What is Market Trust?

A research preview that reads one market and answers whether its price is strong enough to cite: use it, use it with caveats, discount it, or do not cite it, with the caveat and receipt attached. It is labeled a canary preview on the surface itself because it has not yet been calibrated by enough resolved outcomes; that honesty is part of the product, and the label does not come off until the resolved-outcome ledger earns it.

Does the market integrity read detect manipulation?

No. It reports two descriptive structural measures with intervals: participation concentration (how much of recent flow the largest wallets account for, with a bootstrap confidence interval) and a round-trip churn share (an upper bound that deliberately over-counts benign trading). It names no wallet and asserts no manipulation; the most common explanation for high concentration is a single liquidity-providing wallet. It is a structural read, not an allegation.

What does the coherence engine check?

Whether related market prices are logically consistent with each other. Some sets of prices cannot all be right at once, for example mutually exclusive outcomes whose prices sum well past 100%. The engine flags those disagreements as candidates. Its outputs are version-labeled, and the track record is still maturing, which the page states on itself.

Markets, not just wallets

The answer first

Convexly runs three market-level reads. None of them is a wallet verdict, and each one states on its own page what its evidence can and cannot support. The boundary labels are not fine print; they are the finding.

The intuition, with one worked example

Take a market quoting 63% and ask: can I responsibly use that number in my work? A Market Trust card (a research preview) answers with one of four reads, with the caveat and receipt attached: use it, use it with caveats, discount it, or do not cite it. A clean "use" is meant to be rare, and the whole surface is labeled a canary preview because it has not yet been calibrated by enough resolved outcomes. The surface tells you its own maturity before it tells you anything else. That ordering is the lesson.

The actual method

Market Trust (research preview). Reads one market's evidence trail and answers whether the price is strong enough to cite. Boundary as printed: canary preview, experimental, not yet calibrated by resolved outcomes; it does not graduate until the resolved-outcome ledger earns it.
Market integrity read. Two descriptive structural measures with intervals: participation concentration (how much of recent flow the largest wallets account for, with a bootstrap confidence interval and an effective-participant count) and a round-trip churn share (an upper bound that deliberately over-counts benign trading). Boundary as printed: names no wallet, asserts no manipulation; a descriptive structural read, not an allegation.
Coherence engine. Checks whether related market prices are logically consistent with each other, and flags sets that cannot all be right at once. Boundary as printed: version-labeled candidates with a maturing track record, and realized outcomes are counted separately from expectations.

Where you see this on the site

Market Trust (research preview) lives at /market-trust. The concentration methodology is written up at /research/market-integrity-concentration, and the coherence engine's daily output is at /research/cme/signals.

What this does NOT mean

None of these surfaces detects manipulation, names a bad actor, or certifies a market as safe. High participation concentration is most often a single liquidity provider doing its job. A coherence flag is a structural disagreement between prices, not an instruction to trade it. And a Market Trust read is a research preview, not a calibrated verdict: until the resolved-outcome ledger matures, its own page tells you not to cite it as one. The boundary language is the deliverable; removing it would make every one of these reads wrong.

Next lesson: what we never say

Convexly publishes new methodology research roughly every 6-8 weeks plus the /learn series on a rolling cadence. Get the next paper in your inbox when it ships: