[Rafael-SOWNet]

Two commits.

Two fractions are added over their least common denominator. The simplifier added two fractions by cross-multiplying, so a factor both denominators share was squared, and each later sum multiplied the denominators again before anything cancelled. (1 - tanh(x)^2) / tanh(x) written through exponentials spent nearly all of its two seconds in two sums whose denominators had grown to (e^(2x) + 1)^6 (e^(2x) - 1)^3. Where the denominators share a factor, the sum now goes over their least common multiple. Denominators that share nothing are cross-multiplied as before.

A limit reads the sign of a vanishing base on the side approached, and at an infinity. The first commit writes (sqrt(x) x)^(x - 2 - x^2) at 0+, through x = 1/x, as (1/x)^(2 ((x - 1)/x^2 - 2)) at +oo, which the descent read as 0^(-4) and left. A base tending to 0, to an exponent tending to a negative number, tends to +oo where the base is positive on the way in; the descent knew it only for a base never negative anywhere. The sign near an infinity is now read from the vanishing order after putting 1/x for x, for a product or quotient from its parts, and for a positive base to a real power as positive. It is not read from a slope at an infinity: after the substitution, sinh(e^x) at -oo has a flat function's slope, and asking for it made sgn(sinh(e^x)) take 24 s to decline, against 0.13 s on master. It now takes 0.17 s.

	master	branch
Simplify of (1 - tanh(x)^2) / tanh(x) through exponentials	1.75 s	0.15 s, the same answer
1/(x+1) + 1/(x+1)^2, simplified	left as the sum	(2 + x) / (1 + x) ^ 2
(1/x)^(2 ((x - 1)/x^2 - 2)) at +oo	unevaluated	+oo
sin(x)^(x - 2) at 0+	unevaluated	+oo
sgn(1/x) at +oo, at -oo	unevaluated	1, -1
sgn(e^x) at -oo	unevaluated	1
sinh(sgn(e^x)) at -oo	unevaluated, 0.7 s	(e - 1/e) / 2, 0.17 s

AVanishingBaseToANegativeWholePowerTest pinned sin(x)^(x - 2) at 0+ as left alone. It is +oo: the base is positive there, so the power is e^((x - 2) ln sin x). The test only pinned it because nothing answered it. What it was for, that an exponent merely tending to a whole number is not given that whole power's sign, is kept on the left, where the power is complex and the limit is still left alone.

limitcheck A/B against master (DEPTH=3, the usual seeds): 129 answers gained, none lost. 81 are at an infinity, mostly a sign. Seven rows the harness lists as another value are right:

• sinh(sgn(e^x)), sgn(sgn(e^x)), sgn(e^x)^3 and 2 sgn(e^x) at -oo are functions of 1, where the harness's samples of e^x underflow to 0.
• sgn(arctan(1/2)^(2/x)) at 0+ is 1, for the same reason.
• sin(pi x)^(erf(pi) - (-1)^x) at 0+ and 2+ is +oo. Its exponent tends to erf(pi) - 1, about -8.9e-6, which is negative. mpmath gives 1.02 at x = 10^-1000 and 7.5e8 at 10^-1000000, where the harness's samples stop at 10^-8.

floor(tanh(x)/x) at 0 moved from a timeout to a decline. floor(sqrt(x^3)) at 0+ declines in 6–7 s on both builds; its one timeout in the first A/B was jitter.

Each changed answer is in BREAKING-CHANGES.md, measured on 2.6.0. The full suite (19,113), the wrapper suites, AOT and the ten harnesses pass.

Found by limitcheck at DEPTH=3, SEED=77, after the slow Simplify was found sampling a slow limit. No issue was filed.

🤖 Generated with Claude Code

[codecov]

Codecov Report

❌ Patch coverage is 97.36842% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 86.80%. Comparing base (a1c0034) to head (69f5bff).

Files with missing lines	Patch %	Lines
...ath/Functions/Continuous/Limits/Transformations.cs	93.75%	0 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #2386      +/-   ##
==========================================
+ Coverage   86.79%   86.80%   +0.01%     
==========================================
  Files         312      312              
  Lines       69156    69190      +34     
  Branches    20166    20177      +11     
==========================================
+ Hits        60022    60060      +38     
+ Misses       4918     4916       -2     
+ Partials     4216     4214       -2     

Flag	Coverage Δ
cpp	37.77% <ø> (ø)
csharp	87.34% <97.36%> (+0.01%)	⬆️
fsharp	32.97% <ø> (ø)
interactive	40.72% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
...nctions/Continuous/Limits/Solvers/Limit.Classes.cs	74.93% <100.00%> (ø)
...tions/Simplification/Patterns/Patterns.Rational.cs	99.12% <100.00%> (+0.01%)	⬆️
Sources/AngouriMath/Functions/SingleQuotient.cs	92.92% <100.00%> (-4.64%)	⬇️
...ath/Functions/Continuous/Limits/Transformations.cs	88.78% <93.75%> (-0.05%)	⬇️

... and 8 files with indirect coverage changes

🚀 New features to boost your workflow:

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