Fixed-correction and gradient-ambiguity review evidence

This bounded review covers two units:

1. FPRD-T78, the fixed-state sequential-correction separation.
2. FPRD-T79/T80, the letter-bounded gradient criterion and the exact ambiguity-three series.

The executable checker independently tests the finite identities, hostile boundaries, and small cases used during review. It does not replace the universal proofs.

Mathematical boundaries

• T78 uses two separate infima over baseline/presentation pairs. The zero baseline proves the optimized residual raw-height upper bound; every allowed baseline satisfies the residual canonical lower bound. No single baseline is asserted to attain both infima.
• The T78 target is undefined outside b^+ c a^* # Code Code. Its equality challenge is load-bearing for the forced-access argument. The horizon presentation may use polynomially many states.
• The target is not max-plus rational at any ambiguity level. After fixing an address, a hypothetical finite automaton would have to realize unary slopes 1/(i+1) with unbounded reduced denominators, whereas an n-state unary integer max-plus automaton obtains asymptotic slopes from cycle means with denominator at most n.
• T79 imports the classical theorem that a unary max-plus rational sequence is eventually affine on each residue class. Finite testing cannot prove that theorem.
• The residue modulus in T79 need not be bounded by the number of states: disjoint cycles of lengths two and three use five states but have least support period six.
• T80 needs only the three open full-dimensional cones. Boundary rays, axes, and ties do not affect the lower bound. Its three-component automaton supplies the matching upper bound.

Reproduction

Run:

python3 audit_fixed_correction_gradient.py

The published JSON output is the exact stdout of that command.