fix(openai): record outgoingTextChars for Responses transforms

transformOpenAIChatCompletions set info.outgoingTextChars = countOutgoingTextChars(req)
before returning; transformOpenAIResponses set info.compressed = true and returned
without it. That field is the regression denominator (tokens ≈ α·outgoingTextChars +
β·imagePixels), so compressed Responses rows silently carried no denominator.

- add countResponsesOutgoingTextChars(): instructions + message-item text
  (string or input_text parts, via responsesContentText so input_image base64 is
  excluded) + flat tool name/description/parameters
- set it on the Responses success path, mirroring the Chat path
- test asserts outgoingTextChars > 0 for a compressed Responses request

src/core/openai.ts

@@ -418,6 +418,32 @@ function countOutgoingTextChars(req: OpenAIChatRequest): number {
   return n;
 }
 
+/** Outgoing text-char denominator for the GPT Responses regression, mirroring
+ *  countOutgoingTextChars for Chat: instructions + message-item text (string or
+ *  input_text parts) + flat tool name/description/parameters. input_image base64
+ *  is excluded on purpose — it is image cost, not text (responsesContentText
+ *  already drops non-text parts). */
+function countResponsesOutgoingTextChars(req: ResponsesRequest): number {
+  let n = 0;
+  if (typeof req.instructions === 'string') n += req.instructions.length;
+  if (typeof req.input === 'string') {
+    n += req.input.length;
+  } else if (Array.isArray(req.input)) {
+    for (const item of req.input) {
+      n += responsesContentText((item as ResponsesInputItem).content).length;
+    }
+  }
+  if (Array.isArray(req.tools)) {
+    for (const tool of req.tools) {
+      if (!isFlatFunctionTool(tool)) continue;
+      if (typeof tool.name === 'string') n += tool.name.length;
+      if (typeof tool.description === 'string') n += tool.description.length;
+      if (tool.parameters !== undefined) n += safeStringifyLen(tool.parameters);
+    }
+  }
+  return n;
+}
+
 function safeStringifyLen(v: unknown): number {
   try {
     return JSON.stringify(v)?.length ?? 0;
@@ -986,6 +1012,9 @@ export async function transformOpenAIResponses(
 
   if (rewrittenTools !== undefined) req.tools = rewrittenTools;
 
+  // Regression denominator, same as the Chat path — Responses was the only
+  // transform that never recorded it.
+  info.outgoingTextChars = countResponsesOutgoingTextChars(req);
   info.compressed = true;
   return { body: new TextEncoder().encode(JSON.stringify(req)), info };
 }

tests/openai-gpt5.test.ts

@@ -347,6 +347,19 @@ describe('transformOpenAIResponses (gpt-5.6)', () => {
     expect(textParts.some((p) => p.text?.includes('Do the thing'))).toBe(true);
   });
 
+  it('records outgoingTextChars for compressed Responses requests (denominator parity with Chat)', async () => {
+    const body = enc.encode(JSON.stringify({
+      model: 'gpt-5.6',
+      instructions: BIG_INSTRUCTIONS,
+      input: [{ role: 'user', content: 'Please do the thing.' }],
+    }));
+    const result = await transformOpenAIResponses(body, { charsPerToken: 1, minCompressChars: 1 });
+    expect(result.info.compressed).toBe(true);
+    // Chat set this via countOutgoingTextChars; Responses never recorded it, so
+    // the tokens ≈ α·outgoingTextChars + β·imagePixels denominator was missing.
+    expect(result.info.outgoingTextChars).toBeGreaterThan(0);
+  });
+
   it('returns compressed=false with not_profitable/below_min reason for small input', async () => {
     const body = enc.encode(JSON.stringify({
       model: 'gpt-5.6',