Correct figures in the comments and docs added by #543

[meandmytram]

An audit of every factual claim #543 added (code comments, docstrings, the benchmark README), checked against the experiment result files and fresh measurements, found five wrong figures. This PR corrects them; no code behaviour changes.

where	said	measured
svd() comment, test_convergence.py docstring	wrong verdicts on "24 of 24 shots" at chi_max=400	21 of 22 shots with a non-trivial error (the old count mixed repeated runs of one shot with a second shot); 3 of 24 with the plain SVD on OpenBLAS
svd() comment	the pre-reduction "saved 0-7%"	at most 7%, nothing measurable on most workloads
_to_numpy docstring	the per-call cupy import cost "~18% of a decoding run"	about 11,000 failed imports per decode; surface_bitflip 4-13%, dem_d3 about a third (timed on pre-#543 main, with and without the fix)
_contract_cached docstring	a shape-keyed cache missed "7-14% of calls on large codes"	1-2% of zip-up calls on dem_d3/dem_d5, 23% on css_optimised
test_convergence.py docstring	"30-60 minutes" per chi_max=400 decode	the whole test takes about 20 minutes on an idle laptop (21.5 min measured)
benchmarks/README.md	classical_ldpc on "random (3,4) LDPC code"	three codes (seeds 11, 12, 13)

Checked and correct as written: Apple M5 (10 cores), the ~1700-site DEM chain (1677 mechanisms), the DEM workload parameters in bench_suite.py, the opt_einsum expression overhead on small tensors (42% of a zip step at chi=32, none at chi=64), the Frobenius-versus-maximum bound in the isometry gate comment, and the suite header's agreement of old and new decoders at chi=1e5 (pre- and post-#543 main agree to 1.6e-14 on 22 converged posteriors).

#543's description is corrected in place; the errors in its older comments are listed in a correction comment on #543.

🤖 Generated with Claude Code

https://claude.ai/code/session_01S6mbxc9eJDQMVx7tjvYwud

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 96.91%. Comparing base (05c7453) to head (0e73e33).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #581      +/-   ##
==========================================
+ Coverage   96.89%   96.91%   +0.01%     
==========================================
  Files          28       28              
  Lines        5509     5509              
==========================================
+ Hits         5338     5339       +1     
+ Misses        171      170       -1     

☔ 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.

[Copilot]

🟢 Approval recommended

The documentation-only corrections are consistent with the benchmark implementation and supplied measurements.

Pull request overview

Corrects inaccurate benchmark and regression figures introduced by #543 without changing runtime behavior.

Changes:

• Corrects SVD failure rates and test duration.
• Clarifies measured optimization and cache impacts.
• Documents that the LDPC benchmark uses three codes.

File summaries

File	Description
tests/decoding/test_convergence.py	Corrects failure count and runtime.
mdopt/utils/utils.py	Corrects performance and SVD measurements.
mdopt/contractor/contractor.py	Corrects cache-miss measurements.
benchmarks/README.md	Corrects the LDPC workload description.

Review details

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

---

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

[meandmytram]

Stacked into #582 together with #574; closing in favour of it. The branch is kept.