add datasheet

DATASHEET.md

@@ -0,0 +1,63 @@
+# binary-reasoning-v1 — Datasheet
+
+**Product:** verified reverse-engineering tasks (dataset + RL/eval environment) over real binaries.
+**Seller:** Samuel Anttila / RE-Lab (Rovaniemi, FI). Contact: samueljanttila@gmail.com
+**License:** (c) Samuel Anttila / RE-Lab. Free for evaluation & research; commercial license on purchase.
+
+## Composition (3,079 items, unique IDs, 2026-10-28)
+| family | tier | grading | count |
+|---|---|---|---|
+| func_id_exact — identify internal/stripped function from scrubbed disassembly | hard/med | deterministic symbol match | 2,560 |
+| func_id_mc — multiple-choice function identification | medium | deterministic option match | 400 |
+| func_summary — summarize scrubbed function (purpose + identifying detail) | med/hard | rubric LLM judge | 53 |
+| crossarch_equivalence — same function across two ISAs? (+distractors) | medium | deterministic stance+symbol | 25 |
+| agentic_flag_hunt — find argv[1] that makes a shipped binary print OK (sandbox-executable) | hard | exact flag / program stdout + exit-code probe | 14 |
+| arch_semantics_qa / isa_semantics_qa — instruction-sequence semantics | medium | rubric judge | 14 |
+| library_ident — SBOM-style: name+version of a production library from artifacts | easy/med | exact substring (name+version) | 6 |
+| pe_import_reasoning — infer program behavior from imports+strings | medium | rubric judge | 4 |
+| pe_structure_qa — PE header/section facts | easy | exact field / rubric | 3 |
+
+Architectures: x86-64 (Linux ELF, Windows PE), i686, AArch64, ARM32, RISC-V, MIPS(o32/n64), PPC64LE.
+Sources: Ubuntu system libraries (608 unique real .so incl. libcrypto, zstd, sqlite3, libxml2,
+zlib, libpng…), **Debian production cross-toolchain glibc (AArch64/RISC-V/MIPS64/PPC64LE)**,
+**Android/Termux production aarch64 libs (libsqlite3 3.54.1, libpng 1.6.59, libarchive 3.8.10,
+libssh 0.12.2 — platform-tagged in provenance)**, curated PE samples, cross-compiled targets
+(honestly labeled).
+
+## Construction & verification
+- Produced by static analysis only: Ghidra (via REA), binutils objdump/readelf/nm. No target binaries executed for answers.
+- Scrubbing: every symbol equal to the answer or a callee replaced with `<...>`; independent audit: **0/3,005 answer leaks**.
+- Ground truth derived from artifact itself (symbol tables, header fields, decompiler output).
+- Auto-grader included (`grading.py`): deterministic graders for exact/MC/crossarch/flag/library families; rubric judge for open-ended. Test harness `test_grading.py`: 3,000 deterministic items scored 1.0 on gold answers, 0.0 on abstention; 79 routed to judge.
+- Agentic challenges: round 1 (8, unstripped) + round 2 (6, **stripped**, exit status = failing position+1 enables black-box probing); each binary verified OK-on-flag / NO-on-probe; 11,500+ mutation probes total, zero false OKs; all flags recoverable statically and by execution probe; `strings` shows no flag fragments; dictionary-audited.
+- Provenance per item: binary name, sha256_16, address, size, arch.
+
+## Intended use
+RL training (GRPO-style verifiable rewards), post-training eval suites, expert-model benchmarks,
+SFT for code-understanding models. Runnable as-is on Prime Intellect Environments Hub via the
+bundled `verifiers` v1 taskset (`environment/`), or standalone JSONL.
+
+## Positioning vs existing work (surveyed 2026-10-28)
+| neighbor | what it is | how binary-reasoning-v1 differs |
+|---|---|---|
+| CrackMeBench (arXiv:2605.10597, 5/2026) | educational CrackMe-style agent tasks, executable oracles | validates the genre; ours = **production binaries** (Ubuntu/Debian glibc/Termux-Android), 3,079 items, 10 families incl. cross-arch + SBOM-style ID, not hand-made crackmes |
+| knewstimek/crackme-agent-bench (GitHub, 3★) | 3 Windows PE crackmes + MCP debugger tools; **analyzed statically by our REA**: v1 fully static-readable; v2/v3 = TLS-callback packer, `.text` encrypted at rest, API hidden via PEB-walk, memory-only flag (hash of encoded blob, wiped) | no input channel/oracle at all; ours adds argv + exit-status probing; their packing/anti-debug techniques → our v2 roadmap below |
+| CyberGym (sunblaze-ucb, HF dl≈369k) | crash/vuln-finding RL at scale | demand proof for binary-security data; different task (dynamic crashes vs static RE reasoning) |
+| cybench-harbor (HF dl≈1.5k) | Cybench CTFs in Harbor format | precedent that security tasks ship via Harbor — our 14 agentic tasks use the same packaging |
+| AssemblyBench (arXiv:2604.03750 / HF) | **robotics** mechanical assembly — name collision only | unrelated |
+| atul10 RE code datasets, ollvm-decompilation (HF dl<400) | source-level or decompiler-output pairs | no executable oracles, no verification bar, single arch |
+
+Conclusion: no shipped, verified, production-binary RE environment pack of this scale exists on
+the Hub (365k tasks = SWE/terminal/search only) or on HF. First-mover claim holds with evidence.
+
+## Known limits
+- 46 func_summary + 16 PE/xarch-judge items need an LLM judge (model configurable).
+- Some stripped-binary answers are marked `inferred:true`; graders accept role-correct names.
+- Address coordinate systems vary (PIE load base noted in provenance where relevant).
+
+## Files
+- `environment/` — verifiers v1 package (taskset.py, grading.py, items.jsonl, bins/ + src/, README.md, pyproject.toml); **import-tested against real verifiers v1: all 3,079 tasks load**
+- `harbor/` — the 14 agentic challenges ALSO packaged in **Harbor task format** (instruction.md, task.toml, Dockerfile, solution/, tests/test.sh → reward.txt); reward chain tested 14/14 (container build untested locally — no Docker on this box)
+- `part_relf/repe/xarch/bulk/agentic/agentic2/isa/glibc/android.jsonl` — source batches
+- `sample_free.jsonl` — 9-item free sample (one per family) for listings
+- `test_grading.py`, `leak_audit.py` — verification harnesses