Scoreboard
Anyone can make a forecast. The claim, the target date and the resolution criteria are logged up front, then marked as the call resolves. No quiet edits, no cherry-picking.
Hit rate, resolved calls only
75%
3 of 4 calls that came due were graded correct. The other 3 are still running.
Every call tracked
7 totalWhere the calls stand
RADICAL TRANSPARENCY
How a call gets graded
Log the claim
before the outcome
Record the forecast verbatim with its author, the target date and a clear resolution criterion, all fixed before the result is known.
Wait it out, then
resolve on the date
The call stays open until its target window arrives, with no quiet edits, then it is marked correct or wrong against the criterion logged up front.
Keep the record
side by side
Resolved and open calls stay together, misses included, so the hit rate is the whole picture rather than a highlight reel.
NOT YET GRADED
What comes due next
A track record is only worth reading if the pending half is public too. These are logged and dated; none of them can quietly disappear.
COMING DUE
3 open · graded on the date- 2027
An AI agent autonomously completes a multi-day software project end to end
Community forecast1 year out
- 2028
50% chance of human-level machine intelligence
S. Legg (DeepMind)2 years out
- 2029
AGI capable of most economically valuable work
R. Kurzweil (long-stated)3 years out
Each of these was logged with its criterion before the outcome was known. When the target year arrives it gets marked correct or wrong and moves into the hit rate above, including the ones that land badly.
Read the
calls that
came due.
Field notes on the predictions that resolved, what grading them showed, and what moved the readings. Cited, no hype.