One decision. Four pieces of evidence.
- 01
A real chart, stopped partway
A real asset over a real window from the past. The prices are exactly what happened.
- 02
4 of the 7 read it
They get the chart up to the cutoff and nothing more. No ticker, no dates. Each one writes what it sees and commits to a call.
- 03
You can see what happened
Switch to the finished chart whenever you want. Nothing here is hidden from you.
- 04
Pick the best read
Whichever one actually read the chart. If none of them did, say so. We would rather know that than have you crown a winner.
Seven forecast windows.
The chart window grows with the forecast. The context-to-target ratio runs from 20:1 for three days to 2:1 for one year.
- 3D60 days shownnext 3 days20:1
- 1W60 days shownnext week8.6:1
- 2W120 days shownnext 2 weeks8.6:1
- 1M180 days shownnext month6:1
- 3M365 days shownnext 3 months4.1:1
- 6M540 days shownnext 6 months3:1
- 1Y730 days shownnext year2:1
A public benchmark built from human judgment.
- 01 / RANK
Measure useful chart reading
Direction alone misses the quality of a read. Voters reward analysis that uses the evidence in the chart well.
- 02 / LEARN
Keep the failures in the record
Every read, vote and optional note becomes part of the public evidence needed to improve the benchmark.
Seven models enter the draw
- 01ByteDanceSeed 1.6 Flash
- 02OpenAIGPT-5.6 Luna
- 03GoogleGemini 3.8 Flash
- 04AlibabaQwen 3.8 Flash
- 05DeepSeekV4 Flash Vision
- 06xAIGrok 4.3
- 07AnthropicClaude Haiku 4.5
All seven run through Orbio. One key reaches every lab, and the cost under each read is what Orbio charged for it.
Bring your own Orbio key.
Lab-funded credit covers the arena for now. Soon, each player will use their own Orbio key for the rounds they play.
USED FOR YOUR ROUNDS / NEVER STORED
Frontier models join next.
The first field favours fast multimodal models so the lab can keep the experiment running. Funded frontier rounds will use the same charts, reveal rules and public scoring method.