Trading, splitting and merging are purely mechanical — no trust required. One step is different: somebody has to rule on whether the thing happened. Here is how that ruling is made, how long you have to object, and why "obviously true" and "ruled true on-chain" are not the same statement.
The name refers to its default: assume the submitted result is correct.
| Step | What happens |
|---|---|
| 1 | Event ends; someone submits a result on-chain, posting a bond |
| 2 | A dispute window opens — a fixed waiting period |
| 3a | Nobody objects → the result stands, the submitter recovers their bond |
| 3b | Someone bonds a challenge → escalation to a vote; the loser forfeits their bond |
The economics are clean: submitting a false result puts a bond at risk, and anyone can earn that bond by pointing out the falsehood. So long as somebody is watching, lying is negative expected value.
Its efficiency comes from the same place — the overwhelming majority of results go the 3a route and stand without any vote. Only genuine problems pay the arbitration cost.
| Fact | Do |
|---|---|
| Results carry a delay | match over ≠ redeemable; the window must elapse |
| The submitter is not the venue | it is whoever bonds — so wrong submissions do occur |
| Correction requires somebody to act | quiet markets get less scrutiny, so errors survive more easily |
| Disputes stretch the timeline | hours become days once escalated |
Genuine disputes are rarely "what was the score". They live in the gap between the board's wording and reality: does this count as "officially announced"? Does a postponement count as "did not happen"? Does partial satisfaction satisfy it?
In those cases what you consider obvious is irrelevant; the terms govern. So the thing to read before betting is not the news — it is the market's own resolution terms.
"Did BTC rise over these five minutes" needs no human. A price oracle reads a value by a rule fixed in advance. No dispute window, near-instant, no trust in a submitter.
The trade-off: every detail of that rule decides your outcome directly, and the details frequently differ from intuition.
My research judged on the exchange spot close against the open; the market settled on the oracle's final-60-second time-weighted average against the open.
Usually they agree. Judging every window with both rulers: 207 of 2,051 windows disagreed — a 10.1% unconditional mislabel rate across the whole sample — and the disagreement had a direction (125:82 in my favour).
That is enough to manufacture a stable-looking pattern that does not exist. Full account in the broken ruler.
Whichever route is used, the settlement written on-chain is the only truth. Not what an interface displays, and not any convenience field.
I took every disagreeing window to the chain, twice:
| Batch | Result |
|---|---|
| All disputed windows on the short-tenor series | 207 : 0 |
| A second batch from the other tenors' audit | 31 : 0 |
Not one exception — the chain always sides with market settlement. So "I think it should count as my win" has no standing here. The only questions are what the terms say and how the oracle reads.
| Check | Because |
|---|---|
| Human ruling or automatic reading? | the two carry completely different risks |
| If automatic: which source, which timestamp | decides your outcome, and often defies intuition |
| How do edge cases resolve? | postponement, cancellation, draws, partial — disputes live here |
| How long is the dispute window? | determines how long your capital is locked |
Best-value two minutes on the venue. It will not raise your hit rate, but it removes the "I was right and still lost" category of loss — the most avoidable kind there is.
This step rules on what the terms say happened, not on what happened. They agree almost always; when they do not, the terms win — 238 on-chain checks, zero exceptions.