Is Convexly a replacement for Dune?

No. They answer different questions. Dune gives you SQL and a canvas: you query raw on-chain data across many chains and build the dashboards you want. Convexly answers one narrow question with the analysis already done: does this prediction-market wallet's resolved record separate from chance, after correcting for how many wallets you looked at. Many researchers would use both.

What does Dune do better than Convexly?

Flexibility and reach. Dune lets you query almost any on-chain table across many chains and shape the result into custom dashboards, so it can answer questions Convexly never touches. Convexly does not attempt that: its scope is prediction-market wallet skill, primarily on Polymarket, delivered as a finished read rather than a query surface.

What does Convexly do that a SQL and dashboard tool does not?

It ships the statistics finished and reproducible, so you do not have to write or validate them yourself. Every skill read carries a bootstrap 95 percent interval alongside its resolved-position count; cohort screens apply a Benjamini-Hochberg false-discovery-rate correction; enterprise cohort work is anchored by a size-matched negative control of 500 seeded random draws; the method is frozen and version-controlled; and nulls are published, including against Convexly's own leaderboard, where 0 of the 35 testable wallets in the published top-50 cohort cleared the corrected bar in the frozen 2026-06-09 scan.

Which one should I use before following a trader?

For a Polymarket wallet, Convexly's free analyzer returns a four-state read (skilled, not separable from chance, insufficient, or flagged) with its 95 percent interval and the resolved-position count the read is based on, rather than a raw profit rank you assemble in SQL. A past read is not a forecast, and no read is an instruction to copy any wallet.

Convexly vs Dune for prediction markets

Side by side

Dimension	Dune	Convexly
Primary job	Build-your-own SQL queries and dashboards over raw on-chain data across many chains	Finished statistical read of prediction-market wallet skill: does a record separate from chance
Coverage	Many chains and any on-chain table you can query	Prediction markets, primarily Polymarket (cross-venue only via the published V1-M paper)
Who does the analysis	You write the SQL and build the charts	The read is finished and reproducible; the method is frozen and version-controlled
What a wallet result means	Whatever your query returns: raw rows or aggregates you define	A four-state statistical read (skilled / not separable from chance / insufficient / flagged), each state with a frozen threshold
Uncertainty	Not provided by default; you add it yourself	Every skill read carries a BCa bootstrap 95 percent interval and its resolved-position count; a point estimate is never published alone
Multiple-testing correction	Not built in	Benjamini-Hochberg FDR correction across every cohort screen (primary q = 0.10, with q = 0.05 / 0.20 sensitivity in enterprise work)
Chance baseline	Not built in	Size-matched negative control: 500 seeded random cohorts run through the identical test in enterprise cohort audits
Negative results	Not part of the product	Published, including against its own cohort: 0 of 35 testable top-50 wallets cleared the corrected bar (2026-06-09 scan)
Best for	Exploring raw on-chain data your own way, across chains and tokens	Deciding whether a prediction-market record is evidence of skill before acting on it

"Not built in" is descriptive, not a criticism: a general query surface and a finished audit layer are optimizing different things. You can build a lot in Dune; the point is you build it.

What Dune does well

Dune's strength is flexibility. It exposes raw on-chain data across many chains as SQL, so you can ask almost any question and shape the answer into the dashboard you want. If your question is "let me query this data my own way and chart it", that is the job Dune is built for, and prediction markets are one small dataset among many you can reach. The tradeoff is that the analysis, and its correctness, is yours to write.

What Convexly does differently

Convexly is not a query surface. It ships one domain already analyzed: prediction-market wallet skill, primarily on Polymarket (cross-venue claims only via the published V1-M paper). You do not write the statistics; they come finished and reproducible. Concretely:

A four-state skill read (skilled / not separable from chance / insufficient / flagged) instead of a raw PnL rank you assemble yourself, with every state gated by a frozen threshold.
Every read carries a BCa bootstrap 95 percent interval alongside the resolved-position count it is based on; a point estimate is never published on its own.
Cohort screens apply a Benjamini-Hochberg false-discovery-rate correction, and enterprise cohort work is anchored by a size-matched negative control of 500 seeded random draws.
Nulls are published, including against Convexly's own board: in the frozen 2026-06-09 scan, 0 of the 35 testable wallets in the published top-50 cohort cleared the corrected bar (full table).
Methods are frozen and version-controlled, and follow-up test designs are filed in public registries before the analyses run, as a standing practice documented on the research index.

Which one for which job

Exploring raw on-chain data your own way, across chains and tokens, and building the charts to match: Dune. Deciding whether a specific prediction-market record is evidence of skill before you act on it, including before you copy it: Convexly. The two are complementary, and the honest answer for many desks is both. You can even query the same wallets in Dune and bring them to Convexly for the skill-vs-luck read.

Check a Polymarket wallet free