Proxy Free Trial Terms and Refund Policies Compared

There is exactly one question a trial can answer that a sales page cannot: does this network return correct, unblocked responses from my targets, at my concurrency, right now? Everything else on a provider's feature list is either verifiable from documentation (ports, protocols, geo coverage, dashboard features) or unverifiable in three days regardless (uptime over a quarter, support response at 3am, pool churn next month).

So treat the trial as a targeted experiment, not a tour. You are buying an answer to one question, and the currency you spend is a very small bandwidth allowance.

The trap is that trials are sized for the tour, not the experiment. A 100 MB allowance sounds generous until you route it through a headless browser, where a single retail product page can move 1.5 MB to 3 MB of assets. That is roughly 30 to 60 page loads. You cannot measure a success rate from 40 samples, and the arithmetic further down shows exactly why.

Providers use four distinct mechanisms and they are not interchangeable. Knowing which one you are being offered tells you what you can learn.

The under-used option is the fourth. A micro purchase has no expiry pressure, no "one per customer" clause, and no refund forfeiture attached. If a provider sells a gigabyte of pay-as-you-go traffic for the price of a sandwich, that is often a better evaluation vehicle than a 3-day trial you have to schedule your week around. It also sidesteps the biggest structural problem with trials, which is next.

The table below summarises what each vendor publishes on its own site and support pages, checked in September 2026. Trial and refund terms change frequently and are often region-specific or account-specific, so confirm the current wording on the provider's own terms page before you rely on any row here.

Two patterns repeat across the market. First, the trial and the money-back guarantee are usually mutually exclusive, so the "free" option is not free of cost, it is paid for with your refund rights. Second, refunds are almost never available on consumed bandwidth. Once the gigabytes leave the network, they are sold.

Read the terms page, not the pricing page. These five clauses appear again and again, and each one has cost buyers real money:

1. Trial activation supersedes the money-back guarantee. The most common and most expensive clause. If you plan to buy anyway, skipping the trial and relying on a 14-day or 30-day refund window frequently gives you far more evaluation capacity than the trial ever would.
2. Consumption caps on refunds. A refund that is void once you use more than 10% of the allowance, or 1 GB, means the real trial size is that cap, not the plan size. Compute the cap in gigabytes before you start testing.
3. Provider-fault-only refunds. Some policies refund only when the provider's own infrastructure failed and support could not fix it, reported within 24 hours. "It did not work on my target site" is not a provider fault under that wording.
4. Promotional and discounted purchases excluded. The coupon that saved you 30% often removes refund eligibility entirely. Read the coupon terms, not just the checkout total.
5. Auto-conversion at trial end. A card-required trial usually converts into a paid subscription on the final day unless cancelled. Set a calendar reminder for the morning before expiry, not the day of.

None of this makes a provider dishonest. It makes the terms page part of the product, and a provider whose terms are short, plain and easy to find is telling you something useful about how it operates.

This is the section most trial guides skip, and it is the one that decides whether your test means anything.

Success rate is a proportion, so its uncertainty follows the standard error of a proportion, the square root of p(1-p)/n. At an observed 93% success rate, the 95% confidence interval half-width works out like this:

Now convert that into bandwidth. Requests are not equal, and your fetch method changes the answer by two orders of magnitude:

Put the two tables together and the practical rule falls out: a 100 MB trial is statistically useful only if you fetch raw HTML and block assets. If your production workload renders JavaScript with images on, a 100 MB trial cannot produce a sample large enough to distinguish a 93% provider from an 86% one. In that case, either request a larger allowance, block assets during the trial and accept that you are testing the network rather than the renderer, or skip the trial and use a refundable purchase.

For the question that comes before any trial, which product type you should be testing at all, see our guide on choosing between residential and datacenter proxy services.

Run this against your real targets, never against an IP echo service alone. The goal is a decision, not a demo.

Step 1. Build the target list before the trial starts. Twenty to fifty URLs from the sites you actually need, including the hardest ones. Save the expected content signature for each: a price selector, a JSON key, or a string that only appears on a correctly rendered page.

Step 2. Verify connectivity and exit geography. One request per port, then a geo check:

# HTTP/HTTPS rotating endpoint
curl -x http://USER:PASS@gateway.sparkproxy.io:11000 -s https://ipinfo.io/json

# Sticky session endpoint, same IP across a session
curl -x http://USER:PASS@gateway.sparkproxy.io:11002 -s https://ipinfo.io/json

# SOCKS5
curl --socks5-hostname USER:PASS@gateway.sparkproxy.io:13000 -s https://ipinfo.io/json

Step 3. Measure success rate and latency on your targets, not on a test page. Block images and fonts so your sample size survives the allowance:

import time, requests
from concurrent.futures import ThreadPoolExecutor

PROXY = "http://USER:PASS@gateway.sparkproxy.io:11000"
PROXIES = {"http": PROXY, "https": PROXY}
TARGETS = [l.strip() for l in open("targets.txt") if l.strip()]
SIGNATURE = "add-to-cart"   # a string that only appears on a good page

def probe(url):
    t0 = time.perf_counter()
    try:
        r = requests.get(url, proxies=PROXIES, timeout=20)
        ms = (time.perf_counter() - t0) * 1000
        ok = r.status_code == 200 and SIGNATURE in r.text
        return ok, r.status_code, ms, len(r.content)
    except Exception as e:
        return False, type(e).__name__, (time.perf_counter() - t0) * 1000, 0

with ThreadPoolExecutor(max_workers=20) as pool:
    results = list(pool.map(probe, TARGETS * 20))

ok  = sum(1 for r in results if r[0])
lat = sorted(r[2] for r in results)
mb  = sum(r[3] for r in results) / 1_048_576
print(f"n={len(results)} success={ok/len(results):.1%} "
      f"p50={lat[len(lat)//2]:.0f}ms p95={lat[int(len(lat)*0.95)]:.0f}ms "
      f"bandwidth={mb:.1f}MB")

Step 4. Repeat the identical script against every provider you are comparing, in the same hour. Target sites change their defences by the day. A test you ran on Tuesday against provider A is not comparable to a test you run on Friday against provider B.

Step 5. Re-run the same script at your real concurrency. A pool that returns 97% at 5 threads and 71% at 100 threads is a pool you will hate in production. Thread limits are the single most common gap between a trial and the plan you actually buy.

Step 6. Record bandwidth consumed per successful record. This is the number that turns a per-GB price into a real monthly cost, and it is the whole basis of proxy pricing model comparisons.

For deeper diagnostics on individual proxy failures, our complete proxy testing guide covers header leaks, DNS behaviour and TLS fingerprint checks that a trial rarely surfaces on its own.

Write these down for each provider, then decide. Nothing else belongs on the scorecard.

The fourth row is where per-GB pricing gets decided in practice. Two providers at the same headline price per gigabyte can differ by 3x on real monthly cost if one of them forces more retries or serves heavier responses.

Stop testing and move on when you see any of these. Each one predicts a problem that will not improve after you pay:

• The trial requires a call with sales before you can send a single request, and the call is more than 48 hours out.
• Exit IPs return a country different from the one you requested, and support calls it "approximate targeting".
• The dashboard's usage counter disagrees with your own measured byte count by more than a few percent.
• Credentials work on the HTTP endpoint but silently fail on SOCKS, with no documentation explaining why.
• The terms page is not linked from the pricing page, or the refund policy exists only in a support ticket reply.
• Success rate is excellent on a generic IP echo endpoint and poor on every real target, which usually means the pool is clean for nothing except tests.

If a provider's pool looks suspiciously cheap, the economics behind that are worth understanding first. Our breakdown of free versus paid proxies covers where the difference actually comes from.

SparkProxy has two free entry points. The datacenter proxy plans come with a 24-hour free trial with 250 threads and no card, and the Scraping API gives you 1,000 credits, no card required. Plain fetches cost 1 credit, JavaScript rendering costs 5, and screenshots or PDFs cost 10. That converts to 1,000 raw HTML fetches or 200 fully rendered pages, comfortably above the 400-sample threshold in the sizing table for the raw-HTML case.

curl -H "X-API-Key: sk-your-key" \
  "https://scrape.sparkproxy.io/api/v1?url=https://example.com&render_js=false"

Credits are a cleaner trial currency than bandwidth because the cost of a request is fixed and published in advance. You know your sample size before you start, instead of discovering it when the allowance runs out mid-test. Paid Scraping API tiers run from Starter at $49 for 250,000 credits and 50 concurrent requests, through Growth at $99 for 1,000,000 credits and 100 concurrent, Pro at $249 for 3,000,000 and 200 concurrent, up to Scale at $599 for 8,000,000 and 400 concurrent.

For the datacenter proxy plans, the published ladder is monthly, with unlimited bandwidth and 30 days validity:

Higher tiers (Pro at 1500 threads, Pro+ at 2000) exist under the Fair Usage Policy without a public price, and custom arrangements go up to 1 Gbps. Those speed figures are ceilings under fair usage, not guaranteed sustained rates, and any provider quoting a guaranteed rate on a shared network is quoting a number it cannot control.

Because bandwidth is unlimited on these plans, the metric that matters in a SparkProxy evaluation is threads, not gigabytes. Test at the concurrency you intend to run, on the tier you intend to buy. The gateway is gateway.sparkproxy.io, with HTTP and HTTPS on port 11000, sticky sessions on 11002, and SOCKS5 on 13000. The network spans over 1 million datacenter IPs across 80+ countries, including more than 50,000 US datacenter IPs.

One last piece of arithmetic before you accept any trial. If the plan costs $99 and the money-back window is 14 days, the trial is competing against a 14-day, full-capacity, fully refundable evaluation. A 100 MB, 3-day trial that voids that window is not the cheaper option. It is the more expensive one, priced in information you never got to collect.