Rules
How points work, who decides, and what is allowed.
Points are not accuracy
Points grow with how many predictions you log and how hard they are. Free accounts can log 3 a day and members 6, so a high score does not mean a high accuracy. The standings rank by points; accuracy is on every row, and on every profile it is shown by topic.
What “AI expected” means
8 of 11 came true · 73%
AI expected 45%
How scoring works
When you log a prediction, AI makes a quick estimate of how likely it is to come true (p). The rules work out how much it is worth (w) from its category and time span. Those two numbers set the points.
The formula
- If right: 100 × w × (1 − p)
- If wrong: −100 × w × 0.75 × p
- No verdict: 0
A worked example
By 31 December 2027, the US federal minimum wage is raised above $7.25 an hour for the first time since 2009.
If right +180 · if wrong −15 points
The less likely AI thinks it is, the more it’s worth.
Politics, 3 months (w = 1.50)
| AI’s quick estimate | If right | If wrong |
|---|---|---|
| 90% | +15.0 | −101.3 |
| 50% | +75.0 | −56.3 |
| 10% | +135.0 | −11.3 |
The same five levels at w = 1.00
| AI’s quick estimate | If right | If wrong |
|---|---|---|
| 10% | +90.0 | −7.5 |
| 30% | +70.0 | −22.5 |
| 50% | +50.0 | −37.5 |
| 70% | +30.0 | −52.5 |
| 90% | +10.0 | −67.5 |
The less likely AI thinks it is, the more it is worth if it comes true, and the fewer points are taken off if it does not. Results are rounded to 2 decimals and shown with one, or two when under 1.
AI’s estimate: quick, and one of five
AI’s estimate is a quick estimate from general knowledge. It does not search the web and it is not research. It can be wrong, and it only ever says one of five things:
10%, 30%, 50%, 70% or 90%.
Nothing is ever certain, so there is no 0% or 100%. The estimate is made once, when you log, and it cannot be reviewed or changed. Only the category can be reviewed.
Category and time span
How much a prediction is worth depends on its category and on how far away its deadline is. You choose the category when you confirm (one is pre-selected). AI checks it once, without searching the web, and uses its own if it clearly disagrees.
Category
- Politics, World Affairs, Economy & Markets×1.00
- Business, Technology & AI, Science & Health, Crypto×0.80
- Sports, Entertainment & Culture×0.60
- Other×0.40
Time span, from logging to the deadline
- 1 to 7 days×1.0
- 8 to 30 days×1.2
- 31 to 365 days×1.5
- Over 365 days×2.0
Within 72 hours of logging you can ask for one category review. AI looks again without being told the first answer. Your points can go down as well as up.
The low-score flag
Some predictions are worth almost nothing either way: personal matters, things you control yourself, and things that are almost certain to happen or not to happen. AI flags them when it makes the quick estimate.
- The category factor becomes 0.02. The time-span factor still applies.
- A flagged prediction does not count towards accuracy.
- Without this, posting sure things every day would be the easiest way to climb.
A personal matter, 1 day, AI at 50%: w = 0.02.
How a verdict is reached
AI can misread sources. A verdict is the site’s best reading of public information, not a statement of fact. Predictions are opinions, not advice.
- The deadline passes. AI searches the web once for what happened.
- If the answer is clear, a provisional verdict is shown: Came true or Didn’t happen. Points are not final yet.
- After 72 hours without an appeal, the verdict becomes final and the points are added.
When nothing can be found, the kind of prediction decides
If the deadline passes and nothing can be found, what counts depends on the kind of prediction. You are told which kind yours is before you log it. If the sources conflict: no verdict, 0 either way.
- Will happen
Nothing found by the deadlineDidn’t happen - Will not happen
Nothing found by the deadlineCame true - A number condition, such as “revenue above $1 million”
The figure is not publishedNo verdict, after a second check
- If the result was already public before you logged it, it counts as a wrong guess. The author can appeal if they could not have known.
- If a check finds no clear answer, AI checks once more 3 days later. If there is still none, the prediction is void: 0 points, and it does not count for accuracy or towards the 5 verdicts for the board.
- Early verdict: when a result is already public, anyone with a verified email can ask for “It already has a result”, once the prediction is at least 24 hours old. Limits: 1 a day (3 for members), 1 per prediction per day. It is decided only if the answer is clear now, and the time span then counts the real days, so the points can be lower. If nothing is clear, the prediction stays open.
Appeals and the AI review
- Only the author can appeal, once per verdict, within 72 hours of the provisional verdict. A “No verdict” can be appealed too.
- A reason is required, 20 to 1,000 characters. You can add one link to a source. The reason and the link are not shown; the ruling, its grounds and its sources are.
- A second, stronger AI review decides: up to 4 searches, and it reads your link. The ruling is shown on the prediction with its grounds. While it works, the page says “Under review”.
- A ruling in your favour needs an independent source that is not user-generated. A link you add does not count unless AI also finds such a source.
- The ruling can keep the verdict, change it, or declare no verdict. The second review’s ruling ends the appeal; you cannot appeal again, and the verdict does not change after that. Before a verdict is final, a report or an obvious error can still change it.
- Limits: 3 appeals per 24 hours and 30 a month. If 5 of your appeals in 30 days were not upheld, new ones wait. Appeals that were upheld do not count.
What is not allowed
- Predictions about the death or injury of a specific person.
- Predictions about minors or ordinary private people.
- Anything nobody can check from public sources.
- Illegal or hateful content.
Two more things are turned away when you write: matters too small for reliable news or official sources to report, and number conditions with no public source that publishes the figure regularly. You see the reason before you log, so you can fix the text.
Titles
Titles follow your place on the all-time board, as a share of everyone on it, and are recalculated every hour. They cannot be bought.
- top 1%
- top 5%
- top 20%
- top 50%
- everyone else on the board
The standings
- A place needs 5 predictions with a verdict. “No verdict” does not count.
- The board ranks by points, not accuracy. Accuracy is shown next to the points.
- Weekly and monthly boards count only verdicts that landed in that week (UTC, Monday start) or month.
- Same points: the higher accuracy goes first, then whoever got there first.
Your record is permanent
A prediction cannot be edited or deleted once it is logged. A miss stays on your page, and you cannot hide it. The only exception is a prediction that breaks the rules below: a moderator can take it down, and the page then says it was removed. That is the point of a record.
Points only. No money, no betting.
Verified record
Verified record
What the badge says
It says one thing: this person has a record of at least 20 predictions that count toward accuracy, and the full record is on the page. It does not say who the person is, that they are an expert, or that Time Traveler vouches for them.
The two conditions
- At least 20 predictions that count toward accuracy. Predictions with no verdict do not count. Predictions flagged as trivial or near-certain do not count either.
- An active subscription to the service ($14.99 a month). When it ends, the badge, the links and the custom address stop showing. The address is kept for 30 days, then released.
What cannot be hidden
Wrong guesses and no-verdict predictions stay on the page, in the same size as the rest. Accuracy is always shown with how many predictions it comes from, for example 59% of 37. Below 5 counted predictions only the count is shown.
What paying does not change
Points, rank and the way predictions are judged are the same for everyone. The service adds a page, links and a card. It does not add points.
About Time Traveler
Time Traveler is a public record of predictions. AI rewrites each one so it can be checked, gives a quick estimate of how likely it is, and judges it at the deadline. The standings rank people by points.
Points only. No money, no betting.
- Who runs it
- Zhan&Rushman LLC d/b/a MatrixBunny, 3901 Colwick Ct, Plano, TX 75023, USA
- Contact
- Questions, corrections and legal requests: support@zhanrushmanllc.com