N'Ko Public Blog Series

This folder is the public narrative version of the final four-paper bundle. The older `blog/posts/` drafts had the right energy: they opened with Solomana Kante, walked through the experiments as they actually unfolded, used numbers as evidence rather than decoration, and treated N'Ko as both a technical system and a living script. This series keeps that voice while tightening the claims around the final research record. The research papers under `../final/` preserve the formal argument. These essays are for readers who need to understand why the work matters before they care about the LaTeX. | # | Post | Job | |---|------|-----| | 0 | `00-field-guide-to-the-claim.md` | The map: what the project found, what 20.57% CER means, and which claims stay separate. | | 1 | `01-when-unicode-is-not-understanding.md` | The brain scan: why rendering N'Ko is not the same as representing it. | | 2 | `02-why-20-57-cer-matters.md` | The measurement story: why direct N'Ko CER is a serious ASR result. | | 3 | `03-the-model-that-listened-in-nko.md` | The architecture story: how Whisper features, CTC, trajectory state, TAR, and TTT fit together. | | 4 | `04-after-the-transcript.md` | The deployment story: why AGP exists after the transcript and what Djoko taught us. | Define the machinery in plain English before using labels like CTC, TAR, TTT, or AGP. Use numbers as turning points in the story: 32 tokenizer entries, 2.94x translation tax, 290,596 speech pairs, 216,225 edits, 1,050,967 reference characters, 20.57% CER.