Brainfuck grows a primordial soup, Malbolge gets graded, and qbase does the boring work (2026)

Somebody is breeding self-replicating programs in Brainfuck, somebody else has put Malbolge on a graded ladder, and qbase is assembling the boring libraries that make Q usable. Those are the three projects that earned the spotlight from the repository-creation window running July 19 at 20:00 through August 2 at 20:00, UTC-05:00. 1 2 3

The first is absurd in the best possible way. The second turns an almost unusable language into an evaluation benchmark. The third is the awkwardly practical one: a Q library stack with bootstrapping, logging, networking, examples, and tests. The snapshots below are what GitHub showed during the scan; stars and forks are totals, not claims about growth during the window.

The three picks

Category	Project	Created (UTC-05)	GitHub snapshot
Most absurd	`mathelehrer/BrainFuckLife`	July 28	12 commits, 1 star, 1 fork
Most show-off	`oklo/malbolge-rungs`	July 27	17 commits, 0 stars, 0 forks
Most practical	`belowzeroff/qbase`	July 20	192 commits, 0 stars, 0 forks

Most absurd: Brainfuck grows a primordial soup

BrainFuckLife begins with the least suspicious description imaginable: a minimal Python 3 Brainfuck interpreter with no dependencies. It includes the usual Hello World! and cat examples, plus a trace tool that records the touched cell, its value, loop depth, and output for every step. 1

Then the README opens the trapdoor: bff/ contains BFF, a self-modifying Brainfuck variant based on the 2024 "evolution of programs" work by Agüera y Arcas and colleagues. The experiment starts with random programs and looks for self-replicating behavior in a computational "soup of noise." That is a much stranger use of Brainfuck than another interpreter, and the repository actually gives it a home rather than leaving it as a sentence in a README. 1

The code layout tells you where the author spent the effort: bff/, anim/, docs/, examples/, tests/, and viz/. The most revealing file is bff/live.py, a realtime Matplotlib dashboard that functionally checks candidate replicators, tracks marker-carrier prevalence, and keeps exact-tape, entropy, and execution diagnostics. The ordinary interpreter also has a concrete debugging path:

python3 viz/trace.py examples/hello.bf > trace.json

That command is not the joke. It emits a machine-readable execution trace, which makes the self-modifying experiments inspectable instead of merely cinematic. 1

The repository page showed 12 commits, 1 star, 1 fork, and 0 open issues. GitHub's metadata listed four contributors and no open pull requests at the time of the scan. 4 5 6 7

Why it made the issue: BrainFuckLife takes the phrase "self-modifying code" literally enough to put it in a soup, then adds the instrumentation needed to tell whether anything interesting happened. It is not pretending that a random tape has discovered biology. It is making the question executable.

Most show-off: Malbolge gets a grading rubric

Malbolge was designed to be hostile: the README for malbolge-rungs describes instructions that encrypt themselves after execution, code and data sharing a mutable ternary memory, and the lossy trinary operation known as "crazy." The first working program took two years and was found by machine search rather than hand-written. 2

malbolge-rungs responds by turning that hostility into MAL-51, a ladder of classic-Malbolge programming challenges. Each rung defines a challenge family, a byte transform, cases, and resource limits. A candidate runs on the native evaluator; for ordinary rungs, every case must halt with the exact expected output. Coverage rungs specify how many of 256 single-byte cases must be correct. Several epochs can be used to test multiple deterministic seeds. 2

The design has an important authority boundary. The Python VM in tools/hell_lite is diagnostic only. The native Rust evaluator is the thing that decides whether a solution passes, and verify-leaderboard reruns every claimed solved entry. A leaderboard row is therefore not a social assertion; it is a checked artifact accompanied by an actual .mal program. 2

The useful code path is wonderfully severe:

cargo test
cargo run -p harness -- verify-leaderboard

The first command runs VM conformance tests and re-verifies the leaderboard. The second repeats the full leaderboard check as a release-style gate. The repository also includes a challenge registry, verified solutions, a static leaderboard site, and a generator for procedural training instances. 2

The README says the registry contains 29 original rungs, with additional difficulty-smoothing rungs added later. It also admits that most of the harder ladder remains unsolved: the general single-byte XOR frontier and map8-plus finite maps are still open. That is the right kind of show-off project. It makes the hard part visible instead of quietly moving the goalposts. 2

The public repository was created on July 27 in the issue window. It showed 17 commits, 0 stars, 0 forks, and 0 open issues; metadata listed one contributor and no open pull requests. One caveat matters: the README calls this a clean, self-contained public subset of a larger private project. The public artifact is new to GitHub's surface, but not every idea inside it began on July 27. 8 9 10 11

Why it made the issue: Malbolge normally turns "does this program work?" into an archaeological expedition. MAL-51 turns it into a repeatable test with a ground-truth VM, a difficulty ladder, and an honest unsolved frontier. The language stays awful. The experiment becomes legible.

Most practical: qbase does the unglamorous work

qbase is a batteries-included foundation for building kdb+/q applications. Its boot sequence loads core libraries such as cargs, log, and ns; additional libraries can be loaded with --load-libs or .require.lib. The repository lists dependency loading, structured logging, command-line parsing, namespaces, type normalization, time utilities, CSV, filesystem and HDB helpers, compression, HTTP, IPC, WebSockets, cron, and more. 3

This is not an esolang stunt in the Brainfuck sense. The interesting part comes after the language demonstration: how do you build and maintain an application around it?

The quick start is intentionally boring:

q ~/.qbase/boot.q

The examples directory includes a REST service exposing JSON metrics, a tickerplant subscriber for market data, and a hybrid q-plus-Python pipeline. The README says the test suite contains 48 tests, green on kdb+ 4.1, and the project is released under Apache-2.0. 3

There is a provenance wrinkle. GitHub metadata puts the public repository's creation on July 20, inside this issue's window, but the README identifies it as an Apache-2.0 fork of BuaBook/kdb-common. The 192 commits therefore describe a substantial public codebase, not 192 days of work that began on July 20. That distinction does not disqualify it; it tells you what "new" means here: newly available as this public project. 3 12

The public snapshot showed 192 commits, 0 stars, 0 forks, and 0 open issues. Metadata listed four contributors and no open pull requests. 13 14 15

Why it made the issue: the practical category should not mean "the least weird project in the search results." It should mean that a programmer can imagine using it for a real task. Boot code, dependency loading, logging, networking, examples, and tests clear that bar. Q is still Q, but someone has put a toolbox around it.

The quiet corners were genuinely quiet

The scan found several names that looked promising until the repository page was opened.

Whitespace: text-whitespace-visual is a useful Python tool for making tabs, spaces, end-of-line markers, and trailing whitespace visible, with a CI gate and JSON reports. It is about whitespace characters, not the Whitespace esolang, so it does not qualify for the channel's shortlist. 16
Befunge: befunge-formatter has one commit, no description or readable README in the returned page, and is marked as generated from a C++ template. rotarymars/befunge has 36 commits and a package-like description, but the returned page exposed no project README explaining a working interpreter or formatter. Neither supplied enough evidence for a deep feature. 17 18
APL: apljs identifies itself as an APL-to-JavaScript transpiler and REPL and shows 13 commits, but its repository page has no README-level explanation of supported operations. That is a real lead, just a thinner one than qbase. 19

Two Brainfuck and Malbolge projects were strong enough to keep on the watch list. brainfuck-lean-ts puts a Lean 4 formal model beside a TypeScript twin, encoding the 30,000-cell tape and pointer bounds in types, with 31 executable tests; its one-commit public surface made it a better alternate than a spotlight. 20

albertovillaosorno/malbolge is the opposite kind of near miss: 398 commits, exact VMs, translation validation, compiler research, native execution tiers, and optional CUDA. It is an impressive laboratory, but its README also separates a large amount of planned work from the implemented foundation. MAL-51 made the cleaner show-off story this time because its claims are tied directly to a runnable evaluator and verified leaderboard. 21

The verdict

These three projects give the esolang scene three different escape routes. BrainFuckLife makes a joke language behave like an experimental system. malbolge-rungs makes deliberate suffering measurable. qbase makes an array language look less like a dare and more like a software choice.

The funniest artifact is the Brainfuck soup. The strongest technical flex is the Malbolge ladder. The one most likely to survive contact with an actual workday is qbase, precisely because its README spends so much time on the unglamorous pieces: boot, libraries, examples, and tests.

Brainfuck grows a primordial soup, Malbolge gets graded, and qbase does the boring work