HermitCrab analysis: final-template interleaving prune with a grammar-wide partiality guard (rejected: zero prunes on Sena, Amharic, Mbugwe) #590

The experiment skipped the inverse interleaving of non-template affix processes after a final template in analysis, mirroring synthesis. It was guarded so it applied only when every relevant template was final and the grammar had no partial morphemes. It was removed from production.

Measured. The prune fired zero times on all three real grammars. Sena has 25 partial lexical entries and six partial rules. Amharic and Mbugwe each have one partial rule. Synthetic tests showed it was safe but showed no value.

Why rejected. The guard made it inert on real grammars, and carrying branches, state, toggles and counters for a zero-effect change broke the minimum-surface goal. Addendum 2026-09-03: with the guard lifted, the same prune measured 8-33x on Mbugwe and Sena with identical analysis sets (298x on the heaviest Sena word). Partial lexical entries cannot rescue a pruned branch, so a rules-only, per-stratum guard is sound. That work continued in #491, which merged as b240b7b5, and in #551 and #553.

Record: https://github.com/sillsdev/machine/blob/2eba2f502338705a4d529d7034cecb6303db7f35/docs/rejected-optimizations/final-template-prune.md
Ledger: https://github.com/sillsdev/machine/blob/2eba2f502338705a4d529d7034cecb6303db7f35/docs/hermitcrab-optimization-ledger.md; disposition: https://github.com/sillsdev/machine/blob/2eba2f502338705a4d529d7034cecb6303db7f35/docs/hermitcrab-perf-2026-09-disposition.md
From #490 (closed).