docs(benchmarks): re-execute after four releases

[michen00]

What changes, and why

The notebook is committed with its outputs because GitHub renders it from what the file holds rather than by running it — so the stored outputs are the published page. They were produced on 2026-08-21, and two later commits edited the file without re-running it, so the page had drifted from the tree it describes.

the page said	the tree says
235 tracked Markdown files, 10 in scope, 225 excluded	326 tracked, 11 in scope, 315 excluded
releases ending before v0.1.2	all twelve tags through v0.4.0
"the parent of this commit"	true again — 1e9ac89
mean 10,4xx bytes per in-scope file	118 KiB in scope, mean 10,986 bytes

Every code cell still carried an execution_count from that single run.

The run

Re-executed with the command in CONTRIBUTING.md — kernel named, timeout raised — with the release binary built beforehand and nothing else running. All three of the notebook's own checks pass in the new outputs:

• Both implementations agree. cell05: "Both implementations returned identical bytes and identical exit codes for every row above."
• Nothing changed underneath the run. No cell reports a modified file.
• No median sits far above its own minimum. The cheapest process reports 1.20 ms against a 1.33 ms median; the Rust program 2.02 against 2.11; the Python program 26.72 against 27.62. A busy machine shows up here as a failed check, and none failed.

The three charts are regenerated beside it. ruff reads the notebook natively and passes, as does check-added-large-files.

Two comments in the cells

They go with the run because editing a code cell is what makes a re-run necessary.

One comment annotated its read as "The only read in this notebook that named no encoding, which meant it decoded with whatever the platform default happened to be" — and the read immediately below it names encoding='utf-8'. So it described a past state of its own line. It now states the reason instead.

The claim was also wrong in the other direction: the Cargo.toml read in cell01 named no encoding, so it was not the only one. That read now names utf-8, which makes the comment true of every read in the notebook.

Note

CLAUDE.md's claim that no file carries a DEVIATION, blocked on comment still holds — git grep -l matches only CLAUDE.md, which describes the convention, and the notebook, which defines the marker constant. The collector prints nothing, as intended.

Corpus

The corpus is the specification, and both implementations answer to it. Tick what applies.

• This changes no behavior the corpus specifies.
• This changes what gets joined, and a case in corpus/ pins the new behavior. The case was written first and failed first.
• The change makes the tool join more than it did. The section above says what it will not eat.

Checks

• make check passes, or make test does and this touches no Rust.

make tidy — ruff over the notebook, both spell gates, check-added-large-files. The notebook's own checks are the substantive verification here, and they are quoted above. make test and the Rust suites were deliberately not run alongside the timings, per CONTRIBUTING; they ran on this tree before the notebook execution began.

[qodo-code-review]

PR Summary by Qodo

Refresh benchmark notebook and generated charts

📝 Documentation 🐞 Bug fix 🕐 10-20 Minutes

AI Description

• Refreshes benchmark outputs against current repository state and releases through v0.4.0.
• Regenerates all SVG charts from the latest Python and Rust timings.
• Makes notebook text reads explicitly UTF-8 for reproducible execution.

Diagram

graph TD
  R["Repository state"] --> N["Benchmark notebook"] --> P["Python tool"] --> C["Result checks"] --> O["Stored outputs"] --> S["SVG charts"]
  N --> B["Rust binary"] --> C

Loading

High-Level Assessment

Re-executing the authoritative notebook and committing its outputs is the appropriate approach because GitHub renders stored notebook state. Manually editing measurements or charts risks inconsistency, while automated CI benchmarking would introduce uncontrolled runner variance and is unnecessary for this documentation refresh.

Files changed (4) +1028 / -1038

Documentation (4) +1028 / -1038

benchmarks-bytes.svgRegenerate file-size performance chart +82/-71

Regenerate file-size performance chart

• Regenerates the chart with current per-file timings, the 10,986-byte repository average, and the revised 13.3-times-or-more comparison.

docs/benchmarks-bytes.svg

benchmarks-startup.svgRegenerate startup-cost chart +214/-229

Regenerate startup-cost chart

• Updates startup and one-file timing bars from the new benchmark run, including the 1.20 ms process floor and current Python and Rust measurements.

docs/benchmarks-startup.svg

benchmarks.ipynbRe-execute benchmark notebook with current repository data +97/-89

Re-execute benchmark notebook with current repository data

• Refreshes execution metadata, repository statistics, release history, timing tables, checks, and conclusions through v0.4.0. It also makes the Cargo manifest read explicitly UTF-8 and clarifies the notebook's encoding rationale.

docs/benchmarks.ipynb

benchmarks.svgRegenerate benchmark scaling charts +635/-649

Regenerate benchmark scaling charts

• Refreshes the Python and Rust file-count scaling curves and relative-performance plot using the latest benchmark measurements.

docs/benchmarks.svg

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 87.37%. Comparing base (1e9ac89) to head (fae4ea4).

Additional details and impacted files

@@           Coverage Diff           @@
##             main      #49   +/-   ##
=======================================
  Coverage   87.37%   87.37%           
=======================================
  Files           3        3           
  Lines         689      689           
=======================================
  Hits          602      602           
  Misses         87       87           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[qodo-code-review]

Code Review by Qodo

🐞 Bugs (0) 📘 Rule violations (0) 📎 Requirement gaps (0)

Great, no issues found!

Qodo reviewed your code and found no material issues that require review

Tip of the day

💡 Did you know, you can add REVIEW.md to your repo root and Qodo follows it on every PR

More tips ↗ | Customize Qodo ↗ | Qodo docs ↗

[Copilot]

🟢 Approval recommended

Changes are confined to re-executed documentation artifacts and a small, correctness-improving encoding tweak with no impact on tool behavior.

Pull request overview

This PR refreshes the committed benchmark artifacts so the rendered documentation matches the repository state after recent releases and notebook edits (the notebook outputs are the published page).

Changes:

• Re-executes docs/benchmarks.ipynb, updating stored outputs/metrics to reflect current tags and repository file counts.
• Regenerates the benchmark charts (docs/benchmarks*.svg) to match the updated run results.
• Tightens notebook reproducibility by explicitly reading Cargo.toml as UTF-8 (consistent with other reads in the notebook).

File summaries

File	Description
docs/benchmarks.ipynb	Updated execution metadata, outputs, and a small code/comment tweak to keep the notebook’s narrative and reads consistent.
docs/benchmarks.svg	Regenerated primary benchmark chart to match the latest notebook run.
docs/benchmarks-startup.svg	Regenerated startup-cost chart based on the latest timings.
docs/benchmarks-bytes.svg	Regenerated per-byte scaling chart based on the latest measurements.

Review details

• Files reviewed: 1/4 changed files
• Comments generated: 0
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.