A ranked list is a strong claim: it asserts that the evidence can tell first from second. Most of the time, for most of these providers, it cannot, and the honest way to say so is a tie.
Start with what a single score is on this site. It is not one number but three. The floor is what a provider has earned outright, counting every dimension we have not measured as zero. The potential is the most it could reach if every unmeasured dimension hit its own ceiling, and those ceilings are not all ten: unmeasured streaming is credited at 6.5, unmeasured speed at 8.5, because a number we did not earn the right to certify cannot reach the top of the scale. The headline is the familiar renormalised average, held down so it can never exceed the potential. Drawn to scale, the band from floor to potential is the honest width of what we do not yet know.
The only ordering the evidence licenses
Given intervals rather than points, there is exactly one comparison that is safe to make. A provider A ranks above a provider B when A’s guaranteed minimum clears B’s best possible maximum over everything we have not measured:
floor(A) > potential(B), A dominates B. Anything weaker than this is an artefact of how much of each provider we happened to research, not a fact about the providers.This is a strict partial order, and its most important property is what it refuses to do. It cannot be moved by the sequence in which we research providers. Add a citation for one provider and its floor can only rise; remove one and its floor can only fall. A renormalised average has no such discipline, research a provider a little more and its average can drift in either direction, which means an average-based leaderboard partly ranks providers by our own attention. Dominance does not, and that is the whole reason we use it.
What the current corpus actually says
Run the rule over today’s five-provider corpus and the dominance relation comes back empty. The highest floor sits well below the lowest potential, so no provider’s guaranteed minimum clears any other provider’s ceiling. Nothing dominates anything. That is not a defect in the data; it is the data being honest about how much is still unmeasured. When the intervals are that wide, first and fifth genuinely overlap.
A Top 5 would paper over exactly that overlap. To print 1-2-3-4-5 we would have to rank on the headline, the one number of the three that is an estimate rather than a bound, and present a difference the evidence cannot defend as if it were a finding. So the site shows one five-way tie, with every interval drawn to scale beside it, and says plainly that no provider has separated from the field.
No tiebreak, on purpose
An earlier version of the engine broke ties by how much of each provider we had researched, promoting the ones we knew most about. It has been deleted. That rule rewarded our own coverage rather than the provider’s merits, and its mirror image was worse: it quietly penalised the providers we had looked at least. A tie with no tiebreak is the correct output when the evidence separates no one. As our own measurements fill in, server authenticity first, floors will rise, potentials will tighten, and real separations will appear. When they do, they will be separations the physics and the documents actually support, not a rank order we imposed to look decisive.