Wikipedia Views

Your agent translates this into daily page-view counts, each ticker's own 30-day baseline, and the z-score threshold that separates a genuine spike from ordinary noise.

A daily count of who bothered to look the company up.

Wikipedia publishes how many times each page was viewed on each day. When something happens at a company, some share of the people who hear about it go and read that company's page, which makes the count a rough, public measure of attention. This dataset maps those pages to tickers and carries the daily count in views.

Raw counts are not comparable across companies: a mega cap gets more views on its quietest day than a small cap gets on its loudest. avg_30d and zscore_30d fix that by measuring each day against the same ticker's own recent history. A z-score of 4 means the same thing on both names, which is what makes a cross-sectional screen possible at all.

Turn a spike in curiosity into a dated, filterable feature.

Attention shows up in places price does not. A stock can trade flat while the number of people reading about it goes up several times over, and the reverse happens too: a large move nobody looked up is a different kind of event from a large move the whole internet watched. The z-score is what makes that comparison possible across a universe instead of one name at a time.

The usual use is as a conditioner rather than a standalone signal. Setting zscore_30d_gte in the request turns the endpoint into an anomaly feed you can pin to a single date, which is a cheap way to ask whether the crowd was present for a setup you already had a reason to look at.

• Query zscore_30d_gte=2 with a fixed date to list every name that carried abnormal attention that session.
• Use zscore_30d_lte=-2 to find names the crowd has stopped watching, which is a different regime and not simply the opposite trade.
• Feed views and avg_30d into a cross-sectional attention factor rather than committing to one threshold.
• Line spike dates up against earnings, filings, and enforcement events to see which categories the public actually follows.
• Walk history with cursor_date and cursor_ticker to build a full per-ticker attention series once, then update it daily.

One row per ticker per day, scored against its own baseline.

Each ticker is mapped to a single Wikipedia page, carried in name, and that page's daily view count becomes views. The trailing statistics are computed per ticker over the previous 30 days: avg_30d is the mean of daily views, and zscore_30d places the day's count inside that distribution, so it is measured in standard deviations of the name's own recent traffic.

New observations land daily. Results are ordered by date descending then ticker ascending, and the endpoint is cursor-paginated: send back both cursor_date and cursor_ticker from next_cursor to walk history, since sending only one returns a 400.

A short schema, and the z-score that carries it.

The headline fields below are a subset. Every field, with exact types and semantics, is documented in the API reference.

Filter on the z-score, not on raw views.

One key works across the REST API, the hosted MCP server, and this dashboard. Every response is JSON with a { count, data } envelope.

import requests

url = "https://api.alphanume.com/v1/wikipedia-views"
params = {
    "date": "2026-02-06",
    "zscore_30d_gte": 2,
    "api_key": "alp_abc123"
}

r = requests.get(url, params=params)
print(r.json())

curl "https://api.alphanume.com/v1/wikipedia-views?date=2026-02-06&zscore_30d_gte=2&api_key=alp_abc123"

{
  "count": 1,
  "has_more": false,
  "next_cursor": null,
  "data": [
    {
      "ticker": "AAPL",
      "name": "Apple Inc.",
      "date": "2026-02-06",
      "views": 48213,
      "avg_30d": 21847.3,
      "zscore_30d": 4.11
    }
  ]
}

What this data does not claim.

• The baseline is only 30 days long. A name that has been in the news for a month can look calm at the peak of the story, because the spike has already been absorbed into its own average.
• Views are counted at the page level. A company page is shared by everything that company is famous for, so a film, a lawsuit, or a founder can move the count without the stock being the subject.
• Each ticker maps to one page, carried in name. Companies discussed across several pages, or under a page named for a parent or a product, are only partly captured.
• The count measures readers, not investors. Nothing in it distinguishes a shareholder from a student, a journalist, or someone who clicked the wrong link.
• Pro includes full history and the latest session.