Crypto AI Adaptation Research Report — NMR (Numeraire) · WhiteSwan Crypto constituent
EventTrader research desk · generated 2026-11-22 07:35 UTC · report ai-adapt-crypto-white-nmr-20261122 · built from the public-data evidence pack by the platform
| Project | NMR — Numeraire · CoinGecko numeraire |
| Book | AMF-WHITESWAN-CRYPTO (long) — WhiteSwan Crypto |
| Composition algorithm | Swan Composite v1.1.0 — 67% AI Adaptation Score + 33% market verdict versus BTC, conviction-weighted |
| Level | 4 — greenfield — built with AI from the ground up · model-confirmed |
| AI Adaptation Score | 82.4 / 100 · run 2026-11-22 |
| Place in the book | rank 5 of 10 · composite 73.3 · market verdict 55.2 · weight 8.4% |
Verdict
- NMR is a WhiteSwan constituent on scoring run 2026-11-22: level 4 — greenfield — built with AI from the ground up — because AI is the protocol AND the project was built around AI from the ground up [S1].
- Its crypto AI Adaptation Score is 82.4 / 100 (the highest tail of the AI-claiming universe) [S1].
- Level confirmation: a model read confirmed the level [S1].
What the project says about AI
- Claim layer: AI in the name no; AI category on CoinGecko yes (Smart Contracts, Ethereum (ETH) Token (ERC-20), AI (Artificial Intelligence), Recruitment, Education, Finance & Banking, DeFi, Assets Management); AI markers on the homepage 3 [S1].
- Substance: 17 AI sentences across 7 fetched documents (4,163 words): 1 describe AI as the protocol, 0 state an AI origin (built from the ground up), 0 proprietary models / data, 0 bolt-on feature language, 1 name a third-party AI vendor, 0 deployment, 0 with numbers [S1].
Verbatim from the project's own materials:
"Numeraire (NMR) is a cryptocurrency used as a staking token for the Numerai platform, which enables data scientists to build machine learning models for financial predictions." — profile [S2] "Get API Keys Codex Claude Cursor Copy $ # clone example-scripts, install MCP, and prompt Codex $ git clone git@github.com:numerai/example-scripts.git $ cd example-scripts && curl -sL https://numer.ai/install-codex-mcp.sh | bash $ codex exec --yolo "find the best neural network architecture to predict target jupiter" Start with hedge fund quality data." — homepage [S3] "Just like the weights of a neural network can be tuned with training data." — docs [S5] "Numeraire: A Cryptographic Token for Coordinating Machine Intelligence and Preventing Overfitting Richard Craib, Geo ↵rey Bradway , Xander Dunn with Joey Krug https://numer.ai February 20, 2017 Abstract Machine learning competitions are susceptible to intentional overfitting." — whitepaper [S6] "Apply machine learning to predict the stock market." — homepage [S3] "You build machine learning models on it." — docs [S5] "Modeling Your objective is to build machine learning models to predict the target given the features ." — docs [S5] "The auction mechanism also yields natural arguments for the economic value of a Numeraire token. 1M o t i v a t i o n A common approach to verify accuracy in machine learning is to break the dataset into train and test sets." — whitepaper [S6]
What can be verified
- Agent card (A2A / MCP) on the project's own domain: none; llms.txt: found; embedded AI support widget: none [S1].
- Repositories inspected: 1; ML frameworks in dependency manifests: none; last push unknown; stars 0; commits in the last 4 weeks unknown [S1].
- Verified-evidence points: 1 (ML code, agent card, widget, llms.txt, README depth, deployed products with numbers) [S1].
Score and level
- The crypto AI Adaptation Score weighs the level (0–4) at 45%, greenfield (built with AI from the ground up, level 3+) at 20%, credibility at 15%, trajectory at 12% and developer activity at 8%, the last two relative to the crypto universe [S1].
- NMR: level 45.0, greenfield 18.0, credibility 10.5, trajectory 5.8, developer activity 3.2 points → 82.4 [S1].
- Model read: level 4 (greenfield), confidence 0.72; sub-scores product 8.0, code evidence 1.0, data moat 7.0, credibility 7.0, greenfield 9.0 (each 0–10) [S1].
- Model summary: "NMR is the staking token purpose-built in 2017 to coordinate a crowdsourced machine-learning hedge-fund tournament — AI is the protocol itself, conceived around machine intelligence from inception. Mechanism, metrics and named products are described, though no ML frameworks or agent card were verified in repositories." [S1].
"Numeraire: A Cryptographic Token for Coordinating Machine Intelligence and Preventing Overfitting" — model evidence, whitepaper [S1] "Numeraire (NMR) is a cryptocurrency used as a staking token for the Numerai platform, which enables data scientists to build machine learning models for financial predictions." — model evidence, profile [S1] "By performing well in Numerai’s machine learning competition, data scientists will earn Numeraire on an ongoing basis." — model evidence, whitepaper [S1] "Use frontier AI to predict the stock market." — model evidence, homepage [S1] "The performance evaluation metric is logloss 2, a suitable metric for binary classification problems like Numerai’s machine learning competition." — model evidence, whitepaper [S1]
- Market: cap $53.2M, 24 h volume $5.8M, rank 368 (CoinGecko, as of the run) [S1].
Risks to this view
- An AI-native protocol's token can still fall with the market; the score measures adaptation, not price [S1].
- Origin statements are the project's own; the greenfield sub-score is scored only from them and could be marketing [S1].
Place in the book
- Rank 5 of 10 in AMF-WHITESWAN-CRYPTO (Swan Composite v1.1.0, scoring run 2026-11-22): the book holds the highest composites [S1].
- Composite: 0.67 × 82.4 (AAS) + 0.33 × 55.2 (market) = 73.3 [S1].
- Market verdict 55.2: over the last 90 days NMR rose more and fell less than BTC (down-capture 0.75, up-capture 0.96, L = 0.21; 49 down / 41 up days) — the Failure-To-Adapt fund's selection signal, 50 = moves with BTC [S1].
- Weight: rank factor 1.56 (rank 5 of 10) × market factor 1.05, normalised across the book and clamped to 5–20% → 8.4% [S1].
Sources
- S1 — EventTrader crypto AI Adaptation Score run 2026-11-22
- S2 — profile — https://coinpaprika.com/coin/nmr-numeraire/ — https://coinpaprika.com/coin/nmr-numeraire/
- S3 — homepage — https://numer.ai/ — https://numer.ai/
- S4 — llms_txt — https://numer.ai/llms.txt — https://numer.ai/llms.txt
- S5 — docs — https://docs.numer.ai/ — https://docs.numer.ai/
- S6 — whitepaper — https://static.coinpaprika.com/whitepapers/nmr-numeraire/whitepaper.pdf — https://static.coinpaprika.com/whitepapers/nmr-numeraire/whitepaper.pdf
- S7 — repo — https://github.com/numerai/contract — https://github.com/numerai/contract
- S8 — readme — https://github.com/numerai/contract#readme — https://github.com/numerai/contract#readme
Method and disclaimer
The crypto AI Adaptation Score reads each project's PUBLIC materials only: its CoinGecko profile (description, categories, links, genesis date, developer data), its homepage (text, AI markers, embedded AI widgets), A2A / MCP agent cards and llms.txt on its own domain, its whitepaper and docs, and its GitHub repositories (push recency, READMEs, the ML frameworks named in dependency manifests). Deterministic features set a level on a 0–4 rubric (no mention · AI in name only · bolt-on · AI-native · greenfield, built with AI from the ground up); a model read of the same evidence confirms the level and scores credibility and greenfield; the score combines level, greenfield, credibility, trajectory and developer activity, with greenfield carrying the highest reward. The books are composed by Swan Composite v1.1.0: each coin's composite is 67% AI Adaptation Score + 33% market verdict — the market verdict is the Failure-To-Adapt fund's selection signal, the signed down/up capture ratio of the coin against BTC over 90 days (a coin that falls more and rises less than BTC scores below 50). AMF-BLACKSWAN-CRYPTO holds the ten lowest composites among coins that claim AI that pass the liquidity screen, the new-listing gate and the crowded-short gate, and AMF-WHITESWAN-CRYPTO the ten highest, weighted by conviction (rank and market agreement, each leg 5–20%), recomposed on every scoring run. Every quote is the project's text verbatim; every factual sentence names its source; a document the platform could not reach is reported as missing.
This report is information, not investment advice. Figures come from the cited sources as of their dates. Nothing here is a recommendation to buy, sell or hold any asset.