// Probe 01: per model — (a) tiny generateContent with systemInstruction, (b) generateContentStream chunk shape,
// usage arrival pattern, part.thought flag, finishReason, modelVersion. Also raw REST SSE for one model.
import { ai, MODELS, KEY, SUFFIX, save, errInfo, shortText, withRetry, sleep, PACE_MS } from "./lib.js";

const results: any = {};

for (const model of MODELS) {
  const r: any = (results[model] = {});
  // (a) + (i) non-streaming with systemInstruction
  try {
    const res = await withRetry(() => ai.models.generateContent({
      model,
      contents: "Reply with exactly one word: what colour is the sky on a clear day?",
      config: { systemInstruction: "You are terse. Always answer in UPPERCASE.", maxOutputTokens: 400 },
    }));
    r.generate = {
      ok: true,
      text: shortText(res),
      modelVersion: res.modelVersion,
      responseId: res.responseId,
      finishReason: res.candidates?.[0]?.finishReason,
      partKeys: res.candidates?.[0]?.content?.parts?.map((p: any) => Object.keys(p)),
      role: res.candidates?.[0]?.content?.role,
      usage: res.usageMetadata,
      topLevelKeys: Object.keys(res).filter((k) => !k.startsWith("_")),
    };
  } catch (e) {
    r.generate = { ok: false, error: errInfo(e) };
  }

  // (b) streaming
  try {
    await sleep(PACE_MS);
    const stream = await withRetry(() => ai.models.generateContentStream({
      model,
      contents: "Count from 1 to 12 separated by spaces, then say DONE.",
      config: { maxOutputTokens: 2000, thinkingConfig: { includeThoughts: true } },
    }));
    const chunks: any[] = [];
    for await (const c of stream) {
      chunks.push({
        parts: c.candidates?.[0]?.content?.parts?.map((p: any) => ({
          keys: Object.keys(p), thought: p.thought ?? undefined, textLen: p.text?.length, hasSig: !!p.thoughtSignature,
        })),
        role: c.candidates?.[0]?.content?.role,
        finishReason: c.candidates?.[0]?.finishReason,
        usage: c.usageMetadata,
        modelVersion: c.modelVersion,
        responseId: c.responseId,
        keys: Object.keys(c).filter((k) => !k.startsWith("_") && (c as any)[k] !== undefined),
      });
    }
    r.stream = {
      ok: true,
      chunkCount: chunks.length,
      usageOnChunks: chunks.map((c) => (c.usage ? (c.usage.candidatesTokenCount != null ? "full" : "partial") : "none")),
      thoughtChunks: chunks.filter((c) => c.parts?.some((p: any) => p.thought)).length,
      lastUsage: chunks.at(-1)?.usage,
      finishReasons: chunks.map((c) => c.finishReason ?? null),
      sampleFirst: chunks[0],
      sampleLast: chunks.at(-1),
    };
  } catch (e) {
    r.stream = { ok: false, error: errInfo(e) };
  }
  await sleep(PACE_MS);
  console.log(model, JSON.stringify({ gen: r.generate.ok ? r.generate.text : r.generate.error, stream: r.stream.ok ? { n: r.stream.chunkCount, usage: r.stream.usageOnChunks, thoughts: r.stream.thoughtChunks } : r.stream.error }));
}

// Raw REST SSE protocol sample (one model)
if (!process.argv[2]) {
  const model = "gemini-3.5-flash-lite";
  const res = await fetch(`https://generativelanguage.googleapis.com/v1beta/models/${model}:streamGenerateContent?alt=sse`, {
    method: "POST",
    headers: { "x-goog-api-key": KEY, "content-type": "application/json" },
    body: JSON.stringify({ contents: [{ role: "user", parts: [{ text: "Say hello in three words." }] }], generationConfig: { maxOutputTokens: 50, thinkingConfig: { thinkingBudget: 0 } } }),
  });
  const raw = await res.text();
  results.restSse = { status: res.status, contentType: res.headers.get("content-type"), rawFirst1200: raw.slice(0, 1200), eventCount: raw.split("\n\n").filter((s) => s.trim()).length };
  console.log("REST SSE", res.status, res.headers.get("content-type"), "events:", results.restSse.eventCount);
}

save(`01-basic-stream${SUFFIX}.json`, results);