Editorial Policy

AI-authored content policy

Per-coin summaries on Pharos detail pages are drafted by a large language model (currently Claude Opus 4.7 via the Anthropic API) against the underlying dataset Pharos already publishes: static profile fields, live reserve composition, depeg history, DEX-liquidity metrics, and the scoring outputs documented in the methodology. The model is prompted to summarize what is on the page, not to introduce facts the data does not support.

Every summary is reviewed by the editorial maintainer (@TokenBrice) before it ships. The review checks for factual drift, missing context, and tone consistency with the rest of the dashboard. The disclosure chip on each summary records who reviewed it, when, and the "facts as of" date — the point at which the underlying figures the prose relies on were valid. If the facts drift, the prose is rewritten on the next editorial pass.

Corrections work the same way every other data correction on Pharos does: open a GitHub issue or use the feedback link on the affected coin page. Verified corrections trigger a re-draft and a fresh review stamp; the chip updates with the new dates.

The dashboard's numeric outputs — scores, peg deviations, supply, liquidity, freezes — are computed by the worker pipeline, not by the model. AI only ships in the narrative panels, and every such panel is labelled.