GET /definitions?pattern= returns 503 under moderate concurrency despite rate limit not being reached

We're a downstream consumer of the ClearlyDefined API using the **GET /definitions?pattern=&minDescribedScore=80** endpoint to look up component definitions. During normal operation, our service sends 20–100 concurrent GET requests to this endpoint (one per component alias in a single search). We're intermittently seeing HTTP 503 responses on a significant portion of these requests, despite being well within the documented rate limit.

**Observed Behavior**
When sending 100 concurrent GET requests to [/definitions]:

- 25–68% of requests return HTTP 503 with body: [{"error":{"code":"503","message":"An error has occurred","innererror":{}}}]
- Rate limit headers on the 503 responses confirm budget is untouched:

Ratelimit-Policy: 2000;w=60
Ratelimit-Limit: 2000
Ratelimit-Remaining: 1986
Ratelimit-Reset: 60

- Successful requests complete in 0.1–1.1s
- Failed requests take 0.5–16s before returning 503 (sometimes up to 45–110s returning 502)
- The issue is intermittent - the same batch of requests succeeds 100% during healthy windows and fails 25-68% during degradation windows
- Zero HTTP 429 (rate limit) responses observed across hundreds of test requests

**Response Details (captured from a 503)
Headers:**

Server: cloudflare
CF-RAY: a1baca8f6cccc501-SJC
Cf-Cache-Status: BYPASS
Ratelimit-Policy: 2000;w=60
Ratelimit-Limit: 2000
Ratelimit-Remaining: 1986
Ratelimit-Reset: 60
Content-Type: application/json; charset=utf-8

**Impact**
This intermittent 503 behavior causes our service to become unavailable when several concurrent searches overlap with a ClearlyDefined degradation window. The pattern search endpoint is the core of our ClearlyDefined integration.

**Questions**

1. Is there a known backend capacity issue with the pattern search endpoint under concurrent load?
2. Is there a recommended concurrency limit we should respect (even though the rate limit allows 2000/min)?
3. Would batching via [POST /definitions] be more reliable than individual GET /definitions?pattern= calls?
4. Are there any plans to address the intermittent 503s? (Related: #749, #935, #1296)
5. Is there a status page or health endpoint we can poll to detect degradation windows?

Start by reproducing concurrent GET /definitions?pattern=&minDescribedScore=80 requests and recording the documented rate-limit headers, response codes, and latency. Compare the behavior with POST /definitions and review related issues #749, #935, and #1296. Done means identifying the cause or confirmed concurrency limit and documenting a reliable mitigation or backend fix, including how degradation can be detected.

Written by the indexing model from the issue text.