How the predictive list works

Big news usually keeps going for a few days. If something huge is happening in the world, tomorrow’s Wire is more likely to talk about it again than about something random.

We use that idea to make a short list of names that look more likely to show up in the next few Wires. One list. Names only. It is a fan tool for searching this site — not a crystal ball, and not from S2 Underground.

The simple idea

Think of a snowball rolling downhill. Once it is moving, it tends to keep moving for a while. That is momentum.

Now think of friends who hang out in a group. When you see one of them, you often see the others soon after. That is clustering.

1. Momentum — names that have been showing up a lot lately get a boost, because big stories usually continue for at least the next few days.
2. Clustering — when a big story is hot, it often brings a small pack of related names with it. Some of those names are the usual ones. Some are the surprise ones that are not yet loud on every Wire.

The surprise names in that pack are the “alpha” we care about: quiet or early names sitting next to a loud story, before everyone is talking about them.

What the math is doing

We do not guess the future from thin air. We look at the public Wire text we already have.

• Momentum — we count how often each tracked name appears, and whether that count is rising or falling. Rising or still-hot names get more momentum.
  M_i = f_i · Δf_i
  M_i momentum · f_i recent frequency · Δf_i change in frequency
• Clustering — we look at which names show up in the same reports or the same stretch of days. Names that travel together form a cluster around a bigger story.
  C_ij = cooccur(i, j)
  C_ij co-occurrence of names i and j · cluster = names linked by high C_ij
• Surprise — we score names that are tied to a hot cluster but are not already the loudest names in the room. Those are the surprise picks.
  S_i = H_cluster(i) / (f_i + ε)
  S_i surprise · H_cluster(i) cluster heat · f_i how loud already · ε small constant

High S_i: hot neighborhood, quiet name.

What you see on the home page

One ranked list of taxonomy names, ordered by next-Wire likelihood — how relatively likely each name looks to appear across the next few Wires, given momentum and clustering in the published corpus.

The list refreshes when new Wires land. The home block’s as of date is when that snapshot was built.

Data

Built from the same public Wire material on this site. Free downloads live on the Data page at /data/ if you want to explore the corpus yourself.

What this is not

Not a promise that any name will appear. Not a forecast of real-world events. Not advice. Not an official S2 product. Reading the Wire still comes first.

Back to search · Data / offline packs · About