The giant-impact exemption on the core-temperature jump guard in _solve_with_retry (#817) applies to every attempt of the retry ladder on an impact step, not only the first. The guard exists partly because atol-relaxed retries are a real corruption source, and the surrounding comment says so, but on an impact step a corrupted late-attempt result passes as "the impact jump" with no check.
Scoping the exemption to the first attempt is not obviously right either: measured accretion runs show impact steps exhausting the ladder, and a narrower exemption would hold the interior more often on exactly those steps. This needs a decision informed by run behaviour rather than a quick patch, which is why it is an issue and not part of #817.
The giant-impact exemption on the core-temperature jump guard in
_solve_with_retry(#817) applies to every attempt of the retry ladder on an impact step, not only the first. The guard exists partly because atol-relaxed retries are a real corruption source, and the surrounding comment says so, but on an impact step a corrupted late-attempt result passes as "the impact jump" with no check.Scoping the exemption to the first attempt is not obviously right either: measured accretion runs show impact steps exhausting the ladder, and a narrower exemption would hold the interior more often on exactly those steps. This needs a decision informed by run behaviour rather than a quick patch, which is why it is an issue and not part of #817.