Semantic Similarity Is Only the First Filter
CloddsBot’s opportunity finder turns market data into a claim about mispricing, but the word “opportunity” covers several different kinds of uncertainty. This chapter separates internal binary-market mispricing, cross-platform spreads, and edge against a fair-value estimate, then follows the matching and verification steps that decide whether a candidate deserves attention. The reader will be able to distinguish a market pair that merely sounds similar from one that has survived meaningful identity, freshness, liquidity, and fee checks.
Three claims hidden inside one word
An opportunity is not a property that a price difference carries by itself. It is a relationship among a market, a comparison rule, a size, and an execution plan. CloddsBot’s finder recognizes three broad relationships, and each one makes a different claim about what should happen next.
The first is internal binary-market mispricing. A binary prediction market represents two complementary outcomes, commonly YES and NO. If the cost of buying both sides is below the amount paid for the winning side at settlement, the arithmetic suggests a gross spread. For example, if YES is quoted at 0.45 and NO at 0.52, their combined price is 0.97. Before fees, fill differences, and settlement complications, the pair appears to leave 0.03 per dollar of combined payoff.
This is the most constrained of the three relationships because both legs refer to one market on one venue. It still requires the finder to locate the correct YES and NO outcomes, validate their prices, and check enough liquidity. A displayed pair may not be buyable at the displayed values for the desired size. The market may change between the two orders. Fees can consume the spread. One leg can fill while the other does not. “Internal” reduces identity ambiguity; it does not remove execution risk.
The second is cross-platform arbitrage. Here the finder compares markets on different venues that are believed to represent the same event. One venue may price YES lower than another, creating a possible buy-low/sell-high plan. Or the lower-priced YES on one venue can be combined with the higher venue’s NO if the outcomes and settlement rules really are complementary.
This claim is stronger in one respect and weaker in another. It can reveal a price difference that a single market cannot, but it depends on two external systems agreeing about an event. The accounts may use different credentials, currencies, fees, order semantics, and settlement timelines. A price difference can reflect a real opportunity, a difference in market wording, or a difference in what happens at resolution. The finder’s matching layer is therefore not a convenience added around the arbitrage calculation. It is part of the calculation’s meaning.
The third is edge against fair value. Instead of comparing two executable venues, the finder compares a market price with an external probability estimate, model, or benchmark when one is available. If a benchmark says an event is more likely than the market price implies, the system can identify a possible buy-YES case; if the estimate is lower, it can identify a possible sell or NO case.
That is not arbitrage in the strict sense. The two sides do not lock in a payoff difference. The trade depends on the fair-value estimate being calibrated, relevant, current, and expressed in terms that match the market.