Source-data freshness, not HTTP latency.
Vehicle Auction API Performance and Data Freshness
Production availability, latency distribution and source-data freshness are measured separately so developers can understand both API responsiveness and auction-data recency.
Approximate for supported live auction workflows.
Availability, latency and source freshness are different signals
A fast API response does not guarantee fresh source data, and an invalid API key does not mean the service is down. Apibara reports these dimensions separately so integrations can interpret them correctly.
Why p50, p95 and p99 matter
Search workloads vary. A direct VIN lookup is not the same workload as a broad inventory query with filters, location constraints and pagination. Percentiles expose this distribution better than one average response time.
What counts as service failure?
Server-side failure and relevant to production availability.
Authentication, validation, subscription or quota responses. The API is still responding correctly.
{
"server_failure": "5xx",
"client_4xx_is_downtime": false,
"freshness_is_latency": false
}
Data freshness is not HTTP response time
Source freshness describes how recently supported Copart or IAAI records were collected or changed. HTTP latency measures how quickly the API serves a request. These answer different questions.
How to keep large auction integrations fast
Continue from API cursors instead of rebuilding the full catalog.
Process changed or recently updated inventory rather than everything.
Use platform, VIN, lot, make, model, dates and locations intentionally.
Avoid repeatedly requesting metadata that rarely changes.
Performance is aggregated from production /api/v1/vehicle-auction/*
request logs. Server-side 5xx responses are treated as service failures; client 4xx responses are not counted as server downtime.
Freshness remains a separate measurement from request latency.
Frequently asked questions
Do 401, 422 or 429 responses count as downtime?
No. They are client, authorization, validation or quota responses. Server-side 5xx responses are the primary failure class for production availability.
What do p50, p95 and p99 mean?
They describe response-time distribution. p50 is around the median, while p95 and p99 expose slower tail requests.
Is data freshness the same as API latency?
No. Latency measures how long a request takes. Freshness measures how recently auction data changed in the dataset.
Is there a machine-readable status endpoint?
Yes. Use /status.json and /performance.json.