That's a fair statement. It's more than a roll of the dice and less than a guarantee that 100% accurate results will be obtained. Much depends on the testing entity's sample database:
"While some problems can result in merely underestimating the percentage of mutt’s ancestry that derives from a specific breed, other problems can prevent the correct breed from being identified at all. The most substantial of these problems is the absence of true ancestral breed from the reference dataset (Figure 9). Because breed ancestry is inferred by comparing chunks of mutt DNA to purebred dogs of known breeds, if a breed is absent from the reference dataset, that breed simply cannot be detected, even if it contributed a very large fraction of a mutt’s DNA. This issue will ultimately be solved only through inclusion of reference genomes from recognized breeds; in the meantime, if you are interested in knowing whether your dog has ancestry from a specific rare breed, it is important to make sure your breed ancestry company of choice is able to check for that breed. For those who decide to proceed with ancestry inference even though the breed of interest is known to be absent from the reference set, it is important to keep in mind that the absence of that breed from the list of inferred ancestors provides no information as to whether the mutt truly lacks that particular ancestry.
The mutations selected for genotyping also determine which breed ancestries can be accurately identified in a mixed-breed dog. Genotyping arrays tend to include more mutations present in common breeds. This means that chunks of chromosomes from poodles and German shepherds may be especially easy to identify because many of the mutations common in these breeds are assayed on genotyping arrays. While many mutations could help identify chunks of DNA from rare breeds such as New Guinea singing dogs or Skye terriers, some of these mutations may not be represented on widely-used genotyping arrays, which could make these breeds harder to identify. This problem will eventually be solved by creating breed reference datasets with sequence data, which would allow for the interpretation of many more mutations and would not be biased toward detection of ancestry from specific breeds.
A mutt’s relationship to its purebred ancestors also affects the reliability of breed determination. In particular, it is easier to identify the breed ancestry of DNA from a purebred ancestor who is a close relative (such as a parent) because mutations from recent ancestors will reside in longer chunks of DNA with more informative mutations. For example, while the first mutation observed on a mutt’s chromosome may be common in both Labradors and Golden Retrievers, perhaps the first, second, and third mutations observed are only seen together in Golden Retrievers. DNA contributed by ancestors from many generations back will exist as only short chromosome chunks, with fewer mutations to help identify their contribution to the mutt’s ancestry, making inference more difficult. This issue can be mitigated by using data from sequencing instead of genotyping, allowing for all mutations to be analyzed. However, DNA inherited from many generations back can be in chromosome chunks so short that it will not contain chromosome chunks characteristic of a specific breed, such that the breed’s contributions to a mutt’s ancestry cannot be detected even with whole-genome data (Li et al., 2014)."
quoted from:
https://winter2018.iaabcjournal.org/?p=140
A level of identification is possible, and will become more accurate as more breeds are entered in the database and by using a different method.
I did the AncestryDNA test last year. They noted that my DNA results may change as more information is available for comparison. I have already had one update which did change my general Ireland and Scotland genetics to a more specific North East Scotland & the Northern Isles. As the database grows the information will evolve. The DNA information on dogs is growing as it is for humans.
I don't think this exactly meets the science-based data criterion you wrote of for proof, but I think it suggests that a level of accuracy is possible and will get higher as the information and testing evolves.
This article in The Whole Dog Journal discusses genetic health testing and the similar issues that field is facing from the consumer/vet standpoint.
https://www.whole-dog-journal.com/behavior/instincts-dna/genetic-health-screening-for-dogs/