Transparency
How we use AI
Basis Desk publishes AI-generated content and says so on every page. This is how the system works, what it is allowed to do, and where the limits are.
What the machine does
Basis Desk is an AI-native newsroom. Software reads primary sources — regulators, protocol teams, exchanges, filings, on-chain data — clusters related items into stories and drafts articles. A second, independent pass audits every factual claim in the draft against the sources it cites, checks numbers against live market data, scans for verbatim copying and flags legal or advice-like language. Only drafts that pass are published.
What it does not do
It does not invent sources, quotes or numbers: anything not supported by a cited source is removed before publication. It does not give financial advice, price targets or trading recommendations. It does not rewrite other outlets' reporting — media headlines are used only as a signal that something happened; the text comes from primary sources.
Labelling
Every article carries the byline 'Basis Desk Newsroom · AI-generated, source-verified' and lists its sources. Sentiment labels are the model's assessment of a story's tone, not a recommendation. We do not use invented human personas.
Humans
Stories are published automatically once they pass verification — no person reviews each article before it goes live. Chief Editor Julian Mercer is accountable for the system: he sets the editorial rules and the source list, monitors the pipeline, reviews every story the system holds back, spot-checks published stories against their sources and handles corrections. Report an error with the link on any article or at corrections@basisdesk.news — corrections are logged publicly.
Data providers
Rankings by Coinlore and CoinGecko; prices via Coinbase, Kraken and DefiLlama; Fear & Greed by alternative.me; fees from Ethereum mainnet and Blockstream; token security signals by GoPlus Security. Market data may be delayed or inaccurate.
Technical reports & open datasets (CC BY 4.0)
The verification architecture, entity-resolution grammar, and browser claim-checking benchmarks are documented in peer-style technical papers with live CSV/JSONL exports at basisdesk.news/research:
- BD-TR-2026-01: Primary-Source Grounding and Automated Claim Verification (PDF)
- BD-TR-2026-02: Cross-Venue Entity Resolution and On-Chain Whale Flow Attribution (PDF)
- BD-TR-2026-03: Client-Triggered Claim Verification and Smart-Contract Risk Heuristics (PDF)
Nothing on Basis Desk is financial, legal or tax advice. See the disclaimer.