Comparison measured September 25, 2026

What the rewrite improves

I used the same 21 checks to compare saved versions of both engines. The rewrite fixes several ways to lose information or get a misleading answer. That gives me a more useful starting point for game behavior; explanation support remains unfinished.

The improvement appears early

On 21 selected checks replayed against saved code, the earlier engine passes 10, the first rewrite passes 17 including one accepted timeout, and the next rewrite passes 18 including two accepted timeouts. — The first saved rewrite already passes most of these checks. Hatched portions are accepted timeouts. Larger graph.

What changes for the caller

Examples from the latest compared versions
Behavior	Earlier	Rewrite
Keep distinct values	Two values lost.	All three kept.
Add to returned answers	Changes later answers.	Later answers unchanged.
Report bad arithmetic	Looks like no answers.	Reports an error.
Explanation checks	1 of 3 passes.	0 of 3 passes.

I want to debug the rules behind a character's behavior with fewer surprises from missing or altered answers. These fixes help with that. The explanation gap still needs attention.