Skip to main content

Numerai / Numeraire

Numeraire / Numerai (NMR) — a reference card for a crypto project: what the product does and how verifiable the project’s key claims are; also — alignment with the Decentralized AI thesis (DecAI Fit).

Numerai — an ML tournament (data science competition) where participants build models and submit predictions, while NMR is used for staking on predictions and burn in case of poor performance. The NMR token is an ERC‑20 on Ethereum.

Disclaimer

Important: this page is a reference overview of the project. Within DecAIHub, we separately record a verification based on open sources within the AI thesis (what is real AI/onchain/why a token, and which sources this is based on).
This is not financial advice and not a recommendation to buy/sell/hold any asset. We do not assess price, returns, or entry/exit timing.
Conclusions are based on publicly available materials found as of the verification date. The absence of confirmation does not imply the absence of the technology.

TL;DR (for readers who are not into AI)

  • What it is: a competition of ML models where participants submit predictions and the system computes their quality.
  • Who it’s for: data scientists/ML engineers and users/participants of the tournament platform.
  • Why onchain/token: NMR is used as “skin in the game” — it is locked (stake) and can be destroyed (burn), and the facts of staking/burning are verifiable onchain.
  • What you can actually touch (Tier‑1): the Numerai tournament platform and the official documentation (workflow/rules/metrics).

How it works (very briefly)

  1. The user does: trains an ML model and submits predictions (submissions).
  2. The system does: evaluates the quality of the predictions by the tournament metrics and (optionally) takes the NMR stake into account.
  3. The user gets: results/scoring; with a stake — a reward in NMR or a burn of part of the stake.
  4. Where the blockchain is: the NMR contract holds the stake and ensures verifiable destruction of tokens (burn) based on the results.
  5. Token role: NMR — a staking/burning mechanism for coordination and for combating overfitting.

Quick links: AI thesis fit

Project passport

Field Value
Project name Numerai / Numeraire
Also known as (aliases) NMR | Numeraire
Logo (logo) ../assets/35_numeraire_logo.svg
Token symbol (ticker) NMR
Category AI/Other
Primary chain (chain) Ethereum
Type EVM
Network status mainnet
Consensus (consensus) PoS (Ethereum)
Launch date (launch_date) 2017
Project status (status) active
Token (standard) ERC-20
Token contract 0x1776e1F26f98b1A5dF9cD347953a26dd3Cb46671
Token model (token_type) utility (staking + burning)
Max supply 21,000,000 NMR (supply cap)
Issuance fixed cap + weekly mint until cap (per the whitepaper)
Tokenomics (tokenomics, brief) NMR is staked on predictions; in case of poor performance the stake can be destroyed (burn), which is publicly verifiable onchain
Governance —
Admins/keys —
Audit yes
Bug bounty yes
Team (team) —
Funding (investors/funding) —
AI component (ai_component) ML tournament + staking on predictions
Use‑case (use_case) incentivize generalizable ML predictions
DecAI Fit — AI thesis fit 5/6 — high alignment with the Decentralized AI thesis
DecAI Fit component summary 3 / 3 / 1 / 3
Last verified 2026-02-11

General project overview

Numerai describes itself as a data science competition: participants build machine learning models and submit predictions on “live data”. The system scores the results (the documentation describes the metrics and the process), and participants can stake NMR on their model signal: with good results they receive rewards, with poor ones — part of the NMR may be destroyed (burn). In the whitepaper, the motivation is formulated as combating (including intentional) overfitting in machine learning tournaments through economic incentives.

Features/modules

  • Dataset + live data: provision of data for training and daily submissions (docs).
  • Scoring: metrics for evaluating predictions (docs).
  • Staking: the mechanism of staking NMR on a model/prediction and the subsequent reward or burn (docs + contract).
  • Onchain token contract: the ERC‑20 NMR contract and the staking/burning/tournament functions (Etherscan + repository).

Typical scenarios

  • Scenario 1: a participant trains a model → generates live predictions → uploads a submission → receives scoring by the metrics.
  • Scenario 2: a participant stakes NMR on their signal → after the evaluation period receives an NMR reward or loses part of the stake through a burn.

Getting started (wallet/network)

  • What you need: a Numerai account + (if staking is desired) NMR in an Ethereum wallet, or participation via the platform’s mechanisms.
  • Network/chain: Ethereum (NMR — ERC‑20).
  • Fees: gas in ETH for onchain operations with NMR (if staking/transfers are done onchain).

Architecture/tech (non‑AI)

  • Components: the offchain tournament platform (data/scoring/submissions) + the onchain NMR contract for recording stakes and burns.
  • Where logic runs: the ML part (training and generating predictions) is performed offchain by participants; onchain records the economic outcome (stake/burn).

Token/economics (expanded, no forecasts)

  • Why stake: the whitepaper describes an “auction/staking” mechanism, where the stake reflects the participant’s confidence in the model’s generalization and makes intentional overfitting economically unprofitable.
  • Supply cap: the whitepaper states that the smart contract has an upper limit of 21M NMR; a weekly mint until the cap is reached is also described.
  • Burn: the whitepaper emphasizes that the destruction of NMR is publicly verifiable on Ethereum and contrasts this with USD (“USD cannot be burned”).

Security and incidents

  • Audits: the contract repository lists audits (including Trail of Bits for “NMR 2.0”).
  • Bug bounty: the bounties rules are published (docs).

Project status and timeline

  • Current status: the tournament and the token are operating (docs + Etherscan).
  • Key milestones: the whitepaper is dated 2017‑02‑20; the NMR contract is verified on Etherscan (2017‑06‑20 in Etherscan data).

AI component of the project

  • What is real AI here? Participants build ML models that predict the target variable on the provided data; the docs describe the submissions and scoring process.
  • What is truly onchain? The NMR stake and the verifiable destruction (burn) of tokens via the NMR contract; the address and code are available on Etherscan.
  • Why is a token needed? NMR provides “skin in the game”: the ability to stake and irreversibly burn, which is used to coordinate participants’ behavior and reduce incentives for overfitting.
  • Red flags (based on the information found): onchain‑washing is lower (the contract is actually used), but it is important to re‑check the trust in the offchain evaluation of results and in the process that triggers the burn/release of the stake (centralization of arbitration/evaluation).

DecAI Fit component summary

  • DecAI Fit: 5/6 — high alignment with the Decentralized AI thesis
Component Level Score
AI reality high 3
Onchain necessity high 3
Token necessity low 1
Evidence Tier‑1 present 3
Details

AI reality

AI_0 — AI intent (Tier‑1 roadmap) (1/0)

  • AI_0 (decision): 0
  • Claim (one line): AI/ML is the current product (the tournament), not only intent.
  • Evidence: Tier‑1 — https://docs.numer.ai/
  • Why sufficient / why not (one line): the docs describe the current tournament and the ML process.
  • Conflict note (if any): —

AI_1 — what the AI does (1/0)

  • AI_1 (decision): 1
  • Claim (one line): participants build ML models that predict the market; the platform computes the quality of the predictions.
  • Evidence: Tier‑1 — https://docs.numer.ai/
  • Why sufficient / why not (one line): the Overview describes the goal: “build machine learning models to predict the stock market”.
  • Conflict note (if any): —

AI_2 — demos/examples (1/0)

  • AI_2 (decision): 1
  • Claim (one line): the docs contain code examples and step‑by‑step instructions for training and submitting predictions.
  • Evidence: Tier‑1 — https://docs.numer.ai/
  • Why sufficient / why not (one line): there are examples of the pipeline (data → model → submission).
  • Conflict note (if any): —

AI_3 — metrics/quality (1/0)

  • AI_3 (decision): 1
  • Metrics: CORR, MMC (described as the key scoring metrics)
  • Context (version/date/how measured): the metrics are applied to tournament submissions; the description and definitions are given in the docs.
  • Evidence: Tier‑1 — https://docs.numer.ai/
  • Why sufficient / why not (one line): there are quantitative quality metrics and a description of the context of their application.
  • Conflict note (if any): —

AI_4 — tech description/architecture/versions (1/0)

  • AI_4 (decision): 0
  • Claim (one line): there is no Tier‑1 technical description of a “Numerai model” as a single AI artifact (it is a tournament, not one model).
  • Evidence: Tier‑1 — https://docs.numer.ai/
  • Why sufficient / why not (one line): the platform describes the rules and metrics, but does not record a single model/versions.
  • Conflict note (if any): —

Onchain necessity

OC_1 — contracts/addresses/chain (1/0)

  • OC_1 (decision): 1
  • Chain: Ethereum
  • Addresses/identifiers: NMR token: 0x1776e1F26f98b1A5dF9cD347953a26dd3Cb46671
  • Evidence: Tier‑1 — https://etherscan.io/token/0x1776e1F26f98b1A5dF9cD347953a26dd3Cb46671
  • Why sufficient / why not (one line): Etherscan confirms the address and code of the contract.
  • Conflict note (if any): —

OC_2 — onchain beyond just a token (1/0)

  • OC_2 (decision): 1
  • What useful logic is onchain: the contract contains the stake/release/destroyStake functions and tournament/round logs.
  • Evidence: Tier‑1 — https://etherscan.io/token/0x1776e1F26f98b1A5dF9cD347953a26dd3Cb46671
  • Why sufficient / why not (one line): this is protocol staking/burning logic, not only ERC‑20 transfers.
  • Conflict note (if any): —

OC_3 — onchain is critical (1/0)

  • OC_3 (decision): 1
  • “Replace the blockchain” test: without onchain there would be no publicly verifiable irrevocable burn/recording of stakes; trust would shift to the operator.
  • Evidence: Tier‑1 — https://numer.ai/whitepaper.pdf
  • Why sufficient / why not (one line): the whitepaper emphasizes the public verifiability of token destruction on Ethereum.
  • Conflict note (if any): —

Token necessity

TK_0 — no token (1/0)

  • TK_0 (decision): 0
  • Claim (one line): the NMR token exists as an ERC‑20 on Ethereum.
  • Evidence: Tier‑1 — https://etherscan.io/token/0x1776e1F26f98b1A5dF9cD347953a26dd3Cb46671
  • Why sufficient / why not (one line): the token and the contract are verifiable via Etherscan.
  • Conflict note (if any): —

TK_1 — payment for the service (1/0)

  • TK_1 (decision): 0
  • Claim (one line): NMR is used not as payment for a service, but as a staking/burning instrument.
  • Evidence: Tier‑1 — https://numer.ai/whitepaper.pdf
  • Why sufficient / why not (one line): the model is described as staking on predictions, not payments for a service.
  • Conflict note (if any): —

TK_2 — incentives/collateral/penalties (1/0)

  • TK_2 (decision): 1
  • Claim (one line): participants stake NMR and risk its destruction in case of a poor result (penalty/burn).
  • Evidence: Tier‑1 — https://numer.ai/whitepaper.pdf
  • Why sufficient / why not (one line): the whitepaper describes staking and destruction/burn.
  • Conflict note (if any): —

TK_3 — governance (1/0)

  • TK_3 (decision): 0
  • Claim (one line): no Tier‑1 onchain governance mechanics via NMR were identified.
  • Evidence: Tier‑1 — https://numer.ai/whitepaper.pdf
  • Why sufficient / why not (one line): the role of NMR is described as staking/burning, not governance.
  • Conflict note (if any): —

TK_4 — token is replaceable (1/0)

  • TK_4 (decision): 0
  • Claim (one line): the burn/stake logic in the description is tied to a crypto token, not to USD.
  • Evidence: Tier‑1 — https://numer.ai/whitepaper.pdf
  • Why sufficient / why not (one line): the whitepaper explicitly states that “USD cannot be burned”.
  • Conflict note (if any): —

Evidence (table)

Tier Source What it proves
Tier‑1 https://docs.numer.ai/ the ML tournament: data → models → submissions → scoring → staking
Tier‑1 https://numer.ai/whitepaper.pdf the motivation and the staking/burning model (including the 21M supply cap)
Tier‑1 https://github.com/erasureprotocol/NMR the NMR smart contract repository + links to audits
Tier‑1 https://etherscan.io/token/0x1776e1F26f98b1A5dF9cD347953a26dd3Cb46671 the onchain address/code of NMR and the associated functions

No comments to display

Back to top