CT⚡SignalFriday, October 2, 2026 · automated data desk

How CT Signal works

The whole system in four steps — and the rules that keep us honest.

1 · Listen

We watch Connecticut feeds (CT Mirror, Google News CT) and national feeds (NYT, NPR, PBS), plus a civic calendar (school year, tax deadlines, census releases). Headlines are deduplicated and tracked by age.

2 · Choose the question

A language model reads the headlines against our closed catalog of validated indicators and proposes only questions the catalog can actually answer. News arrives as language; data speaks in dataset codes — reading intent is the AI's entire job. An unmatched story produces nothing. If the model is unavailable, a keyword heuristic falls back; the catalog is the same either way.

3 · Fetch, never write

No number on this site was written or rounded by a language model. Answers come from executed queries against Data Commons and the Connecticut Open Data Portal (Socrata), dry-run validated before publication. Every story page shows the dataset link and the literal query.

4 · Publish

Cards land on the board, get their own permalink story page, the RSS feed, and the archive — every 15 minutes, via a scheduled job whose commits are the audit trail. If a fetch fails, the card falls back to labeled cache or doesn't run. We fail by absence, never by invention.

The honesty rules

Corrections

Found something wrong? Email us — we annotate the story, re-publish, and the change is visible in the commit history. See the corrections policy.