diff --git a/TODO.sota/66-b5-persian-registers.md b/TODO.sota/66-b5-persian-registers.md index f49cef7..a2fd812 100644 --- a/TODO.sota/66-b5-persian-registers.md +++ b/TODO.sota/66-b5-persian-registers.md @@ -19,3 +19,15 @@ Acceptance: register-stratified external-protocol numbers published. Owner picks the sources -> corpora/.yaml with sha256 pins -> the register-hold arm (target: CER < 0.7 across registers). + + +## Register gap MEASURED (2026-10-26, full run) + +fas-diac-base-1.0 on the Ganjoor vocalized poetry eval (3,497 verses, +79 poets, harakat-stripped inputs vs vocalized gold): **CER 33.67%** — +vs 0.603% on the news-classic register. The register gap is ~56x, now +quantified. The poetry corpus + eval are pinned +(run-089-fas-poetry-eval, sha 94f318f688b556be). The register-hold arm +(<0.7 across registers) requires poetry in training: the 3,497 +vocalized verses are the seed; the owner source picks (dossier above) +scale it.