Benchmarks & comparisons

Disposable-email classification across four services, plus IronDome API latency from four locations.

Disposable-email classification

We sent the same addresses to IronDome, Abstract Email Reputation, Mailboxlayer and Kickbox, and compared each service’s disposable-email signal.

Identical inputs: 200
Disposable candidates: 100
Non-disposable controls: 100

Classification results
Service	Disposable flagged	Controls flagged	Classified
IronDome	100 / 100	0 / 100	200 / 200
Abstract Email Reputation	43 / 100	0 / 100	200 / 200
Mailboxlayer	24 / 100	0 / 100	200 / 200
Kickbox	12 / 100	0 / 100	200 / 200

Disposable flagged: Disposable candidates identified by each service. Higher is better for this sample.
Controls flagged: Non-disposable controls marked disposable. Lower is better for this sample.

Classified counts inputs with a usable disposable-email result. All four services classified all 200 inputs in the completed results.

How to read this comparison

The sample contains 100 disposable candidates and 100 non-disposable controls. It includes consumer-provider addresses, temporary domains, business domains and education-style domains. Of 200 addresses, 79 were observed in the source snapshot and 121 have synthetic local parts.

Disposable candidates were sampled from IronDome’s own lists. Each count uses only the explicit disposable-email signal. Synthetic addresses test domain classification, not mailbox existence. The source records expected labels as provisional; this balanced sample is not an independent accuracy study or a representative sample of customer traffic.

APIs tested: IronDome’s paid REST check, Abstract Email Reputation, Mailboxlayer with SMTP disabled and Kickbox’s public disposable checker. Results compare each API’s disposable-email signal.

Download anonymized results (JSON) — includes per-sample outcomes, category counts, source hashes and limitations; excludes email addresses and credentials.

API latency

SDK and REST latency from four Google Cloud locations: US East, US West, Europe and Singapore. These measurements cover IronDome only.

Locations: 4
Requests: 15,240
Unexpected failures: 0

Single checks

One email per request, using reused HTTP connections.

Median: 50% of requests finished within this time. p95: 95%. p99: 99%.

Single-check latency (ms)
Location	Client	Median	p95	p99
US East	Node.js SDK	15.52	17.84	27.69
US East	REST	15.52	17.94	32.66
US West	Node.js SDK	10.52	13.32	19.86
US West	REST	10.26	13.58	21.44
Europe	Node.js SDK	6.62	7.73	12.96
Europe	REST	6.43	7.63	10.48
Asia	Node.js SDK	4.84	7.92	19.47
Asia	REST	5.25	8.31	15.21

1,500 requests per client, per location. All times in milliseconds.

Connection reuse

The same HTTP client across requests, compared with a new connection for each request.

Median latency (ms)
Location	Client	Reused	Fresh
US East	Node.js SDK	15.52	47.71
US East	REST	15.52	47.76
US West	Node.js SDK	10.52	32.06
US West	REST	10.26	32.15
Europe	Node.js SDK	6.62	20.71
Europe	REST	6.43	20.75
Asia	Node.js SDK	4.84	16.24
Asia	REST	5.25	16.10

Median times in milliseconds. 1,500 reused-connection requests and 150 fresh-connection requests per client, per location.

Bulk checks

Median time to check 5, 25 or 100 emails in one request. Each value covers the entire batch.

Median duration for the entire batch (ms)
Location	Client	5 emails	25 emails	100 emails
US East	Node.js SDK	15.77	16.11	17.48
US East	REST	15.39	16.17	17.48
US West	Node.js SDK	10.94	10.91	13.07
US West	REST	11.09	10.99	12.59
Europe	Node.js SDK	7.25	6.86	7.56
Europe	REST	7.15	6.95	7.63
Asia	Node.js SDK	5.19	6.16	6.22
Asia	REST	4.83	5.09	6.61

Milliseconds per batch. 75 batches per size, per client, per location.

Latency methodology

Google Cloud: North Charleston, Las Vegas, Groningen and Singapore. Locations are IP estimates.
Three rounds over about seven minutes per location. A dataset of 10,000 test addresses across 1,015 consumer, custom and disposable domains.
Paid API, using REST and a Node.js SDK source version with bulk support. Matching connection limits, no warmup and no retries. Times include the network and response handling.
91,200 valid email inputs and 240 invalid-input controls.

Download latency results (JSON)