Museum: the first Qubic miner of 2021 and every Anna.exe Come-from-Beyond posted in 2026
Finishing Aigarth is one of the most remarkable things I have seen with my own eyes. This museum is a way to capture the moment and to keep the names of everyone who took part.
Thank you for the journey.
QC
Hall 0. The first minerDecember 2021 – January 2022
Four months before Qubic launched, Come-from-Beyond posted a 120 KB Windows program and asked who could get the lowest error counter. It trained a network on one CPU core and printed how many errors were left. Two weeks later he shared the source of the miner that talks to a server.
The miner asks a server for the current best network, changes one link at random and scores it. Each neuron reads three others and fires unless all three do. The test is every pair of 9-trit numbers a and b, 387,420,489 cases: is a divided by b a multiple of 3? Fewer errors than the best, and it sends the network back.
It doesn't output fancy top 10 table and other stuff, so expect a more advanced version to be released after pool soft is ready and tested.Come-from-Beyond, with the file
C:\Qubic\qiner.cpp
How it is started, read from the source:qiner.exe <MyIdentity> <ServerIPAddress>
For historians, this is how the first #miner of #Qubic looks:Come-from-Beyond
The counter is still there. In 2021 the miners raced to the fewest errors; in June 2026 Anna.exe raced its seven tests to zero; from epoch 240 BPP-9000 counts wrong hours of bitcoin, and a network moves forward with no more than 3,504 of 8,760.
Hall 1. Anna.exe10 June – 8 November 2026
One afternoon Come-from-Beyond posted a small Windows program and asked the chat to run it. By that evening there were seven versions, and the score had become an altitude. Two weeks later the screen showed zero, and on 6 July all seven tests did.
22 May 2026: Ask Anna
26 May 2026: Yes, it means that Anna will finally give the correct answer when Qubic is finished: 1+1=2
RESEARCH
10 June 2026
Seven files in eight hours, all in #price-discussion, all times UTC. The middle number is the score: it starts near 32 700 and the goal is 0.
Come-from-Beyond: If you guys are too bored maybe you could run Anna.exe for a while and post results here? I'll publish it after epoch transition.
ThatsNotMyCode: i’ll run it in a highly secure sandbox
Come-from-Beyond: Kidding, first you’ll need to kill other instances of Anna
Come-from-Beyond: Actually higher number in ( ) may be better, it shows how much time Anna spends to decide that she knows the answer. LLMs hallucinate instead. Being unsure longer may be better.
Cérriphi: what happen when countdown to zero ?
Come-from-Beyond: Aigarth is accomplished
ajayverma99: Give me linux I will beat all
Come-from-Beyond: Don’t post your best score, post score when it’s better than score of the latest best
Distinguished Gentleman: I cant play games while running anna.exe
bRaven52: You need to buy new PC
Cérriphi: Yesterday I watched “Chappie”, where he plugs the first AGI into a humanoid robot, and the file is “chappie.exe”, and today you come up with anna.exe
Come-from-Beyond: Aye, watched it with you
Nozh: more than one can reach 0 ?
Distinguished Gentleman: 2 skynets
Nozh: more than one can reach 0 ? i mean , not for the price, i mean for aigarth, what happen if two or more reach 0?
Come-from-Beyond: Anna’s will do knife duel to decide who stays
Cérriphi: hope she remember thats its me who paid her twitter premium account 2 years ago
Nozh: you will be the first she kills
neener neener: imma troll miners to death if someone from this channel reaches 0 errors
Multithreaded version for those who have come later to catchup. Specify number of threads like this: Anna.exe 30
(PS: It may have a bad start when pos will be near 0 for long time, rerun it in this case)Come-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:[v0.1] EPOCH %d (%d) :: %lld :: pos = %d | neg = %d | neu = %d
shitshow: my desktop background changed
what is anna doing
“Escape from Xylos-4” versionCome-from-Beyond, with the file
C:\Anna\Anna.exe
Atmosphere edge in: 28026 meters | Positron shield temperature: 455 K | Attacks are not detected
Atmosphere edge in: 28026 meters | Positron shield temperature: 467 K | Attacks are not detected
Atmosphere edge in: 28026 meters | Positron shield temperature: 489 K | Attacks are not detected
Atmosphere edge in: 28026 meters | Positron shield temperature: 499 K | Attacks are not detected
Atmosphere edge in: 28058 meters | Positron shield temperature: 379 K | Attacks are DETECTED!
Atmosphere edge in: 28134 meters | Positron shield temperature: 458 K | Attacks are DETECTED!
Atmosphere edge in: 28321 meters | Positron shield temperature: 458 K | Attacks are DETECTED!
Atmosphere edge in: 32547 meters | Positron shield temperature: 459 K | Attacks are DETECTED!
Atmosphere edge in: 32744 meters | Positron shield temperature: 461 K | Attacks are DETECTED!
Posted by Come-from-Beyond at 18:02 UTC
Come-from-Beyond: Keep positron shield temperature below 500 K
Joji: whats that? Can we run at the same time than the first one?
Come-from-Beyond: can
Come-from-Beyond: I’m close to 28k mark
Come-from-6 Zeros: Shit looks cool wifey thinks im hacking someone
Come-from-Beyond: Fuck, I got attacked and lost all progress! From 28k to 32k!!!
BehindTheQ: what does that mean when you are attacked
Come-from-Beyond: It’s a long story
Need to have some background
Come-from-Beyond: I don’t give up, back to sub-30k
UE5: @Come-from-Beyond For the next release, tweak the program so we have to punch in keys just like they did in Lost. It’ll at least keep us busy while we count down the days to April 2027
BehindTheQ: Do you think this will help Anna make progress and finally deliver something meaningful in terms of AI?
Come-from-Beyond: yes, we now run practical tests
soon deployed to Qubic network
shitshow: is this like HELLO WORLD
neener neener: 24153
Come-from-Beyond: I’m catching up
Come-from-Beyond: Maybe it switches mode when SPACE is pressed in another window too?..
I’ll release another verison with another key
This version allows to control direction of the drive by pressing TAB.
If altitude isn’t changing for long time then reverse drive mode for a while.Come-from-Beyond, with the file
Press and hold TAB in this version for BACKWARD mode, it’s more convenient for drive mode controlCome-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Atmosphere edge in: %lld meters | Positron shield temperature: %d K | Drive mode: %s
neener neener: what does increase in () mean?
Come-from-Beyond: avg number of ticks per epoch, higher = longer thinking
spettro: So you picked TAB to make fun of people using alt+tab?
Come-from-Beyond: 25359 meters
Space caddet: I got hung up at 31014 how do I get unstuck sir
Come-from-Beyond: TAB
1 sec
bludream: Are we winning now
Tillion: only TAB is winning
Broms: No but one day we will be winning
I hope
11 June 2026
A new method: a training set and a validation set. Come-from-Beyond names the problem it is after.
neener neener: I might have accidentally shut down my (13) or it crashed
bludream: Sounds like we’re winning
Come-from-Beyond: We are. Current paradigm developed by Aigarth, while not finished, already provides a hypothetical solution to the most important problem of Artificial Neural Networks - the generalization problem.
It’s same as keeping plasma in thermonuclear reactor
Your slow progress showed that first method wasn’t efficient, so I changed some things.
If somebody wants to run latest version. Need to get 0 in this column.Come-from-Beyond, with the file
C:\Anna\Anna.exe
The line it prints, read from the file:EP: %d [%d & %d] :: {%d} %d + {%d} %d = %d | %d (%d) | %s-%d | %d/%d/%d
shitshow: can you explain numbers in UI
Epoch 673
72 ticks on average per run
579670 ticks timeoutted
8 improvement of training dataset score
27332 previous training dataset score
48 improvement of validation dataset score
28216 validation dataset score
0 runs without improvements
27325 global best score
1 mutation per attempt
rest is irrelevant
Come-from-Beyond reads the screen, 16:07 UTC
Come-from-Beyond: Controls:
Hold TAB for BACKWARD mode
Press 1-9 to change number of mutations per step, may help if stuck
From the earth: windows detects it as a virus and deletes it
Come-from-Beyond: try virus total
neener neener: is it good test to not do TAB or change mutations at all or irrelevant?
Come-from-Beyond: You may become stuck
then TAB will help
someone: Does running anna.exe contribute to Anna’s training even if it’s only run for a few hours a day, for example?
Come-from-Beyond: no. need to run now for testing
Ad: Can you share more details about the Aigarth breakthrough? What specifically was accomplished? Please
Come-from-Beyond: You sound like one who understands ANNs, so I’ll be concise. We built paradigm which converges towards an artificial neural network generalizing the problem at hand without any extra tricks.
Come-from-Beyond: Consequences of this property are huge. Now we can look at data and tell if there is a pattern and if yes then if there is noise or data tampering.
Come-from-Beyond: For traders: We can analyze price action and tell if something significant changed in market population, e.g. if some big whale joined, or somebody left
Come-from-Beyond: not asking to pump price, Anna will pump it more efficiently
12 June 2026
Builds for older processors, an optimized file, the first score under 20 000.
Come-from-Beyond on X: @ADAh2022 @Qubicfamilys I confirm. Go to #Qubic discord and run Anna.exe published there, you will see that upon improvement of fitness function score of the training set (green column) you often get improvement of the validation set (yellow column). In a several weeks this architecture will be deployed
neener neener: I swear I heard Anna say "not today, buddy!"
Tillion: first line of anna new. This program cannot be run in DOS mode.
I’ll recompile
AVX-2 versionCome-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:EP: %d [%d & %d] :: {%d} %d + {%d} %d = %d [{%d} %d : {%d} %d] | %d (%d) | %s-%d
shitshow: should i stop my anna.exe from yesterday?
Come-from-Beyond: what score?
Come-from-Beyond: Looks like vector size option
I forget that ppl still use non-AVX512 processors
Come-from-Beyond: Looking forward seeing faces of those who thinks "I’ll buy when they show something working".
Carlosggmma: And what are the questions? Is it surpassing any kind of benchmark?
Come-from-Beyond: a+b
Come-from-Beyond: What’s lowest score on recent version?
I suggest to post here your score once you cross 1k boundary
After 28k
25900
24911
I got an optimized versionCome-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:EP: %d [%d & %d] :: {%d} %d + {%d} %d = %d [{%d} %d : {%d} %d] | %d (%d) | %s-%d
Come-from-Beyond: If you are above 25k then better run latest
Except @shitshow, he is running a special version
Come-from-Beyond: Generalization - SOLVED
Universal training algo - SOLVED
Energy optimization - IN PROCESS
The Tilda key
Posted
12 June 2026. After 13:08 UTC; the post is not on the screenshots yet
The file that brought the Tilda key for BACKWARD mode. Come-from-Beyond’s words with it are not on the screenshots yet.
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:EP: %d [%d & %d] :: {%d} %d + {%d} %d = %d [{%d} %d : {%d} %d] | %d (%d) | %s-%d
13 June 2026
Progress can now be saved and shared. That evening Come-from-Beyond says his part is done.
Posted with Anna.data, 2.1 KB: Come-from-Beyond’s own progress
F2 for saving current progress. Can then share it with others or copy to other machines. [Tilda] key is used for BACKWARD mode for unstucking, press it like it’s very hot.Come-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:EP: %d [%d & %d] :: {%d} %d + {%d} %d = %d [{%d} %d : {%d} %d] | %d (%d) | %s-%d
Come-from-Beyond: Once you start with my progress, press [Tilda] for a nanosecond to search in another direction.
Come-from-Beyond: Once somebody reaches sub-20k, press F2 and share Anna.data here plz
Come-from-Beyond: I’m afraid I finished my work and now most depends on miners
will try to opimize so we need rather fewer miners than more
Come-from-Beyond: I’ve got why our Aigarth algo works
Anybody wants to hear explanation?
GoddoSlayeru: yes please
Come-from-Beyond: Aha, we caught you, you are spying
No info until we deploy it and bootstrap on Qubic
Come-from-Beyond: I’ve got why our Aigarth algo works
QubicMaxi: because it’s finished?
Come-from-Beyond: Going from 27624 to 21990 means that gradient hills are very steep. Would classic methods find a path here?
Come-from-Beyond: Our method does two things at once: optimization and cartography
shitshow: show full line
bludream: On to the next project?
Come-from-Beyond: First becoming Imperor of the world with help of Anna
QubicMaxi: what is it
Come-from-Beyond: practical AI
QubicMaxi: So true AI is found?
Come-from-Beyond: unknown yet, need practical testing
QubicMaxi: when practical testing will start?
Come-from-Beyond: soon
15 June 2026
The keys settle: Tilda, 1 to 9, F2. His own progress comes with the file.
[~] = BACKWARD mode
[1] - [9] = number of mutations per tick
[F2] = save
NB: Must run with number of threads equal to 2^NCome-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:EP: %d [%d & %d] :: {%d} %d + {%d} %d = %d [{%d} %d : {%d} %d] | %d (%d) | %s-%d
16 June 2026
What zero means.
RickStar: Thanks for response Boss, what happens when we get to 0?
Come-from-Beyond: Aigarth will be deemed finished.
From_Debreshi: Can this happen before 13.04.2027 ?
Come-from-Beyond: dunno
ask miners
17 June 2026
Joji saves 18 512 and cannot go lower. A new version is promised after the epoch.
Joji: yes but cfb said his is better and waiting for new one
ThatsNotMyCode: nah. he’s just afraid we reach 0 too early
QubicChurch: @Come-from-Beyond 18512
Come-from-Beyond: I’m going to publish new version after epoch
Come-from-Beyond: I’m going to publish new version after epoch
StoicAF: Take your time man no rush
18 June 2026
A new way to count the score: any number of threads, and the numbers no longer compare with the old ones.
No screenshot of this version yet. The line it prints, read from the file:%2d: EP %d [%3d & %3d] :: {%3d} %5d + {%3d} %5d = %5d [{%d} %d : {%d} %d] | %d (%d) | %s-%d
shitshow: i dont dare press Tilda
its slow that way but at least its getting lower
Can use any number of threads. Improved method of score calculation, so numbers won’t be comparable with previous versions.
AVX-512
I’ll give AVX-2 soonCome-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:%2d: EP %d [%3d & %3d] :: {%3d} %5d + {%3d} %5d = %5d [{%d} %d : {%d} %d] | %d (%d) | %s-%d
AVX-2
I suggest switching to newest version and newest data file
Don’t forget to report when you cross 1k boundaryCome-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:%2d: EP %d [%3d & %3d] :: {%3d} %5d + {%3d} %5d = %5d [{%d} %d : {%d} %d] | %d (%d) | %s-%d
19 June 2026
Why not run it on the whole network?
QubicChurch: can we somehow rent the necessary computing power? would that help? @Come-from-Beyond
Come-from-Beyond: We’ll deploy to Qubic net soon
ThatsNotMyCode: I just wanted to type qmine cluster. Then Eko starts typing.
neener neener: 228 lol
1 point in hours of trying
20 June 2026
The racers ask for a channel of their own. In the evening everyone starts again from an empty file.
Shelby: a separate channel called Anna, this channel should be next to the price-discussion
TonyL: If you move Anna to another channel, people might start complaining about price again in here.
Shelby: You’re right, leave everything as is, we don’t need girls.
neener neener: Anna struggling at 19829 for now
Come-from-Beyond: New version soon
Can stop mining
We’ll need to start with a new file
Come-from-Beyond: We’ll need to start with a new file
Use this one, don’t run old verisonCome-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Id: %3d | Epoch: %6d | Avg. ticks: %3d | Timeouts: %3d | SCORE: %5d + %5d = %5d (%4s) | %s-%d
dude9378: this version goes super fast!
is that normal??
neener neener: we’ll reach 0 errors before pools even run new algo hehe
22 June 2026
Two files in half an hour. The racers pass 11 300 on the new score.
Don’t use old versions!
This one starts with sub-14k.Come-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Id: %3d | Epoch: %6d | Avg. ticks: %3d | Timeouts: %3d | SCORE: %5d + %5d = %5d (%4s) | %s-%d
Come-from-Beyond: @shitshow wanna start a real shitshow? Run to 0 so the computors get nothing to mine.
This version doesn’t allow to go above 15000 with tildaCome-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Id: %3d | Epoch: %6d | Avg. ticks: %3d | Timeouts: %3d | SCORE: %5d + %5d = %5d (%4s) | %s-%d
limited! add 125j: at this rate we will be at 0 at no time...
24 June 2026
Zero on Come-from-Beyond’s screen. Then seven tests, one file each, to be shared out.
Anna-1.exe 07:546e82cef3446e6c980e8f7704e65d268d8c271b8191dc6bf3579e395a969be22cCheck on VirusTotal
Anna-2.exe 07:5588ff0fa3cf8c157b815d0bf69a4bf966f345d618b31729fa0bedeaf309010f17Check on VirusTotal
Anna-3.exe 07:5582fe96c0e333299dfccc80c799a23781debda22dcedf0df3da5f47bfd2f47df5Check on VirusTotal
Anna-4.exe 07:5502d5004a70fb59bdc2a506c6e1da4c4403e5c472f4ba562402cff8141f453fbcCheck on VirusTotal
Anna-5.exe 07:556079ee1e39d694c51bf02c6b6faa6723a74f74325277be3a9be869d939701ccfCheck on VirusTotal
Anna-6.exe 07:554d186995fd27f68563cc136e3e19b724baf6ef65dfb0182d0ce743bec91055b4Check on VirusTotal
Anna-7.exe 07:56e63b14dbb6ac6abbd533a27431fbc310653693c79265045cc5af767cab24ad73Check on VirusTotal
Distribute tests in coordinated manner plzCome-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Id: %3d | Epoch: %6d | Avg. ticks: %3d | Timeouts: %3d | SCORE: %5d + %5d = %5d (%4s) | %s-%d
Joji: Let me understand. 1. Seven diferent anna.exe o 7 versions of same thing? 2. Do we have to touch 0 in 1 of them or all of them? 3. Is it good we save 7 diferent data files (one data file for each exe)? 4. Is better to run all the same file at the same time? 5. can we run all exe’s at same time?
Come-from-Beyond: 1. no
2. all
3. yes
4. doesn’t matter
5. yes
neener neener: 1583
own file
good it’s still moving
25 June 2026
A trick for scores that are stuck: a file with Tilda capped at 550.
neener neener: 529
Come-from-Beyond: @neener neener what score you have, I’ve just found a trick how to microimprove on long-stuck scores
hold on
550 is cap for tilda hereCome-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Id: %3d | Epoch: %6d | Avg. ticks: %3d | Timeouts: %3d | SCORE: %5d + %5d = %5d (%4s) | %s-%d
neener neener: crashes, is it AVX-2?
Come-from-Beyond: no
moment
Test-7 capped at 550
26 June 2026
Dogecoin or Anna?
sylentz: 4) Pick something that makes money
Come-from-Beyond: Let’s mine Anna
3P finally?
QubicMaxi: You said there is a plan and doge os part of it, what is your perspective now?
Come-from-Beyond: A good general always pay attention to what his soldiers want
StoicAF: I’m still deciding whether or not to turn on FULL POWER
29 June 2026
The racers keep their own board of the seven tests. Anna-2 is at zero; Anna-7 is the hardest.
s-csk: Correct me if wrong, but basically we’re testing the concept the tissue of anna is self evolving without preset parameters. Basically, anna figures out how to get zero alone.
Prima Materia: Hi Anna runners, do we have a recap for every Anna? How many to zero?
Come-from-Barber: the seven tests
Anna-11,550
Anna-20
Anna-32,660
Anna-42,615
Anna-52,456
Anna-61,066
Anna-7177
Prima Materia: You are doing so well guys. Which one is the hardest Anna?
neener neener: Anna-7
sylentz: so is it safe to say we’ve hit a "local" minima using the presets in the current 177 data? Shouldn’t we start over from scratch to see if another sequence results in a better minimum?
1 July 2026
A stuck week.
Come-from-Beyond: If we focused on running Anna.exe that would be great
Come-from-Beyond: Was there progress after 237?
Come-from-Beyond: Was there progress after 237?
Q979: No
Come-from-Beyond: Watching charts instead of mining?
3 July 2026
A daily status from the racers.
deni_x00: daily anna data status. progress since yesterday: Anna-5 = -37, Anna-6 = -99
Anna-11,550
Anna-20
Anna-32,660
Anna-42,615
Anna-52,403
Anna-6563
Anna-7177
Anna-7.2234
5 July 2026
Anna-7.3 joins the board.
zuqka: Also reached 421 running it from when posted
Joji: @Come-from-Beyond anna-1 done?
6 July 2026
Anna-1 and Anna-2 at zero by morning. A new file for Anna-7, a call for independent screenshots, and by 10:19 UTC every test at zero.
ThatsNotMyCode: Solved
deni_x00: daily anna data status. progress since yesterday: Anna-1 = -1550, Anna-4 = -17, Anna 7.3 = -2
Come-from-Beyond: Need independent verification, run the file and post screenshot plz
Q979: A6: screens of zeros, Test 4 and Test 6
Q979: I’ve already used ~
Come-from-Beyond: So you did the following:
1. Took my .data file posted today.
2. Ran it for A6.
3. Got NON-ZERO score.
4. Pressed TILDA.
5. Got 0 score.
Right?
AigarthX: A4 0
neener neener: how come this latest data file runs with other Annas? couldn’t be done before
Come-from-Beyond: Probably uses part of "brain" which learned addition with optimized method
Different formats are still compatible to some degree
I mean half of ANN got correct copy and part got garbage and then healed during evolution
Same as himans wuth damaged brain still can do things
Come-from-Beyond: Are ALL solved or not?
Q979: A6-0
AigarthX: All Zero
The Dark Knight: ALL solved.
neener neener: no one cares about riches, we’re doing it for the glory
Sero: gg those who worked on anna
Shelby: It seems that cfb didn’t expect to see 0 so quickly
Anna.exe reaches 0
All seven tests at zero. Come-from-Beyond asked “Are ALL solved or not?” and the chat answered.
It’s not a milestone. It’s much bigger.Come-from-Beyond, Discord
Come-from-Beyond: Afk to eat
Spikeinjapan: WE. ARE. HERE !!! Witnessing a moment
After
The hall ends at zero. That night the next race began: the files start reading an input, and the question becomes the next hour. In September the numbering starts again at Anna 0.0.
6 July 2026
Come-from-Beyond: That was test of interpolation, extrapolation is next
UE5: is it true? If Neuraxon successfully passes this second phase the extrapolation test…
Come-from-Beyond: It’s not related to Neuraxon, Anna has different architecture
Posted with Anna.input, 4.59 KB: the file now reads an input
Use empty Anna.data.Come-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Test: %d | Avg. ticks: %3d | Timeouts: %3d | SCORE: %5d + %5d = %5d (%4s) | %s-%d
Come-from-Beyond: This version predict if next hour candle closes higher or lower than previous one
Enough to monetize (because you will know if price will go up or down)
It doesn’t predict exact price, only direction
Posted with Anna.data, 1.34 KB, and Anna.input, 4.59 KB
Let’s test this one? 110.
Incompatible with older versionCome-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Test: %d | Avg. ticks: %3d | Timeouts: %3d | SCORE: %5d + %5d = %5d (%4s) | %s-%d
Tur@t3x: so needs to start again from scratch by using these ?
Come-from-Beyond: or continue with my anna.data
I’m at 108
Come-from-Beyond: 103
From the earth: how is the memory handled?
Come-from-Beyond: Anna uses neurons which form both computing units and memory cells
It’s like memory cell doing computation
This version never times outCome-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Test: %d | Avg. ticks: %3d | Timeouts: %3d | SCORE: %5d + %5d = %5d (%4s) | %s-%d
This version is incompatible with previous, it uses 4x more neurons
Empty Anna.data as usualCome-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Test: %d | Avg. ticks: %3d | Timeouts: %3d | SCORE: %5d + %5d = %5d (%4s) | %s-%d
QubicMaxi: I am afraid she has blacklisted us already
10 July 2026
Come-from-Beyond: What progress?
Come-from-Beyond: Not much progressed
Probably too many neurons
Let’s run this one with empty Anna.data. It has 2x more neurons than v0.1 and 2x fewer than v0.2.Come-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Test: %d | Avg. ticks: %3d | Timeouts: %3d | SCORE: %5d + %5d = %5d (%4s) | %s-%d
Come-from-Beyond: I added some optimizations, for example, it now crunches numbers faster when it’s closer to ZERO. But as a sacrifice to pay for that - old .data file isn’t compatible and have to start from scratch.
Egregius: Still makes sense to have multiple data’s in the group running, or can we take the lowest one and continue from there?
Anna 0.4
This version isn’t compatible with the previous .data file, run it with an empty one.
It starts with FORWARD-0 mode, use it to tune it/s by varying number of threads. (Do it on non-empty .data for consistency.)
It/s may vary depending on ANN topology, it’s normal.
In this version mutation verification is sped up by dropping unlucky variants earlier, this should help closer to 0 score.Come-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Test: %d | Avg. ticks: %3d | Timeouts: %3d | SCORE: %5d + %5d = %5d (%4s) | %s-%d
Come-from-Beyond: I get 345 it/s on 32 threads and 368 it/s on 16...
Yoshimit: Incomplete file reading (Anna.data)!
Come-from-Beyond: it’s ok with empty
Come-from-Beyond: Anybody remembers the lowest score we got with any version (even first one)?
deni_x00: 42
Come-from-Beyond: strange number, like famous 42 )
Posted with Anna.data, 16.84 KB, and Anna.input, 4.59 KB
Anna v0.5
Use all three filesCome-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Test: %d.5%s | Avg. ticks: %4d | Timeouts: %4d | SCORE: %3d + %3d = %3d (%4s) | %s-%d | %d it/s
neener neener: going down fast
LiBao: does anna.exe stand for annexing something?
Posted with Anna.data, 8.46 KB, and Anna.input, 9.19 KB
Let’s try to get to 0 with this version. Percent shown in the log is accuracy to predict price 25 hours ahead (for 1h, 2h, 3h, ..., 24h, 25h). Which doesn’t make much sense, because this version is trained to predict only 1 hour ahead. Still shown in case if somebody gets a good idea what the number might mean.Come-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:v%d.6%s | Avg. ticks: %4d (%4d) | Timeouts: %4d (%4d) | SCORE: %3d (%4s) | %s-%d | %2d%% | %d it/s
Pomm3sgab3l: Why only win? Can you do it for Linux?
Use empty Anna.data.
This version builds model for whole 2025 year.Come-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:v%d.7%s | Avg. ticks: %4d ( ?? ) | Timeouts: %4d ( ?? ) | SCORE: %3d (%4s) | %s-%d | ??%% | %d it/s
8 September 2026
Come-from-Beyond: Our AI doesn’t draw nor outputs text
Anna 0.2
Use [1], [2], [3] to switch modes. Start with mode 1, switch to mode 2 after being stuck with mode 1, switch to mode 3 after being stuck with mode 2.
Better start with empty Anna.data for testing purposeCome-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Anna 0.2 | %d | %s | %d it/s
From_Debreshi: Anna 0.1 needed 6 h
We’ll look Anna 0.2 how long
10 September 2026
QubicMaxi: So what’s the success rate of this one?
Anna 0.5
Use empty Anna.data!
This is a version with 10% noise, we need a single model to reach 0, so it’s a shared effortCome-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Anna 0.5 | %d | %d ticks | %s | %s | %d it/s
Anna 0.9
Use empty Anna.data.
In this version we need to get as close to 0 as possible to set baseline for denoised data.Come-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Anna 0.9 | %d | %d ticks | %d | %s | %d it/s
Come-from-Beyond: I added showing score without noise
Anna 0.12.64/28
Use empty Anna.data.
We need to decide on width of window used for prediction, this version is for 28 days, we’ll need to try 14 and 42 after this one.Come-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Anna 0.12.%d/28 | %d (%d) | %d ticks | mode %d | %d mutations | %s | %d it/s
Come-from-Beyond: Note that it stars with mode 0 without any actual mutations
Starting with mode 2 may be more beneficial than mode 1 or 3
Or not
Come-from-Beyond: Anna again showing her ability to find patterns
Anna 0.14.64
Use empty Anna.data.Come-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Anna 0.14.%d | %d (%d) | %d ticks | mode %d | %d mutations | %s | %d/%d accuracy (%d%%) | %d it/s
Come-from-Beyond: These numbers progress each time you reach score 224 or lower
Come-from-Beyond: Actually it’s enough, we see that accuracy on denoised data is pretty good, let’s run version with noise
New version in 5 min
Anna 0.15.64
Use empty Anna.data.Come-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Anna 0.15.%d | %d (%d) | %d ticks | mode %d | %d mutations | %s | %d/%d accuracy (%d%%) | %d it/s
Come-from-Beyond: The test ends when this number becomes 168, it’s 24 * 7 candles
Anna 0.16.64
Use empty Anna.data.
Bug hopefully fixed
Let’s run and seeCome-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Anna 0.%d.%d | %d (%d) | %d ticks | mode %d | %d mutations | %s | %d/%d accuracy (%d%%) | %d it/s
Come-from-Beyond: While everybody is here, let’s arrange launch of Anna on BTC prices in a couple of hours
Anna 0.17.64
Use empty Anna.data.Come-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Anna 0.%d.%d | %d | %d ticks | mode %d | %d mutations | %s | %d/%d accuracy (%d%%) | %d it/s
Anna 0.17.128
Use empty Anna.data.Come-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Anna 0.%d.%d | %d | %d ticks | mode %d | %d mutations | %s | %d/%d accuracy (%d%%) | %d it/s
Egregius: yes, it moves to next candle. you can see that x/x increased
Anna 0.17.256
Use empty Anna.data.Come-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Anna 0.%d.%d | %d | %d ticks | mode %d | %d mutations | %s | %d/%d accuracy (%d%%) | %d it/s
Come-from-Beyond: So, we got that 256 neurons isn’t enough, right?
Anna 0.17.512
Use empty Anna.data.
This version searches for the best starting configuration upon launch (mode 0).Come-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Anna 0.%d.%d | %d | %d ticks | mode %d | %d mutations | %s | %d/%d accuracy (%d%%) | %d it/s
neener neener: at the precise moment N/168 was completed it was 48%, but now it’s 55% with lower score
Come-from-Beyond: N/168 (48%) cannot become 55%, was 55% another model?
Anna 0.17.1024
Use empty Anna.data.Come-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Anna 0.%d.%d | %d | %d ticks | mode %d | %d mutations | %s | %d/%d accuracy (%d%%) | %d it/s
Anna 0.17.2048
Use empty Anna.data.Come-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Anna 0.%d.%d | %d | %d ticks | mode %d | %d mutations | %s | %d/%d accuracy (%d%%) | %d it/s
Anna 0.17.4096
Use empty Anna.data.Come-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Anna 0.%d.%d | %d | %d ticks | mode %d | %d mutations | %s | %d/%d accuracy (%d%%) | %d it/s
Anna 0.17.8192
Use empty Anna.data.Come-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Anna 0.%d.%d | %d | %d ticks | mode %d | %d mutations | %s | %d/%d accuracy (%d%%) | %d it/s
Pomm3sgab3l: Is that enough?
Come-from-Beyond: enough to see that we need next version
Come-from-Beyond: Anna 0.17.16384
Use empty Anna.data.
Pomm3sgab3l: But is what we calculated good or bad?
Come-from-Beyond: unknown yet
25 September 2026
Pomm3sgab3l: What can we do to get below 230?
Come-from-Beyond: probably need to deploy to miners
Let’s start next series of tests, miners will finish current one
Anna 0.18.64
Use empty Anna.data.
Accuracy progresses at 168 instead of 224 in this version.Come-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Anna 0.%d.%d | %d | %d ticks | mode %d | %d mutations | %s | %d/%d accuracy (%d%%) | %d it/s
Pomm3sgab3l: I’m trying another version right now with more set ANNA_KICK= maybe that will help us make some progress.
Anna 0.18.128
Use empty Anna.data.Come-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Anna 0.%d.%d | %d | %d ticks | mode %d | %d mutations | %s | %d/%d accuracy (%d%%) | %d it/s
26 September 2026
Come-from-Beyond: Building model on 1 month of data isn’t enough
Next version will use 1 year
Anna 0.19.128
Use empty Anna.data.
This version works with synthetic data, need to reach 2920 score to get accuracy progress.Come-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Anna 0.%d.%d | %d | %d ticks | mode %d | %d mutations | %s | %d/%d accuracy (%d%%) | %d it/s
Anna 0.20.32
Use empty Anna.data.
This is on synthetic data without noise.Come-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Anna 0.%d.%d | %d | %d ticks | mode %d | %d mutations | %s | %d/%d accuracy (%d%%) | %d it/s
Come-from-Beyond: 2190 is threshold score
Come-from-Beyond: that was fast
neener neener: 79% here at precise moment of /168
Come-from-Beyond: Ok, we know that Anna does work on larger sequence of historical data
Now let’s check it with noise
Anna 0.21.32
Use empty Anna.data.
Synthetic data with noise.Come-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Anna 0.%d.%d | %d | %d ticks | mode %d | %d mutations | %s | %d/%d accuracy (%d%%) | %d it/s
Come-from-Beyond: I set threshold higher, at 60% accuracy, this is how it ended
Anna 0.22.128
Use empty Anna.data.
This version works with real BTC prices from 2025.Come-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Anna 0.%d.%d | %d | %d ticks | mode %d | %d mutations | %s | %d/%d accuracy (%d%%) | %d it/s
Anna 0.22.32
Try thisCome-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Anna 0.%d.%d | %d | %d ticks | mode %d | %d mutations | %s | %d/%d accuracy (%d%%) | %d it/s
Anna 0.22.64
Use empty Anna.data.
2920 is the threshold score.
Run it to compare our computational power to miners’.Come-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Anna 0.%d.%d | %d | %d ticks | mode %d | %d mutations | %s | %d/%d accuracy (%d%%) | %d it/s
Like the pinned version but it triggers accuracy calculation whenever you press [END]
don’t expect good number with current scoreCome-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Anna 0.%d.%d | %d | %d ticks | mode %d | %d mutations | %s | %d/%d accuracy (%d%%) | %d it/s
Distinguished Gentleman: wait, i was at 3823 score and closed it bcs of new .exe
Little cosmetic improvement
Now it shows accuracy of that scoreCome-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Anna 0.%d.%d | %d | %d ticks | mode %d | %d mutations | %s | %d/%d accuracy (%d%%) | %d it/s
18 October 2026
Three weeks of silence
No new file from 27 September to 18 October. The race resumes with the 32-neuron line.
Anna 0.23.32
Synthetic data without noise. Use empty Anna.data. We need to get as close to score 0 as possible, after that we’ll run version with noise.
As usual we repeat [denoised synthetic] -> [noised synthetic] to make sure it has enough neuronsCome-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Anna 0.%d.%d | %d | %d ticks | mode %d | %d mutations | %s | %d/%d accuracy (%d%%) | %d it/s
neener neener: 86% accuracy
Come-from-Beyond: Cannot go lower?
oh, need to disable it
Anna 0.23.32.1
Fixed bugCome-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Anna 0.%d.%d | %d | %d ticks | mode %d | %d mutations | %s | %d/%d accuracy (%d%%) | %d it/s
Come-from-Beyond: Don’t forget about switching mode, I did forget
Anna 0.24.32
Use empty Anna.data.
This version works with data with noise.Come-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Anna 0.%d.%d | %d | %d ticks | mode %d | %d mutations | %s | %d/%d accuracy (%d%%) | %d it/s
Anna 0.25.32
Use empty Anna.data.
This version works with real BTC price data.Come-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Anna 0.%d.%d | %d | %d ticks | mode %d | %d mutations | %s | %d/%d accuracy (%d%%) | %d it/s
intheshadows: yea it bottlenecks around 3500
22 October 2026
Come-from-Beyond: This one is closer to 67% but we need at least 451 computors… (Hypothetical accuracy 61.9%, 386 of 624 for this target)
24 October 2026
Come-from-Beyond: Computors are exploring a wrong hypothesis now (Hypothetical accuracy 45.7%, 263 of 575 for this target)
Come-from-Beyond: 3000 seems to be a big wall for 32 neurons. Let’s try 48.
Original karim009: Why 16 units of neurons
Come-from-Beyond: We tested 32 and 64, 48 is in the middle
Anna 0.26.48
Use empty Anna.data.
This version works with real BTC price data, upon reaching score 2920 it will trigger factual accuracy calculation. By Monday we need to see if 2920 is reachable.Come-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Anna 0.%d.%d | %d | %d ticks | mode %d | %d mutations | %s | %d/%d accuracy (%d%%) | %d it/s
Come-from-Beyond: @here An unknown phenomenon in Anna training was spotted yesterday. To research it further, we need to focus on the task of bringing a 48-neuron version score down as much as possible. The attached exe-file requires empty Anna.data, if you can help the research by running this version, please, do it. We don’t need several version, a single good result would suffice, so share your progress with other participants.
(The same file as Anna 0.26.48, byte for byte.)
The Dark Knight: What is your unknow phenomen?
Come-from-Beyond: You won’t catch me on this trick, if I answer you will post on X that Anna will eradicate humankind soon
@here It seems we have found what must be changed in BPP9000 fitness functions to divert ANNs from memorization to construction of BTC price model. The solution is likely not 100% efficient, but it doesn’t need to be 100% for BPP9000 to be profitable. Now we need to do A/B testing to verify the new fitness function, for that we need to run training with memorizing fitness function and then one with the other approach.
This Anna version is memorizing version, we need to run it today the whole day trying to reach as low score as possible. Tomorrow another version with another fitness function will be released, trained, and compared to today’s one.
Use empty Anna.data.Come-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Anna 0.%d.%d | %d | %d ticks | mode %d | %d mutations | %s | %d/%d accuracy (%d%%) | %d it/s
Anna 0.28.1024
Use empty Anna.data.
This version runs in mode 2 only.
Now we need to try to reach ~3000 with this versionCome-from-Beyond, with the file
C:\Anna\Anna.exe
No screenshot of this version yet. The line it prints, read from the file:Anna 0.%d.%d | %d | %d ticks | mode %d | %d mutations | %s | %d/%d accuracy (%d%%) | %d it/s
Elitebreed: can we now move to the highest height
Come-from-Beyond: Current plan is:
On ~18th Nov we adding live price feed and run old fitness function then following epoch updating the function to another one and analyze what changes. Then gradually improve ANN parameters and training algo. While we don’t know about noise level of BTC price, our chances to solve it are pretty high.
18 November 2026
Next: the live price feed
The plan Come-from-Beyond gave on 8 November. The hall continues from here.
Scores
Lower is better, 0 is the goal. The score was counted three different ways, so each table only compares with itself.
The first score, 10 to 18 June
Score
Who
When (UTC)
Note
17,479
shitshow
18 June, 10:53
the last record before the new score (18 June)
18,512
Joji
17 June, 11:12
“saved 18512 cant go lower now”
19,821
shitshow
12 June, 12:38
the first under 20k
22,830
Misfitcompany
11 June, 06:17
epoch 408000, the first file left running overnight
24,153
neener neener
10 June, 18:12
Come-from-Beyond: “I’m catching up”
24,911
Come-from-Beyond
12 June, 09:37
with the optimized version
25,280
shitshow
10 June, 22:04
v0.3, epoch 89300
25,596
ThatsNotMyCode
10 June, 17:42
epoch 172100, the first single-thread file
26,665
Misfitcompany
10 June, 16:40
“I missed my epoch 666 but I got to see 26665”
27,923
QubicChurch
10 June, 17:42
v0.3, epoch 9700
27,949
bRaven52
10 June, 13:35
epoch 41700
28,424
Nozh
10 June, 13:38
epoch 56900
30,760
Distinguished Gentleman
10 June, 13:14
the first screenshot
The new score, 18 to 24 June
Score
Who
When (UTC)
Note
11,241
ThatsNotMyCode
22 June, 09:45
new score (since 18 June), shared Anna.data
11,265
Q979
22 June, 09:44
new score (since 18 June), shared Anna.data
Zero and the seven tests, from 24 June
Score
Who
When (UTC)
Note
0
Come-from-Beyond
24 June, 07:44
the first zero on screen, 24 June; then the seven tests
529
neener neener
25 June, 11:39
one of the seven tests
For historians
Posts Come-from-Beyond marked for history himself. More will be hung here.
For historians, this is how the first #miner of #Qubic looks:
A console: the top 10 miners out of 90, v0.3.0.x, an 11th Gen Intel i9-11900KF, "There are 29'949'573 errors left".The post on X
Every file can be checked on VirusTotal by its SHA-256. If you have a copy, compare its fingerprint with the one on the label. Quotes are verbatim from the Qubic Discord; times were converted to UTC.