[Rafael-SOWNet]

#1961 asks a hyperbolic integrand as a trigonometric one in s. In the first pass, the exponential substitution declines these when a sum over roots would be needed. The question in s then took 2.5–7 s and declined too, and the second pass answered in e^x exactly as before #1961. 1/(1 + sinh(x)^3) allocated 4.98 GB where it had allocated 0.21 GB, and gave the same answer.

The rule now declines in the first pass, with rationalToo set and a sum over roots wanted, when the question in s is rational in cos(s) or sin(s) alone and either:

• its denominator holds odd powers of the sine at degree 3 or more, or
• it has an irreducible factor over Q of degree past 2, in y, or in y^2 where only even powers stand. That factor survives t = tan(s/2) or tan(s).

A guard on the flag alone would also have taken back the seven rows where #1961 wrote a closed form in place of a sum over roots: 6.1.7:305, 314 and 318; 6.2.7:101, 102 and 107; and 6.7.1:1173. The factor test leaves all seven as they are on master, character for character.

Each problem alone, before #1961 → master → this branch (allocation, answer length):

Problem	Before #1961	Master	Branch
6.1.7:211 1/(1+sinh(x)^3)	0.21 GB, 153	4.98 GB, 153	0.21 GB, 153
6.1.7:212 1/(1-sinh(x)^3)	0.21, 162	4.98, 162	0.21, 162
6.1.7:313 1/(1+sinh(x)^5)	0.86, 217	1.88, 217	0.86, 217
6.1.7:315 1/(1+sinh(x)^8)	0.84, 275	2.27, 275	0.84, 275
6.1.7:316 1/(1-sinh(x)^5)	0.87, 226	1.89, 226	0.87, 226
6.2.7:117 1/(1+cosh(x)^5)	17.6, 389	21.0, 389	17.6, 389
6.2.7:119 1/(1+cosh(x)^8)	0.75, 253	2.58, 253	0.76, 253
6.2.7:120 1/(1-cosh(x)^5)	17.2, 359	22.7, 359	17.3, 359

The answers are identical in all three columns.

Measured on 8270086f, branch against master:

Run	Problems	Master	Branch
Pocket: all of Rubi's family 6	3,690	3,631 solved	3,632 solved
Family 1, 40 a file	1,381	1,368	1,368
Family 0, whole	1,814	1,789	1,789
Families 2 to 8, sampled	2,410	2,380	2,380

Re-run alone, the three moves, 6.1.1:123, 6.7.1:929 and 7.4.1:369, come out the same on both builds: they are problems that take 15–25 s against a 5 s budget. #1961's 30 gained rows answer as on master. 6.1.5:94 reads wrong on both builds; that is the evaluator artefact of #1944. No answer changes, so BREAKING-CHANGES.md has no entry.

AHyperbolicIntegrandAsATrigonometricOneIntegralTest pins both sides. 1/(1+sinh(x)^4), 1/(1+cosh(x)^3) and sech(x)^2/(3-4*tanh(x)^3) keep the closed form, with no sum(. 1/(1+sinh(x)^3) and 1/(1+cosh(x)^8) are answered by the second pass, with sum(. Each is differentiated back at six points, negative x among them.

Part of #233.

🤖 Generated with Claude Code

https://claude.ai/code/session_012sonx8iAspMiwRwokT1Ura