9950 pairs from git acf0d2c

README.md

@@ -17,7 +17,7 @@ tags:
 
 # CankarParallel - plain Slovene to Ivan Cankar's prose voice
 
-10,043 parallel passages for Slovene literary style transfer. Each row
+9,950 parallel passages for Slovene literary style transfer. Each row
 pairs a passage of Ivan Cankar's prose with a plain modern Slovene rendering of
 the same content.
 
@@ -50,7 +50,7 @@ disagreement, a reflexive-only verb used transitively - which a char-ngram
 style classifier scored identically to the stronger model's output. On a
 low-resource language the cheap option corrupts the data invisibly.
 
-10,043 of 10,049 responses passed the
+9,950 of 10,049 responses passed the
 quality filters. Rejections are quarantined, never silently dropped.
 
 ## Held-out works are excluded
@@ -73,10 +73,13 @@ redistribution. Treat it as reference material until the card says otherwise.
 
 ```
 uv run cankar pairs segment
-uv run cankar pairs destyle --limit 10000
+uv run cankar pairs destyle --parse-only   # re-derives pairs from the raw log
 uv run cankar pairs publish
 ```
 
+The middle step re-parses committed raw responses and costs nothing. Generating
+them from scratch is `destyle --limit N`, which submits paid batches.
+
 Provenance: corpus sha256 `d9b05bf04db96db6`, passages sha256
-`e1cad3a2643782f9`, pairs sha256 `16b83a4dce791773`,
-generated at 2026-07-28T11:29:15+00:00 from git `b3095bb`.
+`e1cad3a2643782f9`, pairs sha256 `c387b33c7d036b95`,
+generated at 2026-07-28T12:39:40+00:00 from git `acf0d2c`.