[Rafael-SOWNet]

Part of #233.

Combinations(min, max, cellCount) in Expansion.cs enumerates the ways to share a power among the terms of a sum. Its loop ran i up to max at every depth, so it visited every prefix of every length: about 2^(n + k - 1) of them, for the C(n + k - 1, k) it returns. Squaring a sum of thirty terms (465 terms, well inside MaxExpansionTermCount) ran past a minute. The loop now stops where too few values are left above i to fill the remaining cells. A branch past that point yields nothing, so no result and no order changes.

Five of Rubi's integrals square such a sum in their working and are answered now:

Rubi	master f7f2a7a	v2.5.0	this
3.2.1 1/((a g + b g x)^2 (A + B ln(e ((a + b x)/(c + d x))^n))^2)	past 150 s	declined	3.7 s
3.2.1, the same without the n	past 150 s	past 60 s	2.5 s
4.1.3.1:357 (a + a sin(e + f x))^(3/2) (A + B sin(e + f x))/(c + d sin(e + f x))^2	past 150 s	declined	1.5 s
6.7.1:1178 sech(x)^2 tanh(x)^2/(2 + tanh(x)^3)^2	past 150 s	past 60 s	6.7 s
6.7.1:1251 coth(x)^3 csch(x) sqrt(1 + csch(x))	past 150 s	declined	9.1 s

Chain against master f7f2a7a:

• The pocket of 571 rows went from 532 to 537 solved, with 0 wrong.
• The samples are unchanged.
• Of the 8 rows that moved, run alone, master solved none and this branch solves 5. The other three are declined here; master declined one of them and ran out of time on two.
• 6,064 rows were compared for allocation, and none grew.

The full unit suite passes: 16286 passed, 0 failed.

🤖 Generated with Claude Code

https://claude.ai/code/session_012sonx8iAspMiwRwokT1Ura