Skip to content

Li — Shareable Plots & X Assets

For agentic workers: Ship publication-quality PNGs from CI and local ./scripts/plot_shareables.sh. Every benchmark run and test sweep must be plottable without hand-editing.

Goal: Top-notch, on-brand figures for X, blog posts, and README — generated from latest.csv and li-tests manifest results.

Depends on: benchmarks/harness/bench.py, li-tests/run_all.sh
Blocks: Phase 5b exit gate (publishable benchmark story)

Design standards (X / social)

Rule Value
Aspect ratio 16:9 (1600×900 px @ 1×; export 3200×1800 for retina)
Background Dark #0d1117 (GitHub-style) — reads well in X dark mode
Accent #58a6ff (Li primary), #3fb950 pass, #f85149 fail
Typography Sans: system / Inter fallback; mono for numbers
Branding Title: Li (理); subtitle: benchmark or suite name; footer: li-language · git sha
No chartjunk No 3D, no pie charts for perf; bar/line/heatmap only
Accessibility Color + hatch/pattern; label bars directly when <8 series
Formats benchmarks/results/share/*.png + optional *.svg for blog

Outputs (required)

Asset Source Filename pattern
Cross-lang speed bar results/latest.csv share/bench_speed_{tier}.png
Li vs C++ speedup same share/speedup_vs_cpp.png
Correctness badge grid verify results share/correctness_tier0.png
Test suite matrix li-tests manifest run share/test_suite_matrix.png
Suite pass rate bars same share/test_suite_pass_rate.png
CI summary card combined JSON share/ci_summary_card.png

Repository layout

benchmarks/
  harness/
    bench.py           # run benchmarks → latest.csv
    plot.py            # CSV → share PNGs
    plot_theme.py      # Li dark theme, fonts, sizes
    requirements.txt   # matplotlib, pandas, pillow
  results/
    latest.csv
    share/             # generated PNGs (gitignored except .gitkeep)
li-tests/
  harness/
    plot_suites.py     # run_all → suite plots
scripts/
  plot_shareables.sh   # one command: tests + benches → share/

CSV → plot contract

plot.py reads columns: benchmark, lang, variant, threads, metric, value, unit, git_sha, cpu_model, flags

Charts: 1. Grouped bar — x=benchmark, hue=lang, y=value (log scale for time metrics) 2. Speedup — li / cpp per benchmark, horizontal bars, reference line at 1.0× 3. Scaling — line chart threads vs throughput for OpenMP benches

li-tests → plot contract

plot_suites.py: 1. Parses manifest.toml, runs each suite via run_all.sh 2. Builds matrix: rows=suites, cols=outcome types, cell=count pass/fail 3. Emits pass-rate bar chart and heatmap

Phase alignment

When Plots
Now (5b skeleton) Test suite plots from live lic; sample bench CSV + demo speed chart
Phase 3 codegen Real Tier 1 micro bars
Phase 5b Tier 2 Physics speed + correctness grid for X thread
Post v1 Animated GIF optional (out of scope v1)

Exit gate (Phase 5b plotting slice)

  • ./scripts/plot_shareables.sh exits 0 and writes ≥4 PNGs to benchmarks/results/share/
  • Plots meet 16:9 + dark theme spec above
  • README or docs link to results/share/ for social assets
  • bench.py --tier 0 feeds correctness plot when Tier 0 data exists

X posting checklist

  1. Run ./scripts/plot_shareables.sh on reference machine
  2. Attach share/bench_speed_tier2.png or share/test_suite_pass_rate.png
  3. Alt text: describe metric, langs, machine, N=…
  4. Tweet copy: lead with proved (Lean) + number; link repo