Dataset Viewer
Auto-converted to Parquet Duplicate
task_type
stringclasses
5 values
turns
stringlengths
13.6k
30.9k
c2o
[{"role": "input", "content": "[.>+>[...+-[>>-+-.+-.]->+-+]--.+]..+-."}, {"role": "reason", "content": "2 50 53 53 44 32 48 \u2298\u2259\u22ca\u00fc\u22f2\u043d\u228a\u044e \u03ac\u221c\u25c5\u2284\u044f~\u2221\u22ca\u00ff\u22c8 \u25a1\u25f7\u22f3\u22d0\u2231\u25af\u00e9\u0413\u2276\u25fc\u25aa\u00aa\u2292 \u0442\u00d...
c2o
[{"role": "input", "content": "[[.[>+.]]-++[-+.]>>.[+.]+.[->.[[[-..]+..>+.]+].->--]]-->+."}, {"role": "reason", "content": "\u00ed\u2231\u2292\u00d4+\u25b7\u2270\u21c4\u2216 32 48 44 32 48 44 32 2 49 44 32 49 44 32 \u2237a\u2286\u22f2\u22d8\u22f3\u22c8\u22b4\\\u043d\u03bb\u219b\u21fc\u25c0 \u03c3E\u2263\u25a9\u22fc\u0...
o2c_masked
[{"role": "input", "content": "91 50 53 53 44 32 50 53 53 44 32 50 53 53 44 32 50 53 53 44 [MASK] 48 [MASK] 32 48 44 [MASK] 48 44 [MASK] 48 93"}, {"role": "reason", "content": " 44 32 48 44 32 50 4 91 48 44 32 48 93 \u00af=\u21f9\u00be3+\u21a4\u2190\u25c9\u25d0\u22e9\u2298\u00d7\u2263\u00ec\u00d7\u22fc\u21d2 \u21b4\u22...
o2c
[{"role": "input", "content": "91 50 44 32 50 53 53 44 32 50 53 53 44 32 48 44 32 49 44 32 49 44 32 50 44 32 49 44 32 49 44 32 49 44 32 49 44 32 49 44 32 49 44 32 49 44 32 49 44 32 49 44 32 49 44 32 49 44 32 49 44 32 49 44 32 49 44 32 49 44 32 49 44 32 49 44 32 49 44 32 49 44 32 49 44 32 49 44 32 49 44 32 49 44 32 49 4...
o2c
[{"role": "input", "content": "91 48 44 32 48 44 32 48 44 32 48 93"}, {"role": "reason", "content": "+..+-+...+.[.].>.++. 44 32 53 44 32 52 44 \u21a2\u00d8\u21dc\u21d3\u21d9\u03bf\u00fd\u03b7\u03c3\u2191\u03be\u22d1h \u2241\u03a7\u229e\u22c4\u21f9\u0424\u25eb\u21f7\u2192\u224d\u222c\u0413\u03ba\u21bf\u00a3\u21fd\u22bd ...
o2c
[{"role": "input", "content": "91 50 53 53 44 32 49 44 32 49 44 32 50 44 32 50 44 32 49 44 32 49 44 32 48 44 32 49 44 32 49 44 32 50 44 32 50 44 32 49 44 32 49 44 32 48 44 32 49 44 32 49 44 32 50 44 32 50 44 32 49 44 32 49 44 32 48 44 32 49 44 32 49 44 32 50 44 32 50 44 32 49 44 32 49 44 32 48 44 32 49 44 32 49 44 32 5...
o2c_truncated
[{"role": "input", "content": "91 48 44 32 48 44 32 48 44 32 48 44 32 48 44 32 48 44 32 48 44 32 48 44 32 48 44 32 48 44 32 48 44 32 48 44 32 48 44 32 48 44 32 48 44 32 48 44 32 48 44 32 48 44 32 48 ..."}, {"role": "reason", "content": "\u22bd\u039f\u2280\u0411\u25cc\u22be\u22fded \u00ea\u0397R\u22a3\u00ddI\u22d0J\u03a...
o2c
[{"role": "input", "content": "91 49 44 32 48 44 32 50 53 53 44 32 50 53 53 93"}, {"role": "reason", "content": ".+.+--[[>..[>]-]+]-[ -[->[-+[[-+.]..>>]+. 2 50 53 53 44 32 50 >[[->>.]-+>>>->+].>> +[-+[..->.-[[>.][>.+ +.--+-+-[].+>[+[>.>[ 3 53 44 32 50 53 53 44 32 48 44 32 48 93 32 50 53 53 44 32 48 49 44 32 49 44 32...
o2c_masked
[{"role": "input", "content": "91 49 [MASK]"}, {"role": "reason", "content": "32 50 44 32 50 44 32 91 48 93 4 32 48 44 32 50 44 \u22e5\u25a7~\u22a1.3X\u21b4\u2222 \u00cd\u00af\u25a6\u2263\u21eb[\u03c7\u2261\u2240\u25bc\u00ef \u043a\u00ca\u22f6\u2228\u0439\u25ac\u22e9\u22b0\u00e6\u044e [..-+.-]-+.+-.[.+>.+ f\u21ca\u22b...
o2c_masked
[{"role": "input", "content": "91 [MASK] 44 32 50 44 32 50 44 32 [MASK] [MASK] [MASK] 48 44 32 48 44 32 48 44 32 48 44 [MASK] 48 44 [MASK] 48 [MASK] 32 50 53 53 44 32 49 44 32 49 44 32 49 [MASK] 32 49 44 32 48 44 32 48 [MASK] [MASK] 48 44 32 48 44 32 48 44 32 48 44 32 50 53 53 [MASK] 32 [MASK] 44 32 49 44 32 [MASK] 44 ...
End of preview. Expand in Data Studio

bf_random_dr4_2loop

Synthetic multi-agent communication training data generated by top secret agent bubbles.

What's in it

Each row contains a single agent interaction sample consisting of a task type and a sequence of conversation turns. The turn sequence follows a strict structure:

input → [reason, send, recv, recv] × 8 → output
  • reason: internal reasoning chain (~300 chars)
  • send: outgoing message to the group (~160 chars)
  • recv: received messages from the group (2 per loop, ~165 chars each)

The content of each turn is built from a stochastic mixture of:

  • ~30% Brainfuck code snippets
  • ~60% space-delimited ASCII ordinal strings (BF output)
  • ~10% unicode noise (ASCII, Latin-1, Greek, Cyrillic, symbols)

No templates, prefixes, or suffixes — pure stochastic splicing.

Task types

Task Type Input Output
c empty string "" space-delimited ASCII ordinals
c2o brainfuck program space-delimited ASCII ordinals
o2c space-delimited ordinals brainfuck program
o2c_masked masked ordinals (15% [MASK]) brainfuck program
o2c_truncated random % (10-90%) truncated ordinals + " ..." brainfuck program

Each sample has 34 turns total and ~7,700 characters of content across all turns.

Generation details

  • 38,000 samples
  • 8 agents simulated per sample
  • 8 communication loops per sample
  • 150-char base chain length for turn content (20-char max per slice)
  • Unicode spans: 6-18 characters
  • Sourced from a 40,000-entry Brainfuck corpus
Downloads last month
-