Listing and admission policy

A hypothesis is admitted only if it is falsifiable with a resolution criterion fixed in advance. In practice that means someone can state, before listing, the finding that would count against the hypothesis: the measurement, the population, the threshold, and the protocol that would settle it. A hypothesis that no conceivable result could contradict has no way to be corrected by evidence, so a weekly round could never say anything about it, and it is refused.

Falsifiability is about testable content. It is not a requirement that the financial position expire on a particular date, and an evidence cutoff for one weekly report does not close the market to later discoveries.

• Population, intervention, comparator, outcome. A hypothesis states these where they apply, so the resolution rule is not read differently by each side.
• Fixed in advance. The wording and the resolution criterion are locked at listing. A materially different hypothesis is a different market, listed beside the old one.
• Refuting evidence named. The admission record says what observation would count against the hypothesis.

These exclusions are policy, not examples. They follow from securities law, event-contract restrictions, privacy, and the manipulation-resistance standard above.

• Single-issuer clinical, FDA, financing, or solvency milestones. A contract that references one company's trial result, regulatory decision, financing, or solvency needs a securities-law screen. A scientific topic is not automatically a security-based swap, but a single-issuer connection cannot be ignored, so issuer-specific biotech contracts are out until securities counsel says otherwise. Diversifying the wording or inserting an AI intermediary does not necessarily remove the exposure.
• An identifiable patient's health, death, or private medical outcome. Excluded on privacy, adverse-incentive, control, and public-interest grounds. The pilot uses public, lawful sources, and unpublished patient-level data is not accepted as ordinary trading evidence.
• Harmful or unlawful experiments. Excluded. Statutory event-contract restrictions and other laws can apply to activity that is itself illegal or dangerous, and no scientific framing changes that.
• Assassination, terrorism, or war-linked events. Excluded for the same reason. A hypothesis that turns on violence against a person or a conflict outcome is not a scientific conjecture this product will list.
• Vague or untestable hypotheses. Excluded at admission. If no result would count against the hypothesis, there is nothing for a round to report.

A contract can fail not because it is harmful or untestable but because it tracks nothing anybody has an economic stake in. An external scientific fact and an internally generated AI estimate are different things to build a market on. A battery-performance benchmark with a documented link to commercial production costs presents a stronger economic-use case than an unconstrained model opinion about string theory. Neither is automatically eligible, and merely charging fees does not establish the economic significance a venue would look for.

The test we would have to answer: whose pre-existing risk could this position hedge, how does the estimate relate to that risk, and how is the measure generated independently of the operator? A contract that only tracks the operator's changing judgment may fail an exchange's listing standard, a jurisdictional theory, or both. Note also the two-sided risk: a contract that does not qualify as a CFTC derivative does not become legal by default. It can instead raise gambling, consumer-protection, or other questions without the federal framework the venue might have expected.

The panel estimate is the principal listing risk, and a secure vault does not fix it. Three commercial model providers are not three independent administrators if one operator controls the prompts, the evidence selection, the retries, and the publication. An arithmetic average smooths differences, but every report has direct influence over payments, and shared evidence, training material, or operator discretion can produce correlated errors.

• Separate the administrator from the maker. Whoever runs the panel should be independent of proprietary trading, with enforceable information barriers and independent oversight. Different wallets are not an information barrier when one person controls all of them.
• Publish the methodology. What is averaged, the range, equal weighting, precision, rounding, admissible evidence, cutoff, and publication round are all stated before listing.
• Fix the valid-run rules. Averaging only two, replacing an unavailable provider, deleting an inconvenient outlier, or rerunning a valid answer changes the agreed methodology unless a disclosed contingency authorizes it.
• Accept challenges on objective grounds. Provenance, arithmetic, rule compliance, or a compromised submission can be challenged. A challenge is not a free option to cancel an unfavorable result because a trader disagrees with the model's science.

A broad template that allowed any scientific question would not establish compliance for every future hypothesis. Each funded market would need its own short admission record:

• The exact proposition, its resolution criterion, and the evidence cutoff rules.
• The underlying classification and a securities screen.
• The commercial rationale and the economic-use case.
• Admissible sources and the treatment of contrary evidence, retractions, and corrections.
• The persons, or classes of persons, able to affect the referenced outcome, and the trading restrictions that follow.
• Scientific controls, plausible settlement scenarios, and the jurisdictions admitted for that specific hypothesis, where restrictions vary by product.