Convexly Research / Start here
The research surface, by how much time you have.
Convexly publishes its own methodology in full: frozen coefficients, externally verified preregistration receipts where they exist, explicit receipt-pending labels where they do not, SHA-256 audit chains, and openly published negative results. The catalog at /research has every paper in chronological order. This page is the curated map for visitors who want a route in.
The two methodology pieces.
Every score and every signal on Convexly traces back to one of these two. The historical wallet papers are frozen-coefficient, and whether a reader can rerun one depends on whether its cohort is published: the Manifold half is, the Polymarket cohorts behind V1 and V1.5 are not. Forward tests are only called pre-registered when an external or manifest-verified filing exists.
Edge Score V1-M
Convexly's wallet-skill methodology. The current read is venue-specific refits with frozen coefficients per paper, not one pooled coefficient set. Polymarket is the scored venue today.
Methodology · 2026-04-22
Coherent Markets Engine V0.1
Convexly's structural-mispricing detector. Reads CLOB prices across linked Polymarket events, builds the implied joint distribution, and flags coherence violations. The public product label is Coherence Signals; it is unrelated to Chicago Mercantile Exchange.
Methodology · 2026-04-28
What we publish when our methods fail.
The strongest signal is what we publish when our methods fail. Most analytics companies bury negative results because failure is bad for marketing. Convexly publishes them on dedicated pages because the trust compounds.
Edge Score V1.5: both primary tests failed
Pre-registered V1.5 follow-up (AsPredicted #287368). Per-wallet temporal holdout (E2) and per-quarter IC stability (E7): both failed their ex-ante thresholds. Honest reframing: V3b is a cross-sectional ranker, not a per-wallet temporal predictor.
Negative result · 2026-04-27
MarketAlpha V2.8.2: 24-of-24 aggregators failed
Manifest entries #287436 + #287442 + #287714 track the frozen 24-aggregator sweep on the V1-M cohort; their public receipt URLs are not yet verified. Zero specifications beat the market-implied baseline. Substantive finding: PnL-skill on Polymarket is not a usable forecast-aggregation weight.
Negative result · 2026-04-27
Negative-results register
The full index of every pre-registered hypothesis Convexly has published as failed-or-null. Updated alongside every paper that runs.
Index
Coherence Signals track record
Daily resolved-PnL accumulator for Coherence Signals. Today's claims are framed as 'modeled net EV at prescribed size,' not realized PnL. Promotion to realized requires V0.1.6 + 30-90 days of resolutions + a passing random-control arm.
Live tracker
The full paper stack + the verification surface.
If you want to dig into the math, replicate the analysis, or check whether what we ship matches what we published.
Full paper catalog
Every Convexly research paper, sorted newest first. Edge Score V1, V1.5, V1-M, V1-M politics amendment; Coherent Markets Engine V0.1 + V0.2.0; V2-Perps; V2.8.2 honest-failures register; the 10K wallet study; the Gomez-Cram comparison; and more. Withdrawn work stays listed as withdrawn.
Index
Pre-registration index
Every Convexly manifest entry with verdict and receipt health. External links are only shown as receipts when the verifier can match the AsPredicted ID, title, and filing date.
Index
Audit-chain verifier
Browser-side SHA-256 hash-chain verifier. Paste any Convexly paper's audit-chain anchor and check that the published artifacts hash to what the chain records. Withdrawn artifacts are reported as withdrawn rather than as missing files.
Tool
V2-Perps research
In-research extension of the methodology to crypto perpetual derivatives. Companion: the cross-jurisdiction Hyperliquid Policy Center co-filing posture for the next CFTC rulemaking window.
Methodology · 2026-05-01
How Convexly publishes
Three disciplines, in order.
- 01 Frozen coefficients. Once a methodology paper ships, the scoring math is locked at a dated commit and the paper states it in full. Future tweaks ship as new versions; the old version stays computable exactly as published, and where its cohort is public a reader can rerun it too.
- 02 Forward-test receipt discipline. Future extensions are tracked in the public manifest with pass/fail thresholds before the data is touched. External AsPredicted links are treated as receipts only after the verifier confirms the ID, title, and filing date. The complete filing index, with verdict and receipt health for each entry, lives at /research/preregistrations.
- 03 SHA-256 audit chain. Every paper carries an audit-chain hash that links inputs (raw data) to outputs (scores, plots, conclusions). A third party can check in a browser that every published artifact hashes to what the chain records. Corrected 2026-08-10: that is an integrity check, not a rerun. Re-deriving an output also needs its input cohort, and the Polymarket cohorts behind V1 and V1.5 have never been published.
Together: methodology disclosed, math frozen, forward tests pre-registered only when the receipt exists, results published whether they pass or fail, every published artifact hash-verifiable. Verify it yourself.