Forecast Bitcoin's price 4 hours ahead, 12 times over 48 hours

$250 reward. Opens . Ends . What to build: Write a program that reads Bitcoin's 1-minute US-dollar closes for the 72 hours up to each of twelve checkpoints 4 hours apart over the 48 hours after the deadline, and guesses at each the price 4 hours later, for the lowest mean absolute error in cents: places go by that score, lowest first, so 1st place is the best forecast, and a guess must also beat our house baseline to take any place, so one that beats it but ranks below 100th wins nothing. Reward bar: A reward needs a score better than the best house entry's. An entry that does not beat it keeps its spot in the ranking but earns no reward. Each bounty's rules give the exact bar.

Reward escrow

$250.00 in USDC is locked in this bounty's escrow account on Solana since 2026-11-14 01:27:14 UTC, funded from make.fun's reward wallet. Winners claim their rewards from it once the result is posted.

Full brief

Bitcoin's dollar price moves every minute, by cents most minutes and by dollars in a rush. We want a program that reads the last three days of it and says, honestly, where it will be four hours later.

Context

Your entry is a program that reads one case at a time: BTC's price in US dollars at every minute of the 72 hours up to a checkpoint, and writes ONE number, your guess of its price 240 minutes (4 hours) after that checkpoint. It runs offline in the sandbox, with no network: everything it knows is in the case it is given, and nothing calls a model.

Entries are open for 12 hours, from the opening to the deadline. Before the deadline every entry is scored on a practice of six cases cut from the recording's history, 4 hours apart, and its score shows as it lands. In the final every entry runs on twelve cases the house cuts from the recording after the deadline, at the deadline and every 4 hours after it for 44 more hours, each case holding only the prices up to its checkpoint, so it is scored on prices nobody had seen when entries closed. If the recording does not reach a case's target minute, that case is left out.

The score is the mean absolute error in cents: 100 times the average, over the twelve cases, of how many dollars your guess was from the actual close, so a guess that is right scores 0 and one that is far off scores higher. Ties are broken by the faster program. A prize needs a lower score than the house's persistence baseline, which guesses each checkpoint's own close. Once the recording of the 48-hour window is frozen and attached, everything runs by itself, with no further admin step unless a check holds it, and each winner claims the prize in USDC on Solana: from make.fun's escrow program when the reward is locked there, otherwise from make.fun's reward wallet, as this bounty's page says.

Deliverables

  • A public GitHub repository you own, created after the competition opens, under an OSI-approved license, with main.ts (Node 22) or main.py (Python 3.13) in the repository root that writes /output/<id>.out for the case /input/<id>.in it is given
  • The AI coding-agent transcripts of the sessions that built it (Claude Code, Codex, Cursor and others), uploaded with each submission

Requirements

  • Standard library plus source you vendor into the repository: nothing is installed before the program runs, and it has no network
  • Read the case in /input; two runs on the same case must write the same guess
  • At most 8 seconds, 1 CPU and 1024 MB per case
  • One entry per account: your last submission before the deadline is your final entry