muend.
Deterministic evaluation · Exact Rust harness

benchfck

A generator and exact harness for machine-state tasks, built so that nothing in the scoring path depends on a learned judge.

v0.4.0-alphaApache-2.0Rust
Version0.4.0-alpha · engineering candidate
LicenseApache-2.0 (code) · CC BY 4.0 (future datasets)
LanguageRust, 2024 edition
Machine30,000 cells · 8-bit wrapping · 8,000,000 step cap
Tokenizercl100k_base for BPE measurement
Model runs0 published · 0 leaderboards
Why

A benchmark that cannot be gamed by its own judge

Most reasoning benchmarks ship a fixed dataset and score answers with another model. Both choices leak: the dataset ages into the training corpus, and the judge brings its own errors. benchfck removes both. It generates every item from a seed, and every step from execution to scoring is exact arithmetic on a pinned virtual machine.

The cost of that choice is scope. It measures controlled machine-state reasoning, and nothing else. It does not claim to measure general capability.

Fig. 01

From seed to score

A candidate program is only accepted if the typed intermediate representation and the encodings agree on every one of the 256 possible input bindings. Anything that disagrees is discarded before a model ever sees it.

Fig. 01 / Generation and verification chain
GENERATION → ENCODING → EQUIVALENCE → SCORING SHEET B-00 · v0.4.0-ALPHA 01 / DETERMINISTIC GENERATION SEED + PUBLIC CONSTRUCTOR 8 CONSTRUCTORS · 4 STRUCTURAL FAMILIES TYPED INTERMEDIATE REPRESENTATION SEMANTICS FIXED BEFORE SYNTAX BRAINFUCK COMPILER 30,000 CELLS · 8-BIT WRAP · 8M STEP CAP 02 / INSTRUCTION ENCODINGS BPE MEASURED WITH cl100k_base E0 · CANONICAL BF IMPLICIT POINTER-RELATIVE E1 · PERMUTED SYMBOLS PER-ITEM LEGEND E2 · COMPACT EXPLICIT RLE CARRIER · 3.356× BPE E3 · VERBOSE EXPLICIT RLE CARRIER · 7.072× BPE 03 / ACCEPTANCE GATE COMPLETE INPUT DOMAIN INDEPENDENT PARSER VALIDATION IR ≡ E0 ≡ E2 ≡ E3 OVER 256 BINDINGS A CANDIDATE IS ACCEPTED ONLY IF EVERY REPRESENTATION AGREES ON EVERY INPUT 04 / TASK FAMILIES T1 · STATE TRACKING MACHINE STATE AT INTERIOR STEPS T2 · COMPUTATION COMPRESSION EXPRESSION MATCHING BEHAVIOUR T3 · CAUSAL MUTATION OUTPUT AFTER A PROGRAM EDIT RESTRICTED VERIFIER · EXACT SCORING · NO LEARNED JUDGE · NO FIXED DATASET
Encodings

Four rungs of the same program

Each accepted program is rendered into four instruction encodings. They describe identical behaviour with very different surface forms, which separates a model that reasons about machine state from one that pattern-matches familiar Brainfuck idioms.

  • E0 — canonical Brainfuck with implicit pointer-relative addressing.
  • E1 — per-item symbol permutation plus an operational legend, so memorised glyphs stop helping.
  • E2 — compact explicit operations with a run-length carrier, 3.356× the token count of E0.
  • E3 — verbose explicit operations, 7.072× E0.
Gates

What a generated item has to clear

Population-level gates are published before any model is run, so the acceptance criteria cannot be tuned after seeing results.

Trace semantic density≥ 0.30

Execution has to actually do work, not idle through steps.

Avalanche score≥ 0.60

Small input changes must propagate into the machine state.

Canonical-idiom rate< 0.08

Programs that collapse into memorised idioms are rejected.

Constructor breadth1,730

Unique semantic functions identified across 51 coarse profile buckets.

Testing

Property tests over a 10,000-program population

Four balanced, non-overlapping CI jobs partition a 10,000-program population, so a regression in one structural family cannot hide behind the others. GitHub Actions runs CI and CodeQL; local control scripts reproduce the same checks offline.

Scope

This is a v0.4.0-alpha engineering candidate. No model results have been produced and no leaderboard exists. The release scope is arity 1, with arity 2 deferred to v0.5. The public generator reveals its constructor family by design, so the eight public constructors are a narrow subset of the identified space, and the private scoring epoch has not been activated.

Open to collaboration

Build spatial systems that hold up.

© 2026 Muhammed Enes Duran Static site · no trackers
Technical diagram