtimes$] \textbf{Commands:} Complete experiment commands in \texttt{PROPER\_BASELINE\_COMMANDS.sh}
+    \item[$\boxtimes$] \textbf{Commands:} Complete experiment commands in \texttt{EXPERIMENTS\_REFERENCE.sh}
     \item[$\boxtimes$] \textbf{Random state:} Deterministic training (same seed = identical results)
     \item[$\boxtimes$] \textbf{Compute:} Hardware specs and total GPU-hours reported
     \item[$\boxtimes$] \textbf{Pipeline:} Automated artifact generation (no manual transcription)