The best ML rolesnever hit thejob board.

LinkedIn's algorithm optimises for engagement, not relevance. It surfaces roles that paid to be seen, not roles that match where you're actually heading. You're not browsing a job market — you're browsing a sponsored feed dressed as one.

The roles worth having — the staff ML engineer at a Series B that's about to 3× headcount, the research position at the lab that doesn't post publicly — those move through human channels. A Slack DM. A referral from someone who read your paper. A mention in a Discord server you might not even be in yet.

"Signal exists because the signal-to-noise ratio on job boards is approximately 1:400."

A $280k TC at a company that's quietly pivoting away from ML is not the same offer as $230k at one where the founding team just shipped a model that made the arXiv front page. The number is the least interesting part of the opportunity.

Signal annotates what job boards can't: team composition, runway, who the hiring manager actually is, whether the role is net-new or a backfill, and what the ML stack looks like three levels deep. The context that tells you whether to reply in 24 hours or archive and move on.

"The annotation layer is the product. Everything else is just a list."

No model produces the annotation that comes from a conversation with someone who interviewed at that company last quarter, or from a researcher who left because the roadmap shifted. That intelligence is human. It degrades fast. It doesn't get indexed.

Signal is edited by a person who has been in these rooms. Every role is assessed, not scraped. Every annotation is sourced, not generated. The dispatch reads like a dossier because it is one.