Skip to content
Convexly

Governance · stated 2026-06-11

Independence

Convexly sells tests, not outcomes. It executes no trades, routes no orders, holds no funds, runs no exchange, has no token, and holds no position in any market it rates. Verdicts come from fixed, published decision rules running on public records, with the interval printed next to the estimate and the method frozen and hash-pinned before the window it is judged on. Failed tests are published next to the passed ones, and errors are corrected in the open. Nobody can pay for a verdict, and no verdict pays Convexly.

What Convexly does not do

  • No trade execution. Convexly places no orders and has no trading interface. The only wallet interaction anywhere on the site is a message signature used for ownership claims; a message signature cannot move funds.
  • No order routing. No order is created, transmitted, or managed by Convexly. Outbound links open the venue's own site. See the disclosure below this list.
  • No trade-mirroring product. Nothing on Convexly mirrors, automates, or executes another wallet's trades. Following a wallet on the watchlist produces notifications, not order placement.
  • No proprietary trading on rated markets. Convexly holds no position in any market it rates and does not trade on its own ratings. This is a standing policy as well as the current fact.
  • No custody. Convexly never holds customer funds or keys.
  • No venue buys coverage. No exchange or market venue owns any part of Convexly, pays Convexly for coverage, for a rating, for placement, or for favorable treatment, or has any input into a verdict. No venue reviews, approves, or delays a rating, a research result, or a method before it is published. No money Convexly receives, from a venue or from anyone else, is contingent on what its numbers say. Outbound links to venues carry no referral or builder attribution, so no Convexly link can earn a per-trade fee. Convexly's operating revenue is subscriptions and fixed-fee research services billed directly to its own customers.
  • No token. There is no Convexly token, and no feature is token-gated.

Venue funding

Convexly may accept unconditional funding for infrastructure, including a venue builder grant, where the money funds a named build and nothing else. Convexly declines any funding conditioned on coverage, on favorable treatment, on pre-publication review, or on any approval right over what Convexly publishes, whatever the amount and whoever offers it. Where Convexly declines funding on these grounds, it says so. Pre-publication review and advance notice are different things. Where Convexly gives a venue advance notice of a finding, the notice carries no right to change, delay, or block it. The test is simple and it is applied before signing: could Convexly deliver the funded work in full and publish a result the funder dislikes on the same day? If the answer is no, the money funds an outcome rather than a build, and it is declined.

Any funding received from a trading venue is disclosed on this page, with its source, its amount, the date received, what it funds, and what it does not buy, published before any work it funds ships.

As of 2026-08-09, Convexly has received no grant, payment, fee, equity investment, or other funding from any trading venue.

Disclosure

Before 2026-06-11, outbound “View on Polymarket” links carried a public attribution code under that venue's builder program, which could route a per-trade fee to the link source. The code was removed from all outbound links on 2026-06-11 and no fee was ever received through it. Corrected 2026-08-22: this disclosure previously said the removal was checkable in a public repository. The repository is private; what a reader can check is that no outbound venue link on this site carries an attribution parameter.

What Convexly has declined

  • 2026-08-22, a venue data license. Convexly did not sign the data license a trading venue offers for products built on its market data. The venue's own published template for that license class carries a revenue share, a mandatory attribution legend on the licensed product, a clause against publishing statements likely to harm the venue's reputation, and audit rights over the licensee. An audit layer that cannot publish a negative finding about a venue is not an audit layer, so the license was not filed. The venue is not named here because the template is public and the clauses are the point, not the counterparty; Convexly continues to seek research access to venue data on terms that leave every result publishable whatever it says.
  • 2026-06-11, per-trade attribution fees. The builder attribution code was removed from every outbound venue link, as disclosed above, and no application Convexly has made to any venue program asks for fee share.
  • Paid placement, as a class (policy). Convexly does not accept payment to feature, promote, or rank any wallet, market, venue, or product, and declines such offers without negotiation.
  • A verdict sold to its subject (policy, and the shipped form). When a read is commissioned, the receipt that can be shared onward states what the record contains to a named block and carries no verdict, no figure, and no rating, so a subject cannot buy a document that reads as an endorsement.

How verdicts are produced

  • Fixed, published decision rules. The wallet skill verdict is a four-state machine (flagged, insufficient, luck, skilled) with fixed thresholds and a fixed gate order, with the thresholds and gate order published on this site and described in plain language in the top-50 skill scan. The same inputs produce the same verdict regardless of who is asking. Market Trust graduation is governed by a published promotion contract, not by discretion.
  • No point estimate without an interval. The realized entry edge is reported with its bootstrap 95% interval against zero in the same line, on the free check and on every published table. A cohort-level count of “skilled” records is reported with its multiple-comparisons correction alongside the uncorrected number.
  • Frozen methods, published hashes. The verdict pipelines are pinned to specific methodology hashes. Corrected 2026-08-10: this item previously said those pins live in a public repository and that the verdict artifacts are committed publicly. The repository is private, so neither the pins nor the commits are reader-inspectable, and that claim is withdrawn. The hashes themselves are published on the pages that rely on them, so a reader can check that a hash has not changed between readings even though they cannot browse the commit that set it. The hash-chained audit log is checkable in the browser at /research/verify.
  • Forward tests registered externally before the window opens. Any forward-looking claim must clear a test whose floors and pass/fail bars were filed externally, with a third-party timestamp, before the observation window opened. The verdicts mature on time and outcomes alone. The filing index, with receipts, is at /research/preregistrations.

Negative results

Convexly publishes its own failed tests and keeps them published. A forward test of the Edge Score's per-wallet predictive claim returned an out-of-sample Spearman of +0.11 (95% CI +0.05 to +0.18) against a +0.30 bar set in advance; it failed, and the result is public. The negative-result registry indexes every rejected hypothesis with its effect size and interval; the forward-validation page carries the honest forward result; the methodology page leads with the failures. A null or failed result is a publishable outcome, not a removable one.

Corrections

Errors in published numbers or labels are corrected in place, with the correction stated on the affected page rather than silently edited, and dated. Corrected 2026-08-10: this paragraph previously said the repository is public and that the edit history of every page is therefore inspectable. The repository is private, so that assurance was not true and is withdrawn. What a reader can actually check is the dated correction on the page itself, the pre-registration receipts at AsPredicted, which are externally timestamped and outside our control, and the hash-chained audit log. A worked example: a rolling rank-correlation series initially labeled as forward validation was relabeled as an in-sample diagnostic, and the correction note remains on the methodology page and on the diagnostic page itself. Where a corrected claim was sent to a specific recipient, the policy is to send the correction to the same recipient.

What customers pay for

Customers pay for tests and infrastructure, never for outcomes.

  • The first wallet check is free and requires no signup; checking more wallets is free with an account. See a wallet's verdict and its interval.
  • Subscriptions price tooling and volume: watchlists, exports, API access. A subscriber sees the same verdict on the same wallet as a non-subscriber; no plan changes a verdict.
  • Commissioned wallet reviews charge a fixed fee agreed before the review runs. The report states what the record supports, including “not separable from chance” and “insufficient evidence.” Paying for the review does not buy the conclusion.
  • No rated wallet, market, or venue can pay to be rated, to improve a rating, or to remove one.
  • If Convexly offers certification, the same rule applies as policy: the charge is for running the test, not for the result.

Identity

  • Verdicts are computed on public records keyed by wallet address.
  • A person or product name attaches to a record on Convexly only through a cryptographic self-claim: a message signature from the wallet, with smart-contract wallet support, re-verified on a 90-day cadence. The signature proves control of the address; it cannot move funds. The flow is documented at /research/wallet-verification-methodology.
  • Convexly never first-publishes a handle-to-wallet linkage. Names shown next to addresses are either self-claimed through the signature flow or are the venue's own public profile names for that address, reproduced as the venue publishes them.
  • Deanonymization is not a product.

Disclaimer

Convexly provides diagnostics on public records. It is not investment advice, not a trading instruction, and not a recommendation to follow, copy, or trade against any wallet or market. A past read is not a forecast. “Luck” means not separable from chance on the resolved record; it is a statistical statement, not an accusation. “Skilled” is a read on past data and is disclosed as uncorrected for multiple comparisons where that applies. Dollar PnL is not part of any verdict. Nothing on this site is an offer, a solicitation, or an endorsement of any trading venue.