This article begins from two different places, and I want to separate them at the outset.
The first is Cubrim-1, a working archiver with a measured result. The second is Cubrim-2. For me, this has become a family story: my wife Victoria is beside me, while I am developing the new direction with my daughter Ekaterina, an aerospace engineer and a coauthor of the Global Addresser idea.
Blending the two would be convenient, but it would not be honest. The archiver can already be tested. The addresser is still a set of hypotheses, limits, and experiments that must survive full-cost accounting, not merely an elegant formula.
First, what has actually been measured
In the current world benchmark across the Silesia, enwik8, and Canterbury corpora, Cubrim-1 ranks first in the overall weighted result among ten archivers. It finishes ahead of PPMd, xz, 7z, and Brotli. PPMd is the closest: its weighted ratio is 2.6% above Cubrim. Cubrim weighted output is 10.6% smaller than zstd, which was run strictly as --ultra -22.
This is an aggregate result, not a claim that Cubrim wins on every individual file. Its position varies from file to file. What matters to me about first place is not a slogan about being “best everywhere.” It is that an old idea has become a working system that can stand beside established archivers and be measured in the same table.
Cubrim-2 is not a new name for that result. It begins where the local job of an archiver ends.
From an archive to shared memory
A conventional archive must bring the receiver everything required to reconstruct a file. It may exploit structure inside the data very well, but it cannot assume that sender and receiver already hold a large common library.
An addresser asks a different question. What if both endpoints receive the same verified blocks in advance? A repeated fragment would not have to cross the channel again. A message could identify the required blocks, specify their assembly order, and carry only what is absent from the shared foundation.
I still find the image of two identical books useful. One person can send page and line numbers instead of dictating a whole passage. But the comparison remains honest only while the books really are identical. If the receiver does not have the required paragraph, a reference cannot conjure it into existence.
What we mean by a matrix
Our working name for the shared foundation is the Valentov Universal Data Matrices. It is not an infinite file containing all possible data, nor a hidden warehouse in which every future object already exists.
A practical matrix, if the idea survives testing, would be much less mystical: versioned immutable blocks, a catalog that establishes their identity, and rules for delivering them to particular devices. One spacecraft may need maps and models, another software components, and a third only a small subject-specific section.
A matrix is not useful because it is “universal” in a magical sense. It is useful only where data delivered once can be reused many times.
A short address does not erase the cost
The most seductive form of the idea says that an enormous file can be replaced with a tiny address. Stated that way, the promise is false. A fixed short code distinguishes a finite number of states; it cannot uniquely name every arbitrary long sequence without additional information.
If an object already exists in the shared structure, its address may be short. If only part of it matches, the residual still has to travel. If the data are absent, someone must deliver them and pay for their presence in the system.
That is why our experiments charge the complete cost rather than the length of an impressive reference: address + metadata + residual + matrix/catalog/distribution. Address, metadata, unique residual, catalog, matrix storage, and distribution are one problem. Omitting half of those terms does not compress the data; it hides them on another line of the report.
Why space makes the question practical
On Earth, an extra gigabyte is often an inconvenience. Between Earth and a distant spacecraft, it becomes time, energy, and risk. The channel is constrained, contact windows may be rare, and a retry does not return immediately.
If both endpoints already hold the same verified foundation, the channel could carry changes, new observations, and assembly instructions. Yet space also demands stricter engineering: the spacecraft must know the exact version of every block, verify integrity, and reconstruct the message autonomously. A missing block makes its address worthless.
This is where working with Ekaterina matters so much to me. I can spend a long time thinking about representations of data. She returns the discussion to the actual vehicle: what is already on board, what an update costs, what happens when contact is lost, and where autonomy ends.
Research, not a promise
We run the Global Addresser as a separate open track of testable hypotheses. Each one needs a mechanism, a measurable effect, an operating boundary, and a condition under which we must say that it does not work.
The deep-research phase is complete: all twenty-four hypotheses now have measured verdicts, and the final recommendation is to build a scoped prototype around two complementary mechanisms. This is not a finished universal product or proof of magical compression. It is a move from an attractive idea to an engineering programme with measured boundaries. Detailed results and the continuing work remain in the Global Addresser section on cubrim.com.
Negative results matter as much as positive ones. They show where a short address loses to catalog cost, where a shared matrix cannot repay its distribution, and where privacy rules out a scheme that may be technically possible.
I still want to see Earth and a distant spacecraft share a language for data. But I now care less about the beauty of that dream than about testing it without discounts. If such a language emerges, it will not be magical or free. It will be pre-positioned, versioned, verifiable — and useful only where repeated reuse genuinely repays its cost.