Swapping measurement keys with with_measurement_key_mapping collapses a SympyCondition (a > b becomes a > a or b > b) #8529

SympyCondition.replace_key substitutes one key at a time (self.expr.subs({str(current): sympy.Symbol(str(replacement))})), so a mapping that swaps two keys is applied sequentially and merges them.

import cirq, sympy
q0, q1, q2 = cirq.LineQubit.range(3)
a, b = sympy.symbols('a b')
c = cirq.Circuit(cirq.X(q0), cirq.measure(q0, key='a'), cirq.measure(q1, key='b'),
                 cirq.X(q2).with_classical_controls(cirq.SympyCondition(a > b)),
                 cirq.measure(q2, key='out'))
s = cirq.with_measurement_key_mapping(c, {'a': 'b', 'b': 'a'})
print(s)                                          # condition is now a > a or b > b (varies between runs)
print(cirq.Simulator().run(c).records['out'])     # [[[1]]]
print(cirq.Simulator().run(s).records['out'])     # [[[0]]], expected [[[1]]] (b=1 > a=0)

The measurements themselves are swapped correctly; only the condition is wrong, so the result changes silently. Which of a > a / b > b you get depends on the iteration order of the condition's keys, which can differ between interpreter runs (string hash randomization). Substituting all keys at once (subs(..., simultaneous=True) or a single xreplace of the full mapping) would avoid this.

Seen on 1.8.0.dev0; line unchanged on current main (cirq-core/cirq/value/condition.py). Different cause from #8525 and #8528, same family (condition objects under key rewriting).

Disclosure: found with AI-assisted differential testing that I direct; the snippet above was run before posting.

Saved-version clarification: a separate saved run on November 12, 2026 also reproduced this result with Cirq 1.8.0.dev20261111222613: the mapped condition was a > a, and the original/mapped outputs were [[[1]]] / [[[0]]]. That saved run is separate from the original 1.8.0.dev0 evidence.