docs(benchmarks): the published benchmarks are re-measured at the new model sizes, five times each tenfold step

[github-actions]

The published benchmark run on fbc735a, which is main at #2032 plus #2038's workflow fix. The comment on #2032 compares these numbers with the committed ones; the docs preview of this pull request shows the page.

Prompt: I want the benchmark to get main up to date first

Note

The following content was generated by AI.

This publishes #2032's new sizes, measured on the box, to main. It also carries #2038's two commits, because the run was dispatched from that branch: on main, published-benchmark.yml does not parse. Merging this makes #2038 redundant.

Merge this with a merge commit, not a squash. bench/reproduce.py now pins fbc735a5b, the commit that was measured. A merge commit puts fbc735a into main's history. After a squash it would be reachable only through claude/specsolve-2032-review-7hrcv0.

What is in the diff, and what was checked

• bench/results/*.json, docs/about/benchmarks.json, docs/about/benchmarks.md: from the run's artifact, as the workflow commits them. With main merged in (docs(benchmarks): the benchmark charts share one pair of axes across every model, and a box dragged on any chart zooms all of them #2040, docs(benchmarks): the benchmarks page opens on its charts, counts sizes as the columns and rows that reach the solver, and tables what each model declares #2042), pixi run -e bench report on these results writes the page's models and fresh tables without a change.
• The linopy memory sentence under the model table (docs(benchmarks): the benchmarks page opens on its charts, counts sizes as the columns and rows that reach the solver, and tables what each model declares #2042) quotes this run: network 2.7× at 1M columns, 1.75 GB against 0.65 GB. On HiGHS from 100k columns up, every other model stays at or under 1.26×, at every size including the new 5× ones.
• bench/reproduce.py, bench/reproduce.py.lock: specsolve moves from af1dac7d3 to fbc735a5b, through uv lock --script bench/reproduce.py, as docs(benchmarks): the published benchmarks are re-measured with the network model's pooled plants #2031 did. No other package moved. Without this, test_the_lock_installs_what_the_published_numbers_were_taken_on fails.
• .github/workflows/published-benchmark.yml and tests/test_architecture.py: test(ci): the published-benchmark workflow parses again, and the suite refuses a workflow that does not #2038's fix and its test, unchanged.
• CHANGELOG.md: this PR's line, as docs(benchmarks): the published benchmarks are re-measured with the network and sparse models, down to a few hundred variables #2005, docs(benchmarks): the published benchmarks are re-measured with the sector-coupled model #2024 and docs(benchmarks): the published benchmarks are re-measured with the network model's pooled plants #2031 added theirs.

Run on a cloud container at afff79b: pixi run check (lint, format, typecheck; 5613 passed, 566 skipped, 4 xfailed), pixi run test-bench (475 passed, 15 skipped), pixi run -e docs docs-build (strict, no issues), tests/test_docs_site.py and tests/test_doc_examples.py (123 passed, 4 skipped).

Read from the run's artifacts. All 8 cases finished. No cell was killed by the memory watchdog, and no library stops at a smaller size than in #2031. pyomo on network reaches s5, one size further. linopy and gurobipy-loop on network still stop at m, by the 16 GB and 30 s budgets. The matrix floor's row leaves the "launch to exit" table, because #2034 made test_fresh skip both matrix arms.

🤖 Generated with Claude Code

https://claude.ai/code/session_01NTByC5UJJWvcFj3mvmv83v

[read-the-docs-community]

Documentation build overview

📚 specsolve | 🛠️ Build #35146496 | 📁 Comparing afff79b against latest (2588883)

  🔍 Preview build  

2 files changed

± about/benchmarks/index.html
± about/changelog/index.html

[codspeed]

Merging this PR will not alter performance

✅ 104 untouched benchmarks
⏩ 307 skipped benchmarks¹

---

_{Comparing bench-results/pr-2032 (afff79b) with main (2588883)²}

Footnotes

1. 307 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

2. No successful run was found on main (48c9414) during the generation of this report, so 2588883 was used instead as the comparison base. There might be some changes unrelated to this pull request in this report. ↩