Skip to content
Convexly

Downloads and citation

Prediction market data and datasets

Every dataset, paper and audit artifact Convexly publishes, in one place, with a checksum and a citation line for each. Free, no account, no email required.

Sizes and checksums are computed from the published file each time this page is served, so they describe the exact bytes behind the link rather than a value recorded at some earlier date. Record counts are as last measured. If a checksum here does not match the file you downloaded, the file is not the one we published, and we would like to know: research@convexly.app.

Datasets

  • Manifold forecaster calibration dataset (15,106 users)

    Per-forecaster aggregates for 15,106 Manifold users: resolved bet counts, binary and multi splits, unique events, average Brier score, a matched baseline Brier, the difference between them, realised profit and loss, largest single-event result, concentration and win rate. Forecaster identity is a hash.

    This dataset is withheld pending Manifold's written position on data rights, dated 2026-09-04. The aggregate results on the V1-M page are unchanged.

    Manifold's terms are unresolved for this use: the public bulk dump is licensed for personal and non-commercial use only, Manifold's terms reach derived content, and a commercial licence is offered which Convexly does not hold. No redistribution or sublicence is granted, and the per-user data files were withheld on 2026-09-04 pending Manifold's written position on data rights.

  • Scored Polymarket wallet cohort

    The published cohort artifact. Each row carries its Edge Score components, realised edge where one could be computed, its 95 percent interval, and a stamp saying whether the underlying read was proved complete or is only a window over recent activity. The file declares its own cohort rule and ordering in its cohort_source, ordered_by and ordering_note fields; read those and the per-row completeness stamp before using it. It is a closed set frozen for study, never a live screen and never a recommendation.

    Format
    application/json
    Size
    93.9 KB
    Records
    50 wallets (as measured)
    Licence
    Free to read and cite, no redistribution licence

    Derived from public Polymarket sources. Free to download, read and cite, with attribution. No redistribution or sublicence is granted: Polymarket's terms state their restriction reaches data in on-chain and derived form, that question is unresolved, and we will not grant onward rights we have not established we hold.

    sha256 b094d2967a0cc869978147ff45ca461d4c05b9767d2a5ce1d3ee5dfa975d5c0a

    Cite as

    Convexly Research. Scored Polymarket wallet cohort. 2026. https://www.convexly.app/data#polymarket-scored-cohort

  • Polymarket wallet label index

    Address-to-label mappings read from public Polymarket sources only, with the provenance of each label. Single-venue by scope: no other venue is read, so no other venue's market counts, identity lookups or names appear in any field. Entries carry the public Polymarket display name where one is published.

    Format
    application/json
    Size
    109.1 KB
    Records
    n/a
    Licence
    Free to read and cite, no redistribution licence

    Derived from public Polymarket sources. Free to download, read and cite, with attribution. No redistribution or sublicence is granted: Polymarket's terms state their restriction reaches data in on-chain and derived form, that question is unresolved, and we will not grant onward rights we have not established we hold.

    sha256 341af086c88e36637025322930cc30cf3a6d73123f670a89b5ae24b958dcb77b

    Cite as

    Convexly Research. Polymarket wallet label index. 2026. https://www.convexly.app/data#wallet-labels

Key-gated feeds

Not downloads. These are queryable API surfaces behind an API key, listed here because their coverage receipts are public and belong beside the other verifiable artifacts on this page.

  • Resolved-markets settlement feed (key-gated)

    A queryable feed of settled conditions read from the Polygon blockchain: one row per ConditionResolution event captured on the shared ConditionalTokens contract, carrying the on-chain payout vector, the winning outcome index, the reporting oracle address, the question id, and the settlement block. It is a contract-wide read, not a venue read: 77 distinct reporting addresses were in the store when that was measured on 2026-08-25, and every row carries its oracle so you can partition the stream yourself. Chain-only by default: unless metadata is explicitly requested, a query addresses no venue-derived table at all.

    Key-gated (X-API-Key). Included from the Researcher tier upward under a daily row quota; Trader carries the higher quota. Keys are generated from Settings.

    Access
    API key, plan-gated
    Licence
    Chain-derived, no redistribution licence

    Derived from the public Polygon blockchain. Served rows are for your own analysis and citation with attribution. No redistribution or sublicence is granted: whether any venue's terms reach data in on-chain and derived form is an unresolved question, and we will not grant onward rights we have not established we hold. NOT MIT.

    Coverage receipt package

    The coverage receipt package below is public and reproducible: the manifest states, per 100,000-block range, an event count and content digests over the span the sweep has read, and the checker re-derives them against any Polygon RPC endpoint you choose (Python 3.9+, standard library only). The two sub-floor rescans record, window by window on two unaffiliated endpoints, that every block below the declared floor was read empty, with errored reads recorded separately so a refused window cannot pass as an empty one. One disclosure on the manifest copy served here: its provenance.publisher field reads Convexly, where the repository's archival copy carries the operating company's registered name, which is not published on public surfaces; every range digest and the manifest root are byte-identical between the two.

Papers and methodology

  • Edge Score methodology V1-M

    The frozen methodology behind Edge Score, with the cross-venue invariance extension: what the composite measures, the coefficients, the sample floors, and the tests that were run against it including the ones that failed.

    Format
    application/pdf
    Size
    339.4 KB
    Records
    n/a
    Licence
    All rights reserved, free to read and cite

    Convexly's own writing. Free to read, quote and cite with attribution. Not licensed for republication in whole.

    Cite as

    Convexly Research. Edge Score Methodology V1-M. 2026. https://www.convexly.app/research/edge-score-methodology-v1m

  • Edge Score methodology V1

    The original frozen methodology, retained so a reader can see what changed and when rather than only the current version.

    Format
    application/pdf
    Size
    111.4 KB
    Records
    n/a
    Licence
    All rights reserved, free to read and cite

    Convexly's own writing. Free to read, quote and cite with attribution. Not licensed for republication in whole.

    Cite as

    Convexly Research. Edge Score Methodology V1. 2026. https://www.convexly.app/research/edge-score-methodology-v1

Audit and provenance artifacts

These are the objects that let someone check our work rather than take it on trust: what we registered before running a test, what we found when the answer was no, and the append-only log of what changed.

  • Published negative results

    The tests that did not work. Each entry states what was predicted, what was found, and what was retired as a result. Published because a research surface that only reports its wins is not evidence, it is marketing.

    Format
    application/json
    Size
    14.0 KB
    Records
    4 entries
    Licence
    MIT

    Convexly's own computed output. MIT: use, modify and redistribute freely, with attribution.

    sha256 b4cf7aea14cd1c58b64997e3482eb07a70e2136822c720342e08a36e215b3e14

  • Registration receipt index

    The index of externally registered forward tests: what was filed, when, the window it covers, and where the third-party receipt lives. A method registered before the outcome is known is checkable; one described afterwards is not.

    Format
    application/json
    Size
    31.3 KB
    Records
    15 entries
    Licence
    MIT

    Convexly's own computed output. MIT: use, modify and redistribute freely, with attribution.

    sha256 7cb0553f234b346438b9c4202436645f046febc6465d229991907e002f180f87

  • Append-only integrity log

    The running record of what changed in the published artifacts and when, one JSON object per line, append only.

    Format
    application/x-ndjson
    Size
    54.4 KB
    Records
    116 entries
    Licence
    MIT

    Convexly's own computed output. MIT: use, modify and redistribute freely, with attribution.

    sha256 f6f979e29ff749cb6fa15650b8f4a95b456f13d66e0d627a5ed7a8e72983227c

Using these in research

Licence is stated per artifact above, and it is not the same for all of them. The three audit artifacts are MIT: they are our own computed output and you may redistribute them. The two Polymarket artifacts are free to download, read and cite with attribution, but carry no redistribution or sublicence grant, because they are derived from venue sources whose terms we have not established give us the right to grant one. If you need onward redistribution rights, the venue is who can give them, not us.

The Manifold cohort bundle and its per-user file were withheld on 2026-09-04 pending the venue's written position on data rights. They are not available for download. The published results remain unchanged; the notice above records the affected files and hashes.

The two Polymarket artifacts are not pseudonymised and do not claim to be. They carry public on-chain addresses, which are public by construction on a public ledger, together with the public Polymarket display name where that venue publishes one. Nothing in them comes from a private source, and no email, payment detail or off-venue identity appears in any file here.

If you are writing something that uses these and want a detail checked, or a figure regenerated at a different cut, write to research@convexly.app. There is no charge and no form.