880
Six disk sides, 256,256 bytes each. As far as I can establish they are the only bytes of this machine that anyone still has, and they exist because a crawler happened to take a copy of a file in 2016.
What is on them
Three of the six are boot sides and carry no files at all. Two are application sides and carry everything: a database suite from Riga and Leningrad, and the stipend payroll of a vocational school. The sixth is empty, and I have a page for it too, because an empty side is evidence.
Every catalog above can be read out again with the same tool I used, and the programs that are on those two application sides are in the repository as decoded source.
- iskra_basic.py, the catalog reader and the lister
- listings/, 5,842 lines of somebody else's software, decoded
- disk-survey.md and catalogued-sides.md
The three boot sides
A boot side has no catalog. It starts at byte zero and holds four slots of 63,744 bytes, and each slot opens with a 32 byte header: three words of entry vector, then eighteen characters of signature with a build date inside them. For three weeks I read those four slots as four writings of one interpreter and averaged them, which yields a version of this machine that never ran anywhere. They are not copies. They are a version library, and taken apart properly the three sides give eight distinct builds spanning 1981 to 1986, each byte-identical wherever it repeats.
None of that was a guess in the end, and I found out too late to save myself the three weeks. Balasanian, Bogdyukevich and Shakhverdov published a programming textbook for this machine in Moscow in 1987, and where it explains how the machine comes up it says this: Интерпретатор занимает на диске 249 подряд идущих секторов. Версия с номером 0 записывается в секторах с 0 по 248, версия с номером 1 с 249 по 497. The interpreter takes 249 consecutive sectors. Version number 0 is written in sectors 0 to 248, version number 1 in sectors 249 to 497, and so on.
A sector here is 256 bytes, so 249 of them is 63,744, and the versions begin at bytes 0, 63,744, 127,488 and 191,232. Those are the four offsets I had already measured off the headers in the images, and versions with numbers is exactly what a version library is. A textbook printed in 1987 and a byte count taken in 2026 arrive at the same four addresses and the same reading of what sits at them, and neither had to assume the other. It is the strongest piece of evidence in this project, it says in one sentence that averaging the slots was never going to work, and it was sitting in a book on my own shelf the whole time.
Every line in the tables below is read out of the image itself, so they cannot drift away from the files you can download.
| slot | offset | entry vector | signature | sha1 |
|---|---|---|---|---|
| 0 | 0 | 9006 9009 9007 | BASIC PL5 30.09.84 | 1a567d8bebe8d3ff |
| 1 | 63,744 | AD43 8092 ACFB | BASIC 01 12.07.82 | 4cfdc2e8ec548a34 |
| 2 | 127,488 | AD43 8092 ACFB | BASIC 01 27.12.82 | 0c800ca16c04a092 |
| 3 | 191,232 | AD43 8092 ACFB | BASIC 01 15.12.81 | 3cfc812ac70e6f3b |
| slot | offset | entry vector | signature | sha1 |
|---|---|---|---|---|
| 0 | 0 | 9006 9009 9007 | BASIC 02 22.04.84 | 59197487d68dfcdf |
| 1 | 63,744 | 9006 9009 9007 | BASIC 02 10.10.84 | aaee3ef1319eea17 |
| 2 | 127,488 | 9006 9009 9007 | BASIC 02 10.09.86 | f298c2ac2825797e |
| 3 | 191,232 | 9006 9009 9007 | BASIC 02 16.12.83 | bd9d46d22aae73f3 |
| slot | offset | entry vector | signature | sha1 |
|---|---|---|---|---|
| 0 | 0 | 9006 9009 9007 | BASIC PL5 30.09.84 | 1a567d8bebe8d3ff |
| 1 | 63,744 | 9006 9009 9007 | BASIC PL5 30.09.84 | 1a567d8bebe8d3ff |
| 2 | 127,488 | 9006 9009 9007 | BASIC PL5 30.09.84 | 1a567d8bebe8d3ff |
| 3 | 191,232 | 9006 9009 9007 | BASIC PL5 30.09.84 | 1a567d8bebe8d3ff |
- The eight builds, one file each, under the dates they carry
- firmware_extract.py, which cuts them out of a boot side and is what produced the tables above
- boot-sides.md, including the correction, because the wrong reading is worth keeping
I wrote about this at the time
Where the bytes come from
The archive is called 880.rar. It was posted on oldpc.su by a forum user called vazman, who owned the physical disks. dk_spb read them and made the images. The thread is on phantom.sannata.org and dates from 2016. The download link is dead. The copy I work from came out of a web archive snapshot from that year, and as far as I can establish it is the only one left anywhere.
Neither of them owes me anything and both of them are the reason any of this exists. If you mirror these files, mirror their names with them. All six images are in the repository, and disks/README.md says where each one came from and what I have not opened.
The one thing nobody knows
How dk_spb produced these images in 2016 is written down nowhere I can find. Real СМ or ЕС hardware with a working eight inch drive? A Catweasel, a KryoFlux, ImageDisk? At what sampling, in what format, with how many passes over a bad track? In 2022 he was still asking other people about this disk format himself.
So the position is this. The only surviving bytes of this machine exist, they are complete enough to run software on, and nobody now knows how they were read off the plastic. I would rather leave that standing as a question than fill it with a plausible answer.
The other thing that did not come down
This machine is supposed to have had a UNIX. A group in Kiev is said to have built one on a state planning contract, with a hierarchical file system, a shell and a C compiler they wrote themselves, and the story is told well enough that it gets repeated as established. It rests on one thing: a memoir by somebody who says he was there, written down long afterwards and written with an eye for a good scene. There is no source, no binary, no image, no listing, no manual and no paper. It is not on any code host, in any archive, in any museum, in any emulator project or in any catalog I can search. The Wikipedia sentence people cite back at me leads to a single magazine article about the machine, not about the system, so it confirms nothing on its own. And the six sides on this page, the only sides anyone has, do not contain it.
There is a trap sitting right next to it. The one paper ever published about an operating system for the Iskra-226 came out of Tallinn in issue 6 of Микропроцессорные средства и системы for 1986, on page 21, and what it describes is a port of CP/M-80 to the i8080 clone in one of the machine's interface blocks. Not a UNIX, and it never said it was. The two get folded together constantly, and once they are folded the UNIX looks corroborated.
So it goes here, beside the loader question, in the same condition: told, unverified, and not repeated as fact by me. Things come down to us by luck. The disks got some and this did not, and being able to say that precisely is worth more than the anecdote would have been.