Transparency
Trust comes from a visible record of what changed, which evidence was used, and where the system failed.
Modeled workspaces identify their operating mode; substantive engine changes require regeneration, validation, and the repository sync checks before release. Result-level provenance remains an active improvement where a workspace does not yet expose every version and timestamp.
Prefer primary public records, original research, and official statistics. Independent reporting supplies context. Provider summaries never replace the original link. Unresolved LLM news URLs are quarantined rather than published as observed reporting.
Provider-assisted persona reasoning, digest generation, and Ma’at source discovery can send a bounded prompt or archived story fields to configured external model providers within a hard reserved daily budget. Output is labeled and operationally cached, not stored as raw-text analytics. Ma’at currently suggests links; it does not independently retrieve them or issue verdicts.
Verified operator identity, funding, material relationships, and contact information must be published here after owner approval.
Report a source, data, accessibility, or model problem through Feedback. Minor corrections are logged in place with a date. Material result changes add a visible notice and model/site version entry; withdrawn evidence remains identified as withdrawn.
Report a problem →