Few people on it, no reason in the record
| Target | Runs | Attention | In short |
|---|
Many people on it, record no longer supports it
| Target | Runs | Attention | In short |
|---|
All targets
Attention: papers, trials, sponsors, patents, grants, scaled 0 to 100. Under 35: asked "why is nobody here?". 60 and above: asked "does the record support the crowd?". Between: not asked in round 1.
| Target | Attention | Question asked | Three runs | Result | In short |
|---|
My targets
Your team's targets, same question, same engine. Private instance. Nothing here says leave.
| Target | Attention | Three runs | What changed since the last run |
|---|
Flags on your targets
Rules
Written before use. Rules marked waiting came up during round 1 and need a decision by the scientist; they are not applied until made.
How it works
Scripts count attention. One fixed engine argues one question per target, both sides, then judges, three times. All three answers are published. Open questions go to a scientist, who writes a rule.
What the three answers mean
- No reason found. The engine looked for a reason for the field's behavior in the evidence list and found none that a source supports.
- Reason found. A source in the list gives a reason, and it holds up under the rules.
- Could not tell. The evidence list is too thin to judge, or the question needs a rule that does not exist yet.
The engine is not deterministic, so three runs can disagree. The disagreement is shown. Two "no reason found" out of three is a finding.
What this app does not do
- It does not watch the literature. It runs on a date, over a fixed list of targets.
- It does not use teams of agents. One engine is asked the same question three times.
- It does not choose what to read. The evidence list is fixed before the engine sees it, and what the engine looks up inside a run is recorded.
- It does not learn from you. Rules are written by a person.
- It does not recommend. It reports where the record and the field's behavior do not match.
Four things you can ask about any number on this page
Which engine version. Was it run more than once. Where each piece of evidence came from. Who decided.
Limits
- "Few people" means few visible people. Private programs and unpublished failures are not in any count. A target nobody visible is on may be one several teams quietly left.
- The evidence list is the public record on the run date. For some targets it is thin, and the engine says so rather than guessing.
- Thirteen middle-band targets were not asked in round 1. Neither cleared nor flagged.
- Each target takes hours on one machine. A run over the whole list takes weeks, which is why runs are dated rather than continuous.
Checking the findings later
Before round 1 the scientist wrote down the expected result for ten targets; the round is scored against it, misses kept. Two years after each round, every finding is checked against public events: a program started, a reason appeared, the crowd moved. Nothing is scored yet.
Predictions written before round 1
| Target | Question | Expected | Why | Observed |
|---|