Where delivery rates come from

Not us. Every percentage on this site is reported by the wholesale source that supplies that option — measured across that source’s own order volume, which is far larger than ours, and published per option over several rolling windows. What we do is choose which of those windows to show, and rank on it.

This is the policy, not a temporary gap: we do not publish a delivery rate of our own and do not intend to. A second number, counted differently on a much smaller sample, would compete with the supplier’s figure without being any more reliable. One number, clearly attributed, is more useful than two that disagree.

The limit of borrowing a number is that we inherit its definition. A source decides for itself what counts as delivered and over what population, and we cannot audit that from the outside. Treat the percentage as the supplier’s own report on its own inventory — useful for comparing options, not an independent audit.

One: the last 24 hours. The percentage on an option is that source’s delivery rate for it over the last twenty-four hours, and nothing else. No blending, no weighting table, no long-run average sitting underneath it.

Twenty-four hours covers a full day-and-night cycle, so the figure is not biased by the hour you happen to look at it. It was also populated on every measured option in the catalogue when this was checked, so choosing it costs no coverage.

The source publishes other windows — one hour, three hours, three days, thirty days — and this site used to blend four of them. That blend was dropped in August 2026 after measuring what it did.

Measured across 398 options with published rates. The thirty-day figure read higher than the last twenty-four hours 61% of the time, by an average of 2.8 points — it was quietly propping up routes that had already stopped working.

The clearest case: 109 options had delivered nothing at all for three days and still carried a non-zero rate, purely on a month-old memory. One Austrian route read 0% over three hours, twenty-four hours and three days alike, while its thirty-day figure said 18.4% — and the site printed 3.7% for it. Those all read 0% now, which is what they are.

A three-hour window sounds like a useful early warning: it would catch a route that died an hour ago, while a twenty-four-hour figure still remembers yesterday. It was tested and it does not hold up.

In the same measurement, 17 options read 0% over three hours while their twenty-four-hour rate looked healthy. But on 12 of those 17, the three-day figure agreed with the twenty-four-hour one rather than with the three-hour zero. One read 0% over three hours, 76% over twenty-four and 77% over three days — a route delivering three codes in four does not stop dead and then resume; it had a quiet few hours.

The reason this cannot be resolved is that the source publishes rates without the number of orders behind them. A 0% over three hours may be nought out of two. Demoting on it would push roughly twelve working options down the list to catch at most five that might genuinely be sliding.

So the short window is not used, and the honest residual risk is this: a route that stopped working in the last few hours can keep showing yesterday’s figure for up to a day. That is what the automatic refund is for — it does not depend on the percentage being right.

Some options carry no percentage. That means the source supplying that option does not publish a delivery rate for it — not that the line performed badly. We would rather leave the space empty than fill it with a figure that looks precise and is not.

This is not rare, and it is not evenly spread. Only one of our supply sources reports delivery rates at all; every option from the others reads “measuring” permanently, no matter how well it actually performs, because there is no figure to show and we will not invent one. So the absence of a number tells you about the data, not about the line.

That has a consequence worth stating plainly: because the ranking leads with evidence, an option with no published rate sits below the measured ones even if it would have done well. Where a rate is missing, the other signals on the row — whether the pool is physically backed, and stock depth — are what you have to go on.

Options are grouped by which wholesale source they come from, and the groups are in a fixed order that reflects what each source can actually show you:

1. The source whose entire inventory is physical SIM. That is a structural guarantee, and no percentage overrides it.
2. The source that publishes delivery rates. Inside this group, options are ordered by that rate, high to low.
3. The source that publishes neither a guarantee nor a rate. Usually the cheapest, and last for that reason.

Within a group the order is line type, then reported rate, then price. Price is never the first thing sorted on, and a cheaper option does not move up for being cheaper.

One thing you can check rather than take on trust: the same markup formula is applied to every option from every source — there is no per-source pricing rule. What differs between the groups is what each source will tell you about its own inventory.

It is backward-looking. A route that delivered 90% this morning can fail this afternoon, and the number on screen will lag that change by minutes.

It is also aggregate. Your individual order either arrives or it does not; a 90% rate does not mean you get 90% of a code. That is exactly why the refund is automatic — the percentage informs your choice, the refund covers the outcome.

Because a rate you cannot trust is worth nothing. Options sitting at 0% stay visible on the page with their real figure attached, rather than being quietly hidden. If the honest number for a cheap route is 4%, that is what it says.