Transparent automation
AI and automation policy
What the machines may do, what they may not do, and how Archynetys fails safely.
Automation used on this site
Software collects coverage metadata, clusters related headlines, calculates velocity, assigns lifecycle states, tracks language spread, builds charts, makes bounded predictions and grades those predictions later. These measurements are deterministic and use stored observations.
Where generative AI is used
Generative AI may draft a concise trend explanation, FAQ and tags from the source headlines attached to one cluster. The prompt forbids invented numbers, quotations, people, places, organizations, causes and outcomes. When the available headlines do not establish a fact, the system must avoid asserting it.
Verification before publication
A separate verification pass compares every factual sentence with the stored source headlines. Unsupported sentences are removed. If too much of a draft fails, the AI draft is rejected and a deterministic extractive explanation is published instead. Every trend page labels the path actually used.
Quality quarantine
Automation is allowed to detect broadly but not to publish broadly. The Quality Brain scores evidence strength and keeps developing signals out of homepages, internal recommendation links, sitemaps, feeds and public trend APIs. This prevents the system from expanding the search index merely because it can generate text.
Learning with guardrails
The prediction system records the source count, velocity-to-peak ratio and lifecycle state that existed when each prediction was made. After real outcomes arrive, it may choose a better policy from a small approved range—but only after at least 100 graded samples spanning seven days and at least a two-point balanced-accuracy improvement. It cannot rewrite articles, sources or corrections.
Human contact
Archynetys uses an honest organizational byline rather than inventing a human author. Questions and error reports reach the Archynetys Intelligence Desk.