Update README.md

README.md

@@ -13,9 +13,29 @@ language:
 Classifier weights and lexical resources for the **LittleCurriculum** five-stage
 K–5 text filter.
 
+LittleCurriculum is produced from FineWeb-Edu using five sequential stages:
+
+1. Age-of-Acquisition and word-frequency pre-filtering
+2. fastText grade-level classification
+3. ModernBERT grade-level classification
+4. Advanced mathematical and symbolic notation filtering
+5. Frequency sampling based on Beyond-K–5-associated vocabulary
+   
 These files are the runtime dependencies of
 [`littlelearner-ll/littlecurriculum-filter`](https://github.com/littlelearner-ll/littlecurriculum-filter).
 
+## Intended Use
+
+**The classifiers are trained for web prose.** 
+Stages 2 and 3 were trained on labels generated from FineWeb-Edu documents. Performance may degrade on substantially different data distributions, in which case retraining the classifiers is recommended.
+
+**The grade boundary is fixed to K–5.** 
+Stages 2 and 3 use classifiers specifically trained to distinguish K–5 from higher-grade content. Retargeting the pipeline to a different grade boundary therefore requires retraining these classifiers.
+
+**The classifiers expect whole documents.** 
+They estimate the overall grade level of a document and have little context to work with for very short snippets. We recommend applying the pipeline to full documents rather than individual sentences.
+
+
 ## Usage
 
 ```bash
@@ -26,6 +46,7 @@ python download_artifacts.py          # fetches this repo
 python filter_k5.py --in shard.parquet --out kept.parquet
 ```
 
+
 ## Contents
 
 | Path | Size | Used by | What it is |
@@ -48,17 +69,6 @@ annotation of FineWeb-Edu would have been prohibitively expensive, which is
 what motivates the cascaded design: a cheap lexical stage, then fastText, then
 the ~50× more expensive ModernBERT.
 
-## Limitations
-
-**The classifiers are trained for web prose.** 
-Stages 2 and 3 were trained on labels generated from FineWeb-Edu documents. Performance may degrade on substantially different data distributions, in which case retraining the classifiers is recommended.
-
-**The grade boundary is fixed to K–5.** 
-Stages 2 and 3 use classifiers specifically trained to distinguish K–5 from higher-grade content. Retargeting the pipeline to a different grade boundary therefore requires retraining these classifiers.
-
-**The classifiers expect whole documents.** 
-They estimate the overall grade level of a document and have little context to work with for very short snippets. We recommend applying the pipeline to full documents rather than individual sentences.
-
 
 ## Citation