[flaviens]

Hi! Please find generated PR text below. Thanks!

The camera-ready now prints all five evaluation figures at body-text size and fits Table VIII on the page (microtaint-paper be1c396 on rebuttal). This PR brings the generators behind them in line.

• plot_figures.py draws each figure at the 3.5 in column width with 10 pt STIX text, which matches the Times body. The paper includes them at natural size, which prints them 1:1. They used to be drawn wider and shrunk, which printed their ticks at about 5 pt.
• Fig. 3's group labels sit clear of the frame and read "Bit-precise" and "Register-level". Fig. 5's value labels sit a fixed 2 pt above their bars.
• gen_ladder_table.py drops the CPU column from the LaTeX table, which ran 23.7 pt past the page. CPU time stays within a few percent of wall time on each emulated rung. The Markdown table keeps it.
• The reference figures and ladder table under reference-runs/ are regenerated with the README's commands. The four paper-corpus figures are now byte-identical to the paper's Figures 2 to 5.

Checks: on the paper-corpus run, the old scripts reproduce the paper's previous figures and table exactly. Each new table row equals its old row minus the CPU cell. ruff check passes on both scripts.

[Copilot]

🔵 Needs a closer look

Correctness here hinges on regenerated camera-ready figure rendering that cannot be visually verified in this environment, so final human review is warranted.

2 open issues

• Update figure-width comment to cover all five figures · New
• Remove dead size constants and stale docstrings · New

What changed in this PR

This PR updates the NDSS'27 artifact generators so their output matches the camera-ready paper. plot_figures.py now draws all five evaluation figures at the paper's 3.5 in column width with 10 pt STIX text (included at natural size, so they print 1:1 instead of being drawn wide and shrunk), and gen_ladder_table.py drops the CPU column from the LaTeX overhead ladder (which overran the page) while keeping it in the Markdown version. The regenerated reference run artifacts and READMEs are updated to match.

Changes:

• Draw every figure at a fixed COLUMN_WIDTH_IN with PRINT_SIZE/STIX text and layout='constrained', replacing the old per-figure shrink-to-fit sizing; fix Fig. 3 group labels and Fig. 5 value-label placement.
• Remove the CPU column from the LaTeX ladder table (8 columns) while retaining it in Markdown; update caption and docstrings accordingly.
• Regenerate the reference ladder_table.tex and refresh both READMEs.

File	Description
artifacts/​ndss27/​rq2-comparison/​plot_figures.py	Draws all figures at print size; introduces COLUMN_WIDTH_IN/PRINT_SIZE/PRINT_RC; leaves old size constants and slot_geometry's factor unused.
artifacts/​ndss27/​rq5-overhead/​gen_ladder_table.py	Drops CPU from the LaTeX table (tabular, header, multicolumn, body all reduced to 8 columns); keeps CPU in Markdown.
artifacts/​ndss27/​reference-runs/​run-all-debian12/​20260918-121340/​ladder_table.tex	Regenerated reference LaTeX table with the CPU column removed.
artifacts/​ndss27/​reference-runs/​README.md	States the four figures are the paper's Figures 2–5.
artifacts/​ndss27/​README.md	Documents that only the Markdown ladder includes CPU time.

Notes: the LaTeX/Markdown column counts are internally consistent (8 vs 10 columns), the reference .tex rows equal the old rows minus the CPU cell, the reference .md still carries CPU, and mpl.ticker resolves because matplotlib.pyplot is imported. The findings I filed are confined to leftover dead code (AXIS_LABEL_SIZE/LEGEND_SIZE/VALUE_LABEL_SIZE, slot_geometry's discarded factor) and a stale COLUMN_WIDTH_IN comment.

_{Review effort: Balanced}