From eb65a2be8826428adccb64069b8eb7ad0efb7739 Mon Sep 17 00:00:00 2001
From: Cita <juricek.chen@gmail.com>
Date: Mon, 30 Mar 2026 22:54:05 +0800
Subject: [PATCH] fix: preserve repeated short responses deltas (#336)

---
 src/server/routes/proxy/chat.stream.test.ts   | 45 +++++++++++++++++++
 .../openai/responses/aggregator.ts            | 12 ++++-
 2 files changed, 55 insertions(+), 2 deletions(-)

diff --git a/src/server/routes/proxy/chat.stream.test.ts b/src/server/routes/proxy/chat.stream.test.ts
index e6e696b..7bd6abe 100644
--- a/src/server/routes/proxy/chat.stream.test.ts
+++ b/src/server/routes/proxy/chat.stream.test.ts
@@ -2039,6 +2039,51 @@ describe('chat proxy stream behavior', () => {
     expect(response.body).toContain('"text":"The user asked \\"whoru\\" which is a common internet slang and shorthand for who are you."');
   });
 
+  it('preserves legitimate repeated short deltas when /v1/responses is converted from /v1/messages stream', async () => {
+    fetchModelPricingCatalogMock.mockResolvedValue({
+      models: [
+        {
+          modelName: 'upstream-gpt',
+          supportedEndpointTypes: ['/v1/messages'],
+        },
+      ],
+      groupRatio: {},
+    });
+
+    const encoder = new TextEncoder();
+    const upstreamBody = new ReadableStream<Uint8Array>({
+      start(controller) {
+        controller.enqueue(encoder.encode('event: message_start\ndata: {"type":"message_start","message":{"id":"msg_repeat_short_1","model":"upstream-gpt"}}\n\n'));
+        controller.enqueue(encoder.encode('event: content_block_delta\ndata: {"type":"content_block_delta","delta":{"type":"text_delta","text":"ha"}}\n\n'));
+        controller.enqueue(encoder.encode('event: content_block_delta\ndata: {"type":"content_block_delta","delta":{"type":"text_delta","text":"ha"}}\n\n'));
+        controller.enqueue(encoder.encode('event: message_stop\ndata: {"type":"message_stop"}\n\n'));
+        controller.enqueue(encoder.encode('data: [DONE]\n\n'));
+        controller.close();
+      },
+    });
+
+    fetchMock.mockResolvedValue(new Response(upstreamBody, {
+      status: 200,
+      headers: { 'content-type': 'text/event-stream; charset=utf-8' },
+    }));
+
+    const response = await app.inject({
+      method: 'POST',
+      url: '/v1/responses',
+      payload: {
+        model: 'claude-sonnet-4-6',
+        stream: true,
+        input: 'laugh',
+      },
+    });
+
+    expect(response.statusCode).toBe(200);
+    const deltaMatches = response.body.match(/event: response\.output_text\.delta/g) || [];
+    expect(deltaMatches.length).toBe(2);
+    expect(response.body).toContain('"delta":"ha"');
+    expect(response.body).toContain('"text":"haha"');
+  });
+
   it('preserves function_call/function_call_output when /v1/responses falls back to /v1/chat/completions', async () => {
     fetchModelPricingCatalogMock.mockResolvedValue({
       models: [
diff --git a/src/server/transformers/openai/responses/aggregator.ts b/src/server/transformers/openai/responses/aggregator.ts
index dfa1b50..ccfe536 100644
--- a/src/server/transformers/openai/responses/aggregator.ts
+++ b/src/server/transformers/openai/responses/aggregator.ts
@@ -656,13 +656,21 @@ function mergeImageGenerationFields(
 }
 
 function computeNovelDelta(existingText: string, incomingDelta: string): string {
+  const replayWindowMinLength = 24;
   if (!incomingDelta) return '';
   if (!existingText) return incomingDelta;
-  if (existingText.endsWith(incomingDelta)) return '';
+  if (existingText.endsWith(incomingDelta)) {
+    return incomingDelta.length >= replayWindowMinLength ? '' : incomingDelta;
+  }
   if (incomingDelta.startsWith(existingText)) {
     return incomingDelta.slice(existingText.length);
   }
-  if (existingText.includes(incomingDelta)) return '';
+  if (
+    incomingDelta.length >= replayWindowMinLength
+    && existingText.includes(incomingDelta)
+  ) {
+    return '';
+  }
 
   const maxOverlap = Math.min(existingText.length, incomingDelta.length);
   for (let overlap = maxOverlap; overlap > 0; overlap -= 1) {