Update README.md

README.md

@@ -26,20 +26,15 @@ The idea was to retain the model's core intelligence and knowledge, while divers
 
 --------------------
 
-I'd say v2 is a slight departure from v1, rather than an improvement.
+**v2 better fulfills the agreements from previous turns.** The promised test is carried out, the scheduled meeting actually takes place, but the minor characters' micro-reactions are less developed than in v1.
 
-**It's more compact, more specific, "quicker in making decisions," with fewer cliches**, while v1 is stylistically richer in its embellishments.
+**The prose has become a bit drier and more concrete**, with fewer physical cliches like "clenched jaw" and "white knuckles," at the cost of losing the subtext and subtle inner speech that are v1's strength.
 
-I'll try to return to the style from v1 in the new iteration (v3), I want to keep the gradual atmosphere. v2 will suit those who don't like the long intro and want more action.
+**v2 advances the plot instead of developing a single situation**, and distributes the length according to need—shorter where there's nothing to develop, longer where the scene shifts.
 
-Also in the files, you can find a `dump.json` of generations v1 and v2 with several basic prompts and character cards. LLM player - GLM 5.2, temperature 1.0
-
-----------------
-
-What I couldn't fix:
-- "Slop" still appears; it’s a carryover from the base model or Styletune, though it looks slightly different.
-- I’m not sure if StyleTune should have been kept in v2. I tested without it but couldn't determine whether it improves or degrades the narrative. It definitely adds more "RP" flavor, but it might also be adding more slop.
+But in any case, **the cliché density hasn't changed** (2.42 vs. 2.47 per thousand words with thoughts) and has increased in the non-thought mode (1.70 vs. 2.54), with only the set of turns shifting.
 
+--------------------
 
 ## What changed from v1