News Feed

Origin

Most news consumption is reactive. You read what an algorithm decides will hold your attention, not what your current bets actually depend on. After enough cycles you realize that the feed is not actually informing decisions — it is shaping the decisions to fit the feed. A serious operator inverts this: signals must be filtered by personal priors and ranked against active commitments.

Problem

What's upstream of a current decision is rarely what's upstream of clicks. A markets thesis depends on regulator behaviour, counterparty solvency, jurisdictional drift; an ecological thesis depends on satellite data, partner reports, weather. The general-purpose news feed is calibrated for none of these. A purpose-built pipeline must be.

Approach

Source-agnostic ingestion (RSS, mailing-list, manual paste, scraped HTML) flows into a BRAINS-aware ranking layer. Each item is graded against the active wisdom set, current bets, and open predictions; high-relevance items surface as a daily digest, the rest fall away. The ranking changes when the operator's commitments change — the feed adapts to the work, not the other way around.

Methodology

BRAINS is the substrate. The ranking model has no separate state — it queries the same Postgres that holds active directives, predictions, and corrections. When a prediction resolves, items that informed that prediction get re-scored retrospectively, which feeds calibration drift back into the ranking. The pipeline is designed to be wrong slowly, not right loudly.

Open questions

• How to handle out-of-band intel — material non-public information that doesn't belong in a feed at all
• What the right cadence is between digest and real-time alert

Ask me about

• How items are scored against active predictions
• What the calibration loop teaches about source quality over time
• What does not belong in this pipeline

Stack