[Rafael-SOWNet]

Part of #746.

The release checklist's performance pair for 2.7.0: v2.6.0 (1a72d4da, the 2664th commit) beside 842e8e95 (the 2943rd), the last change to the kernel before the release. #2440, after it, touches no kernel file. The period is 279 commits.

Both columns come from one CI runner. The Kernel Benchmark Across Key Commits workflow ran on this branch as run 43536393809, with the list v2.6.0, 2943rd, v2.6.0, 2943rd. Each commit was measured twice, interleaved, so the runner's spread on an unchanged commit is measured in the same run as the change. The list was a temporary commit on this branch and is restored, so the diff is documentation only. The 2943rd column added to the two history tables at the top of the file comes from the Kernel Benchmark's run 43534063686 on the same commit. Every table in the file has a consistent column count.

• Controls, on allocation:
  • each commit's two measurements agree to within 0.03% on every gated row;
  • the v2.6.0 kernel, the same as Measure the 2.6.0 performance pair at the last kernel change, the 2662nd #2060's 2662nd, reproduces that measurement to within 0.13%;
  • the 2943rd matches the Kernel Benchmark run to within 0.10%.
• Down: SolveHard −4.72% and SolveMediumHard −3.51%, from A union of solves solves each equation once #2310.
• Up:
  • SimplifyHard +2.42%, in three steps at A squared sine and cosine with one coefficient are found anywhere in a sum #2175, One and a squared tangent are found anywhere in a sum #2176 and The power reduction of a squared cosine is proved, as the squared sine's is #2177 (+0.88%, +1.05%, +0.45%), each inside the gate's 3%.
  • Derivate +1.85%, in one step at Keep an irrational exponent exact in the power rule #2393. With 2.6.0's +5.65%, it is up 7.5% since v2.5.0 without the gate failing.
  • CompileEasy +33% and CompileHard +42%, which are not gated, at the commits that taught the compilers new nodes: Membership and divisibility compile #2223, A statement where a number is expected is refused by both compilers #2229, tanh and coth of a large number are their limits #2305, Compile a matrix argument as an array, with its inverse and whole powers #2429 and Compile a function of a matrix argument as the function of each entry, and vectors as arrays #2435.
• Timings are read only where both interleaved pairs agree and sit outside the same-commit spread, which reached 9.8% in this run.
  • CompileEasy +24.4% moved with its allocation.
  • CompileHard −12.2%, RunHard −12.0% and RunMedium −4.3% moved without it. Each reproduced in both pairs, and none is attributed.
  • SimplifyEasy +5.5% is a timing moving alone, recorded as a question rather than a regression.

The steps are located from the gate's output on each push to master. Allocation is deterministic, so the run where a row first moves names the commit. What within each commit costs the bytes has not been bisected.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 86.92%. Comparing base (7c5e04b) to head (59b07fe).

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #2441      +/-   ##
==========================================
- Coverage   86.92%   86.92%   -0.01%     
==========================================
  Files         315      315              
  Lines       70509    70509              
  Branches    20606    20606              
==========================================
- Hits        61293    61287       -6     
- Misses       4944     4951       +7     
+ Partials     4272     4271       -1     

Flag	Coverage Δ
cpp	37.77% <ø> (ø)
csharp	87.45% <ø> (-0.01%)	⬇️
fsharp	32.97% <ø> (ø)
interactive	39.81% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.
see 5 files with indirect coverage changes

🚀 New features to boost your workflow:

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