Four things landed this week. Three of them because members asked.
Practice reminders — now in Training
Training has a Reminders page. Pick the days and the hour in your own timezone, and choose what actually turns up:
- Just remind me — a short nudge.
- A link to start — opens a session with your pool already chosen.
- Assign me a target — we draw and seal a target before the reminder goes out, so a coordinate is already waiting when you open it.
That last one is the one that matters. You know how it goes: the reminder arrives, you fully intend to sit down and practice, and then there's a pool to pick and a decision about decoys and a nagging sense that you ought to have more time than you actually do — so you tell yourself you'll get to it after one more thing, and by the time you look up it's tomorrow. Nobody ever decides not to practice; it just quietly doesn't happen.
A coordinate that's already drawn removes every one of those small decisions. Nothing to set up, nothing to choose. It's sitting there with your name on it.
We never say which pool it came from — naming it would narrow the field before you've drawn a line.
Email and push stay where they've always been, in your notification settings, and the Reminders page links straight to them.
Host a practice event, not just a prediction
Hosting used to mean ARV: a real-world question, two outcomes, a result to declare. Now you can host a plain target event from the events page or from inside a Space — draw a sealed target from a pool, everyone works it blind, then reveal and judge.
You choose:
- One shared target for the whole group, or a different target for each person
- How many decoys — target only for a reveal-and-discuss, or 1 or 3 decoys to make it scoreable
- Who judges — everyone judges their own, one designated judge, the group as a panel, or the AI analyst
And it works in person. Meet, work the target on paper, upload the readings afterwards. Attendees without an account can take part: the host adds them by name and uploads on their behalf.
Blindness holds exactly as it does online. The platform seals the target and never serves it, uploads close when the event locks, and the reveal comes after that — so a reading cannot be written after the target was seen.
Thanks to Wintari, who wanted to run precisely this in a room.
Accuracy: two new numbers, and an honest correction
EA asked for a way to tell whether results are actually unusual. It turned out we were already calculating exactly that and throwing it away.
Your Accuracy panel now splits your work into rows scored separately, because being on target isn't worth the same everywhere:
- Margin — how much you beat chance by
- Luck odds — how often chance alone would match you, read as "1 in 20" rather than "0.05"
There's a What do these mean? explainer beside it that walks through both in plain language and finishes with your own numbers written out as a sentence.
Two deliberate limits. Luck odds stay hidden until a row has at least 12 scored viewings, because below that there's no honest answer to give — being on target three times out of four is a big margin and means nothing.
And we no longer score your own judging. When you rank the images yourself you already know what you wrote, so a correspondence can be found after the fact with nobody meaning to bend anything. It's real practice and the rate still shows, but it isn't proof, and the leaderboard has always treated it that way. Some hit rates will therefore read lower than before: the old number blended self-judged work and different decoy counts into a single figure, and the new one is smaller and true.
Nobody has enough blind-judged work yet for any row to reach significance. That isn't a bug — it's the honest starting line.
Written targets: the envelope
Joe McMoneagle kept taskings on index cards in unmarked envelopes. He had written every card himself, so the set was no secret — but drawing one at random meant he didn't know which he was working.
You can now build a tasking pool: written questions instead of images. Write your own, draw one blind, and afterwards decide what became of it. If the session answered it, set it aside and it won't come up again. If it was a clear miss, it goes back in the envelope and the reading is discarded. And if it was close but not complete, it's resealed — with that reading carried forward, so your next pass at it builds on the picture rather than starting cold.
The catch is volume: with four questions you'll simply guess which one you drew. So you can seed your pool from any of eight curated collections we've just published — Society, Technology, Natural World, Built Environment, Historical, Exploration, Objects, and People & Activity. 360 taskings in total. Pull in as many as you like; your own questions stay in the mix.
These sessions carry no decoys and no answer key, so they produce no hit rate and no Luck odds, and they don't touch the leaderboard. That's the point — this is work, not calibration.
Three of these four started as posts on the feature board. If something's missing or awkward, that's where to say so.