{"version":"2.1.232","anchor":"conversation-normalization-reuses-its-previous-result","canonical_anchor":"conversation-normalization-reuses-its-previous-result","heading":"Conversation normalization reuses its previous result","tier":"notice","area":"Performance","scope":null,"heads_up":null,"url":"https:\/\/changelogs.core-directive.com\/v\/2.1.232\/e\/conversation-normalization-reuses-its-previous-result","release_url":"https:\/\/changelogs.core-directive.com\/v\/2.1.232","markdown":"### Conversation normalization reuses its previous result\n\nPreparing your conversation for each request reuses the unchanged parts instead of rebuilding it.\n\n**What**\n\nPreparing the message list for a request now caches the previous input and output and reuses the unchanged prefix and suffix when the conversation grows, rather than rebuilding everything each turn.\n\n**Details**\n\n- Only streaming assistant messages whose stop reason is still null are re-walked.\n- That pending set is pruned once it exceeds a threshold constant.\n\n**Evidence**\n\n`PRUNE_PENDING_AT`\n\n- Area: Performance\n- Tier: You'll notice\n- Useful: 3\/5\n- Signal: 2\/5"}