Concentration
The echo-chamber watcher, Effective-N measures how many axes your attention actually spreads across.
A lens that only ever points one direction is still a blind spot. skope watches your own news diet and warns you when it collapses onto too few dimensions, even when every article is individually reachable. This is the concentration watcher.
Effective-N
skope measures attention spread with Effective-N, the Inverse Simpson index over how your radar's impact distributes across axes:
Effective-N = 1 / Σ(pᵢ²)
where pᵢ is the share of total impact landing on axis i over the recent window. It answers: "effectively how many axes is my attention spread across?", discounting axes that barely register.
- Effective-N ≈ your axis count → balanced. Your attention tracks your whole situation.
- Effective-N ≈ 1 → concentrated. Almost everything reaching you is one story, your portfolio, or one employer, or one technology.
Five axes each at 20% gives Effective-N = 5. But 80% on one axis and 5% on four others gives Effective-N ≈ 1.5, nominally five axes, effectively one and a half.
Why concentration is the risk
Reachability already keeps the irrelevant out. The remaining failure mode is the opposite one: everything that reaches you is true, relevant, and pointed the same way. That's an echo chamber built from real signal, the hardest kind to notice from the inside, because nothing in it is wrong. Effective-N makes the shape of your attention legible so the collapse is visible before it becomes a habit.
Reading it in the brief
get_brief returns a concentration block alongside the radar. A low Effective-N is Claude's cue to flag the imbalance ("four of your last five briefs were your portfolio; here's what you're not seeing") and, in a later version, to actively recommend a diverging query set.
Deterministic, like the rest of the ledger
Effective-N is pure math over the stored impact distribution, no model, no judgement call. skope computes the number; Claude interprets what to do about it. Same division of labor as everywhere else in skope: the ledger is deterministic, the synthesis is Claude's.
See also: Reachability, the lens that decides what enters the radar in the first place.