is_equiv rates different equations as equivalent: _strip_string drops any left-hand side of two characters or fewer #35

_strip_string in modeling/math_equivalence.py has this rule, commented as "get rid of e.g. "k = " or "q = " at beginning":

if len(string.split("=")) == 2:
    if len(string.split("=")[0]) <= 2:
        string = string.split("=")[1]

It applies to every equation with a short left side, not just stray prefixes. x=7 and y=7 both become 7, so is_equiv treats them as the same answer. For problems whose answer is an equation, the right-hand side alone decides the grade.

A real case is test/precalculus/323: the problem asks for the image of the line y = 2x + 1 in the form "y = mx + b", and the gold answer is y = 2x + 3. A completion ending in \boxed{x=2x+3} (a different equation) is graded correct. I checked that with a one-problem run using the public problem and gold text; that run gave a score of 1/1 where 0/1 is expected.

Reproduction

Standard library only. Run with SRC set to a checkout of the repo.

import os, sys, importlib.util
p = os.path.join(os.environ["SRC"], "modeling", "math_equivalence.py")
s = importlib.util.spec_from_file_location("me", p)
m = importlib.util.module_from_spec(s)
s.loader.exec_module(m)
# The last pair is test/precalculus/323: gold "y = 2x + 3", model answer "x=2x+3".
pairs = [("x=7", "y=7"), ("z=2x+3", "y=2x+3"), ("x=2x+3", "y = 2x + 3")]
bad = 0
for pred, gold in pairs:
    r = m.is_equiv(pred, gold)
    print(f"is_equiv({pred!r},{gold!r}) observed={r} expected=False")
    bad += r
if bad:
    print("REPRODUCED")
    sys.exit(1)

Expected: observed=False for all three pairs.
Actual on main:

is_equiv('x=7','y=7') observed=True expected=False
is_equiv('z=2x+3','y=2x+3') observed=True expected=False
is_equiv('x=2x+3','y = 2x + 3') observed=True expected=False
REPRODUCED

(exit code 1.) I did not count how many MATH answers are equations with a short left-hand side.

Reproduces on main at 985bdc1696e88e8643f081a0ff4719da39f2ae2a.