CRYPTO-PDF-AI-NARRATIVE-HARDENING (Evidence-Bound Output)

Release digest

CRYPTO-PDF-AI-NARRATIVE-HARDENING (Evidence-Bound Output) — Crypto PDF AI narratives now pass through a normalization step before rendering. The report keeps fixed narrative headings, appends a required Evidence Boundary caveat when needed, removes raw source URLs or contract addresses from model output, and rewrites unsafe claims such as guaranteed safety, verified truth, buy/sell recommendations, or audit-nonexistence wording into evidence-bounded decision-support language.

Added focused AI analyst contract coverage for fixed headings, required caveat copy, URL/address redaction, and unsafe-claim sanitization.
Updated docs/architecture/CRYPTO_REPORTS.md and knowledge-base/docs/crypto/dashboard-ui.md because this changes visible PDF narrative behavior. Public-docs decision: documented as evidence-bound narrative handling without claiming investment advice, verified truth, safety, or formal audit coverage.