add dataset card
Browse files
README.md
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: "other"
|
| 3 |
+
tags:
|
| 4 |
+
- reinforcement-learning
|
| 5 |
+
- evaluation
|
| 6 |
+
- reverse-engineering
|
| 7 |
+
- binaries
|
| 8 |
+
- elf
|
| 9 |
+
- pe
|
| 10 |
+
- cross-architecture
|
| 11 |
+
- taskset
|
| 12 |
+
size_categories:
|
| 13 |
+
- 1K<n<10K
|
| 14 |
+
pretty_name: binary-reasoning-v1
|
| 15 |
+
---
|
| 16 |
+
|
| 17 |
+
# binary-reasoning-v1
|
| 18 |
+
|
| 19 |
+
**Verified reverse-engineering tasks over real production binaries** — an RL / eval environment
|
| 20 |
+
pack where every answer is derived from the artifact itself and machine-checked.
|
| 21 |
+
|
| 22 |
+
- **3,079 items · 10 families · 7 ISAs** (x86-64, i686, AArch64, ARM32, RISC-V, MIPS o32/n64, PPC64LE)
|
| 23 |
+
- Sources: Ubuntu system libraries (608 real .so), Debian production cross-toolchain glibc,
|
| 24 |
+
Android/Termux production aarch64 libs, curated Windows PEs, labeled cross-compiles
|
| 25 |
+
- Provenance per item: binary name, sha256_16, address, arch
|
| 26 |
+
- **Verification:** 3,000 deterministic items score 1.0 on gold / 0.0 on abstention; 0/3,079 answer
|
| 27 |
+
leaks; all tasks load against the real `verifiers` v1 framework; Harbor task packages for the
|
| 28 |
+
agentic family pass 14/14 reward-chain tests
|
| 29 |
+
- **Baseline:** a frontier local reasoning model scores 22% on our deterministic subset — headroom
|
| 30 |
+
is the feature
|
| 31 |
+
|
| 32 |
+
## Contents of this repo (free sample)
|
| 33 |
+
| file | what |
|
| 34 |
+
|---|---|
|
| 35 |
+
| `DATASHEET.md` | full composition table, guarantees, known limits, positioning vs existing work |
|
| 36 |
+
| `sample_free.jsonl` | 9 items — one per family |
|
| 37 |
+
| `EVAL_RESULTS.md` | baseline eval methodology + results |
|
| 38 |
+
|
| 39 |
+
The **full pack** (3,079 items + runnable verifiers v1 environment + Harbor packages + agentic
|
| 40 |
+
binaries & sources) lives in the gated companion repo:
|
| 41 |
+
[S3RL4/binary-reasoning-v1-full](https://huggingface.co/datasets/S3RL4/binary-reasoning-v1-full) —
|
| 42 |
+
request access, approved same day. Commercial licensing and custom taskset builds: samueljanttila@gmail.com.
|
| 43 |
+
|
| 44 |
+
## License
|
| 45 |
+
All rights reserved pending commercial license; research evaluation use granted for the free sample.
|