How Signal decides what matters.

Signal reads publicly available feeds: company and lab blogs, research preprints, policy and security advisories, practitioner writing, Hacker News, GitHub, and selected YouTube channels. Sources are not added automatically. Each one is verified before it can be ingested. There are two narrow exceptions, and both appear as explicitly unverified community evidence: GitHub and YouTube links discovered through existing feeds, and articles on a new subject that many independent sites start covering at once, before any verified outlet has caught up. Every source carries a tier:

• Primary: the organisation or person the news is about, announcing it themselves.
• Secondary: established outlets and researchers reporting on someone else.
• Community: forums and aggregators. Useful for noticing things early, weak as evidence.

The active article pipeline first checks source eligibility and obvious off-topic material. A compact AI screen identifies central, informative AI coverage. Surviving articles receive an objective profile describing what changed, affected decisions and work areas, named subjects, and evidence limits.

Code checks source readability, AI relevance, attribution, and grounding before an article can publish. Unsupported, inaccessible, or incomplete material is held or excluded. Importance and evidence guide selection; a score alone is not proof of truth. Older stored stories may retain scores from earlier pipeline versions.

Reports of the same event are grouped so the edition can choose one representative, with supporting sources attached where available. The newsletter is assembled from a shared daily pool with freshness, diversity, direct relevance, and exclusion checks.

Models screen source material and create story profiles and source-grounded editorial building blocks. Code owns publication checks, reader matching, ordering, and provenance. Active article-processing calls use source material and editorial instructions, not your account details or reader-entered preferences.

AI-assisted copy can still be wrong. Every story links to its original source, and attribution or an evidence label is not a guarantee of accuracy. Read the original before relying on a claim. More detail is in the terms.

The public newsletter offers a general edition. With an account, your decisions, active work, must-not-miss subjects, and exclusions help select your edition from the same daily pool. Explicit exclusions remove matching candidates, and edition history helps prevent repeats on later days. Thin fresh supply may lead to clearly labelled continuing context or a shorter edition.

Where Browse is available, its personalized view ranks and filters eligible coverage using preferences and relevance signals; other sections let you explore broader coverage. You can change your answers under Account → Preferences. What we store and why is in the privacy policy.

Signal is an independent project. Nothing in the feed is sponsored, paid for, or placed. Source identity, evidence, relevance, and what changed inform selection. Publisher and company diversity checks help keep an edition from being dominated by one source.

If a summary misrepresents a source, or a story is scored in a way that seems plainly wrong, tell us. Corrections are the fastest way this gets better.

For corrections and feedback, write to dhonabhoovan@gmail.com.