FraudCoins open dataset

On-chain holder-concentration measurements and contract-permission data for cryptocurrency tokens, published as plain JSON under CC BY 4.0.

Snapshot 2026-10-27 — 1,921 assets scored · 519 with an on-chain holder measurement (115 taken on the snapshot date) · 489 with wallet-level correction evidence · 410 with at least one contract permission reported (520 contracts queried).

What this adds

A raw "top 10 holders hold X%" figure counts whatever the ten largest addresses happen to be. In practice those routinely include an exchange's omnibus wallet, a liquidity pool, a bridge contract, a locked treasury or a burn address — which are very different things. An exchange wallet is custodied for thousands of people and a burned balance is gone for good, while a treasury or vesting contract is very much someone's to release later.

This dataset classifies each of the ten largest holders first, publishes the share held by private wallets only, and — the part that makes it checkable — names each wallet set aside, with its classification and stake, so the correction can be verified on a block explorer rather than taken on trust.

v1/concentration.json         raw vs corrected top-10 share per token
v1/excluded-holders.json      the wallets removed from each figure, and why
v1/contract-permissions.json  owner privileges + structural contract properties
v1/scores.json                composite risk score, level and flags
v1/history.json               concentration series (see convention 2)
v1/coins-index.json           id / name / symbol for every asset covered
v1/alerts.json                concentration events by day (see note below)
v1/index.json                 discovery document: endpoints, licence, cadence

Reading alerts.json correctly: most entries carry kind: "first-measured" — the asset was measured for the first time and was already above the threshold. Only kind: "crossed" means it moved across the threshold between two measurements. In the current snapshot that is 95 vs 6. Reading a first-measured entry as "this project newly deteriorated on this date" would be wrong, and it names real projects, so the distinction matters.

index.json, excluded-holders.json and contract-permissions.json are also served live at https://fraudcoins.com/dataset/v1/; the other five sit at the site root (https://fraudcoins.com/concentration.json, etc.). index.json is the authority for every endpoint URL. CORS is enabled — no key, no registration, no rate limit.

Four conventions that matter more than the field list

1. null means no measurement — never a clean result. An unreported contract permission is unknown, not absent. 45 of the 188 queried contracts returned no permission fields at all.
2. A flat interval in history.json means no change was observed, not that no measurement occurred. Consecutive identical readings are not re-emitted. Each point is a reading whose value differed from the one before it.
3. The date field differs per file: sampledAt in concentration.json, asOf in scores.json, measuredAt in contract-permissions.json. Read it rather than assuming a figure is current.
4. Flag thresholds are per tier, defined in concentration.json's tiers array — 50% for assets above 50 million USD market cap, 70% for the rotated 10–50 million USD tier where high concentration is close to ordinary. A flag also requires at least 50 holder records.

Coverage is uneven, on purpose

The pipeline runs daily and re-measures assets above 50 million USD in rotation, each every few days. The 10–50 million USD tier is measured only when capacity allows, so those readings are older. The stats line at the top says how many rows were taken on the snapshot date; the rest carry an earlier date in sampledAt. Correction evidence exists for the assets whose most recent measurement recorded it.

Rows with fewer than 10 usable holder records are not published at all: a "top ten" computed from four addresses is not a meaningful quantity.

Limitations worth stating plainly

• Holder classification relies on public address labels. An unlabelled custodial wallet is counted as private, so every correction here is a floor, not a complete accounting.
• Excluded supply is not automatically harmless. Burned tokens are gone, but treasury and vesting holdings are excluded too and can reach the market when they unlock. A low private-wallet figure is not a low future sell-side risk.
• Contract data comes from GoPlus Labs' static analysis of the deployed contract. We do not decompile bytecode ourselves, and the field is reported as they returned it. Trade-simulation verdicts (honeypot, cannot-sell-all, taxes) are deliberately excluded: they produce false positives on legitimate, freely-traded assets, and republishing one as fact would be defamatory.
• isProxy and isOpenSource are not owner privileges. Upgradeable proxies are the standard deployment pattern for many established protocols.
• Risk scores and flags are automated, opinion-based assessments derived from public market data — not statements of fact, financial advice, or allegations of illegal conduct against any project or person.

Licence & citation

CC BY 4.0 — free to use, including commercially, with attribution.

FraudCoins.com (2026). On-chain holder-concentration dataset.
https://doi.org/10.5281/zenodo.22494421

The DOI above is the concept DOI — it always resolves to the latest archived version, so a citation using it stays correct as the dataset is re-released. Each release also receives its own version DOI if you need to pin an exact snapshot. Archived at CERN's Zenodo.

Methodology: https://fraudcoins.com/methodology/ · Corrections: https://fraudcoins.com/corrections/