Latency and Throughput

Choose an endpoint by the response you need. Use simple valuations for concise results, standard valuations for optional comparables and context, and development valuations for proposed units. Use asynchronous jobs when work must survive a disconnected client or cannot reliably complete synchronously.

Synchronous Response Times

The table gives measured public HTTPS round-trip times in seconds, including request upload and complete response download. The single-property routes were used for one row; the corresponding batch routes were used for larger requests.

Measurement conditions: public service measured from a Mac on 5 October 2026, with a warm service, a reused HTTPS connection and serial requests. Each table cell is one measurement, not a percentile. Simple returns the core response; standard and development request eight comparables and location context, without attribution.

Existing-property batches repeat an already verified identity. Development batches use the same site and specification, with separate request references. No new address-provider lookups were needed. This is a reference workload, not a distinct-portfolio benchmark or a test of entirely uncached properties. Do not assume the same timings for 2,500 different addresses or sites.

These figures are planning references, not a latency SLA. Small reversals between neighbouring batch sizes are normal in individual measurements. Response time depends on the properties, evidence requested, address verification, service load and client connection; a larger batch does not have a fixed per-row cost.

Endpoint Mapping

Selecting fewer optional evidence sections can reduce response size and work, but does not change the point estimate. Simple endpoints omit comparables and context regardless of those options. Development still applies its fixed new-build specification.

Metadata and Job Requests

These operations have different timing boundaries from a synchronous valuation. The following public HTTPS observations use the same test environment:

Job completion time includes queueing and processing. The one-property job figures must not be applied to a large job or a large result download. Measure submission, time to completion and result retrieval separately; the synchronous table is not a job-completion forecast. Honour the returned polling guidance rather than repeatedly querying status as fast as possible.

Planning a Production Workload

Benchmark a representative mix of your own distinct properties before setting customer-facing expectations. Record median and tail latency by endpoint, batch size and response options. Separate first-use address verification from already verified identities, and test the concurrency allowed by your plan.

Reuse HTTPS connections. Allow time for response transfer, especially when requesting many comparables. Uploads and full downloads are part of the time a user experiences; server processing time alone is not an end-to-end measurement.

Standard and development synchronous processing has a 25-second deadline after body reception. Do not use that deadline as a guaranteed response time or an identical client timeout: allow for transport, and use jobs or smaller batches when the workload approaches the limit. See Limits and Safeguards.

A batch-row limit is not a throughput entitlement. Avoid simultaneous bursts of maximum-size requests; handle capacity responses with bounded backoff. See Errors and Retries and Batch and Asynchronous Jobs.