Dataset Viewer
Auto-converted to Parquet Duplicate
short_id
string
source_run
string
bot
string
brain
string
model
string
effort
string
seed
int64
mode
string
prompt_version
string
bot_mcp_commit
string
started_utc
string
result
string
free_the_end_s
int64
deaths
int64
cost_usd
float64
tokens
int64
race_tag
string
2026-10-12-meelah-a1
meelah-sonnet-low-prompt-v4-11-meelahA1-2026-10-12
Meelah
claude
sonnet
low
3,120,510,764,609,846,000
hardcore
v4.11
088f4c58bf675ee09edef310771acccc141176eb
2026-10-13T01:01:06Z
WIN
10,323
0
53.19
240,645,868
meelahA1
2026-10-15-meelah-t2
meelah-sonnet-low-prompt-v4-17-meelahT2-2026-10-15
Meelah
claude
sonnet
low
4,518,528,679,372,318,000
hardcore
v4.17
dcd1f7ef97df9a1107b39d4a0fc3ebb6b9fd3dee
2026-10-15T20:02:13Z
WIN
8,373
0
23.47
190,617,553
meelahT2
2026-10-17-sprout-racev3
sprout-sonnet-low-prompt-v4-19-hard-raceV3-2026-10-17
Sprout
claude
sonnet
low
-7,164,599,967,927,602,000
hard
v4.19
d773a8969b132a39aa14c51c5f117b74cd4a4e7e
2026-10-17T19:34:14Z
WIN
5,608
1
13.33
100,768,815
raceV3

AI Minecraft Runs

Receipts for AI models playing survival Minecraft Java 1.21.11 to beat the Ender Dragon: every run we film or cite, losses included, so anyone can check a time, a death count or a claim about why a model lost.

Each run is one AI brain (a language model) driving one bot through a fixed set of game tools (a mineflayer body over MCP). Nothing in a run is edited: the transcript is the brain's own stream, and the server log is the server's own log.

What is here

runs.csv                  one row per run (columns below)
verify.py                 recomputes every run's time and deaths from its server log; standard library only
game_text.json            the 1.21.11 death messages and advancement lines verify.py matches (from minecraft-data)
runs/<short_id>/          one folder per run, named <date>-<bot>-<tag>, e.g. 2026-10-12-meelah-a1
  README.md               title, result, the videos that use this run, the command that verifies it
  meta.json               the runs.csv row (with source_run), the death lines, the server clock's UTC offset, sha256 of every file
  prompt.txt              the exact prompt the brain ran
  transcript.jsonl        the brain's stream: every message and tool call, scrubbed (below)
  server.log              the Minecraft server's log: joins, deaths, advancements, chat, commands, with timestamps
  advancements.json       the bot's advancements file from the world, each criterion stamped by the server
  stats.json              the bot's statistics file from the world (deaths, play time in ticks)
  server.properties       as the server loaded it: seed, difficulty, hardcore
  progress.log            the race loop's progress line every 5 minutes
  cost.json               list-price cost and token counts
  report.md               our analysis of the run, when one was written
  replays/                Flashback replay files, for selected runs only

runs.csv columns: short_id, source_run, bot, brain, model, effort, seed, mode, prompt_version, bot_mcp_commit, started_utc, result, free_the_end_s, deaths, cost_usd, tokens, race_tag. short_id is the run's folder under runs/: the date in the run's name, the bot, and the run's label (2026-10-12-meelah-a1 is Meelah's run A1 of 12 October 2026, Eastern time); source_run is the run's full folder name in our own logs. The exact start is started_utc. free_the_end_s is seconds from the run start to the bot's "Free the End" advancement (the dragon killed); empty when the run did not get there. bot_mcp_commit is the commit of the bot's tool server (a private fork of the MIT-licensed mine-ai-mcp).

How to verify a run

python3 verify.py                  # every run
python3 verify.py runs/<short_id>  # one run

For each run it checks every file's sha256 against meta.json, then recomputes from server.log alone:

  • Time to Free the End: the timestamp of the line <bot> has made the advancement [Free the End], minus started_utc. The log carries times without a date; the date comes from the run start and the server clock's UTC offset, which the advancements file stamps on every criterion.
  • Deaths: server-thread lines whose message is one of Minecraft 1.21.11's death messages for that bot.

Both must match meta.json and runs.csv, and two files the server wrote on its own must agree: the advancements file's timestamp for minecraft:end/kill_dragon, and the minecraft:deaths statistic.

What this cannot prove

  • That no human touched the server outside the log. Operator commands appear in server.log and are kept, but a change made by editing world files, or through a channel the server does not log, would not show.
  • That the replays are complete. A replay records what its recorder saw; gaps or a missing file are possible.
  • That the transcript is the only input the brain had. It is the stream the run loop saved; anything given to the brain outside that stream would not show.
  • Times finer than a second, or a run that crossed a daylight-saving change (the server log's clock would jump).
  • cost.json is list price computed from token counts, not an invoice.

Scrubbing

Before upload every text file is scrubbed and then scanned; a single scan hit stops the upload. Removed: home and volume paths (now ~ and <local>), IP addresses other than loopback, host names, MAC addresses, email addresses, keys and tokens of every known shape, and Discord IDs. Family Minecraft accounts and their UUIDs are replaced with player-1, player-2, ... (UUIDs as player-N-uuid). Bot names (Sprout, Squirt, Meelah) are kept. Nothing else is changed: the times, deaths, chat and tool calls are as the run wrote them.

Licence

Our files (transcripts, logs, analyses, runs.csv, meta.json, replays): CC BY 4.0. verify.py: MIT. game_text.json holds strings from minecraft-data (MIT). Minecraft is a trademark of Mojang; this dataset is not affiliated with or endorsed by Mojang or Microsoft.

Downloads last month
-