/**
 * Mock stand-in for the real atlas `POST /api/chat`. Streams the same SSE event shape
 * (`queue` → `reasoning` → `content` → `done`) as the real endpoint would, but replays a
 * canned reply instead of calling vLLM — no rate limiting, no budget, no real model.
 */
const REPLY_REASONING =
  'The user is asking about a gap in the wave column display. Argo CD sync waves are integers used purely for ordering; there is no requirement that they be contiguous. The cluster uses 0,1,2,3,5,6,7,8. This is normal and usually happens when a wave is retired or intentionally reserved.'

const REPLY =
  'Wave 4 is empty. Waves are sort keys, not a sequence — Argo orders by value and skips gaps, so 3 is followed directly by 5. Nothing is missing.'

const wait = (ms: number) => new Promise((resolve) => setTimeout(resolve, ms))

export async function POST() {
  const encoder = new TextEncoder()

  const stream = new ReadableStream({
    async start(controller) {
      const send = (type: string, payload: Record<string, unknown> = {}) => {
        controller.enqueue(encoder.encode(`data: ${JSON.stringify({ type, ...payload })}\n\n`))
      }

      for (let position = 3; position > 0; position--) {
        send('queue', { position })
        await wait(500)
      }

      send('reasoning', { text: REPLY_REASONING })
      await wait(200)

      let i = 0
      while (i < REPLY.length) {
        i += 3
        send('content', { text: REPLY.slice(0, i) })
        await wait(40)
      }

      send('done')
      controller.close()
    },
  })

  return new Response(stream, {
    headers: {
      'Content-Type': 'text/event-stream',
      'Cache-Control': 'no-cache',
      Connection: 'keep-alive',
    },
  })
}