How scoring works
The feed runs two tracks. News signals — articles, filings, op-eds, reports — go through an automated scorer that rates each one on five (sometimes six) dimensions, each on a 0–10 scale. Total runs 0–50. Anything below the research threshold gets killed; anything above it shows up here. Academic signals(📜) take the fast lane: if the paper is by a scholar I track in the Takt Network, it's auto-included, no scoring. Being on my tracked list is the editorial signal.
You can filter the feed using the pills at the top of the page — all, news only, or academic only. Below is how the news track is scored.
The dimensions
- Information Gap · 0–10
- Does this reveal something the audience does NOT already know? Specificity widens gaps — a precise number, filing, or data point scores higher than a general claim.
- Mechanism Exposure · 0–10
- Does this show HOW something works, not just THAT it's happening? The causal links between institutional decisions are where the value lives.
- Institutional Gap · 0–10
- Does this reveal a gap between what an institution says publicly and what the data shows they're doing?
- Local Specificity · 0–10
- Is this about a specific place, ZIP, county, state, insurer, or market? Specificity is what makes climate data actionable.
- Timeliness · 0–10
- Is this new? Published in the last 48 hours scores high. Last week medium. Older but newly relevant can score medium with justification.
- Response Value · 0–10 · media signals only
- Media signals only — would Lucas's analysis on this article help readers? A piece that's half-right and half-wrong is a natural jumping-off point.
Thresholds
A signal needs at least 15/50 to be considered research-worthy and appear in the feed. Signals below that are killed silently.
A signal scoring at least 25/50 with high tweetability also gets a draft tweet generated — though I post manually, never automatically.
The learning loop
Every Sunday, a synthesis pass reads my approvals and kills from the week and proposes updates to the scoring weights. The model gets sharper with use. This is how I think today — it evolves.
Why this matters, twice
Each signal has a “Why this matters” blurb. The ones in italics (with a ⚙ icon) are auto-drafted by Claude using the scorer's output. The ones without italics are my edits — those are the signals I starred and worked through personally during the 8:30 AM editorial pass.
Starred signals (⭐) also carry a short note from me explaining what I think the headline gets wrong, what it understates, or why the dry technical detail is actually the load-bearing thing.