Обрабатывайте пустые ответы и повторы, не платя дважды

# FOUND — data populated, charged
# < 200  X-Credits-Charged: 5
{ "status": "OK", "data": { /* … */ } }

# MISS — ran honestly, found nothing, billed zero
# < 200  X-Credits-Charged: 0
{ "status": "OK", "data": null }

# FAILURE — nothing billed, safe to retry
# < 503

# Branch on the payload BEFORE the status code:
# a miss and a hit are both HTTP 200.
if res.status_code >= 500:
    return retry_later()    # transient

if res.json()["data"] is None:
    cache_miss(query, ttl=86_400)  # a real answer — do not retry
    return None

return res.json()["data"]

# A stable key per logical operation. If you never saw the first
# response, the retry resolves to it instead of billing twice.
curl -X POST https://api.qannasapi.com/v1/data/contacts/find-email \
  -H "Authorization: Bearer $QANNAS_API_KEY" \
  -H "Idempotency-Key: signup-8f21c4" \
  -H "Content-Type: application/json" \
  -d '{"query": "Jane Doe, Acme Logistics"}'

import random, time

def call_with_retry(send, *, max_attempts=5):
    for attempt in range(max_attempts):
        res = send()

        if res.status_code == 422:
            raise ValueError(res.text)  # malformed — never resend

        if res.status_code < 500:
            return res

        # 5xx: upstream failures are never billed, so retrying is free
        time.sleep(0.5 * 2 ** attempt + random.uniform(0, 0.5))

    raise RuntimeError("exhausted retries")