[Rafael-SOWNet]

A two-sided limit at a finite point compares the limits from each side. When a side came back NaN, the answer was NaN, which means "this does not exist". But the descent also returns NaN internally for an indeterminate form it could not resolve. For x * floor(2/x) at 0 that form is 0 * floor(-oo), and each side is 2. That NaN means "undecided", and turning it into a two-sided NaN gave a wrong answer where the library had decided nothing.

When the sides are asked outright, a side that stops at an indeterminate form with an infinity in it now counts as undecided, and the two-sided limit stays unevaluated. NaN is still the answer when:

• a side settles as NaN;
• the indeterminate form has only finite parts, such as 0/0 for (x - x)/(x - x) or ln(-2)/0 for ln(-2)/ln(x) at 1. These have no value anywhere near the point.

Input, at 0, two-sided	true value	2.6.0	master	this branch
x * floor(2/x)	2	NaN	NaN	unevaluated
floor(1/x) * sin(x)	1	NaN	NaN	unevaluated
x^2 * floor(1/x^2)	1	0	NaN	unevaluated
x * ceil(1/x)	1	NaN	NaN	unevaluated
floor(2/x)	none	NaN	NaN	NaN
floor(x)	none	unevaluated	NaN	NaN
sin(1/x)	none	NaN	NaN	NaN
(x-x)*sin(x)/((x-x)*x)	none	NaN	NaN	NaN
(abs(x)^(x/pi))^sgn(x)	1	unevaluated	1	1

The first four rows are in BREAKING-CHANGES.md. Answering them with their values needs the floor's 0 * oo resolved from each side, which is a separate change; this one stops the wrong answer. The new test checks that the four rows stay limit(...) nodes, and that the rows without a limit stay NaN.

Checked on this branch:

• Tests: UnitTests 19330 passed, plus the MCP, Terminal, InteractiveWrapper and F# wrapper suites, and the AOT smoke test.
• Harnesses: BoundCheck, RootCheck, CasBench, PropCheck, SimpSweep, RuleCheck, CanonCheck and Confluence all pass.
• limitcheck at DEPTH=3, COUNT=600, on 42 seeds against master: no answer changed. On seed 45, tanh(log(x, -1) - tan(x)) at 1 from the right moved into "did not finish in 5 s". It takes 6.2 to 6.6 s on master and 6.3 to 7.1 s on this branch, and is unevaluated on both, so the 5 s cap decides which list it lands in.

🤖 Generated with Claude Code

[codecov]

Codecov Report

❌ Patch coverage is 94.11765% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 86.81%. Comparing base (15070c6) to head (20cffb8).

Files with missing lines	Patch %	Lines
...ns/Continuous/Limits/Solvers/Solvers.Definition.cs	94.11%	0 Missing and 1 partial ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #2419   +/-   ##
=======================================
  Coverage   86.81%   86.81%           
=======================================
  Files         312      312           
  Lines       69514    69523    +9     
  Branches    20300    20305    +5     
=======================================
+ Hits        60348    60356    +8     
- Misses       4932     4933    +1     
  Partials     4234     4234           

Flag	Coverage Δ
cpp	37.77% <ø> (ø)
csharp	87.35% <94.11%> (+<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.

Files with missing lines	Coverage Δ
...ns/Continuous/Limits/Solvers/Solvers.Definition.cs	92.28% <94.11%> (+0.58%)	⬆️

... and 7 files with indirect coverage changes

🚀 New features to boost your workflow:

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