The Most Underused Free Signal on the Internet: Wikipedia Pageviews
Bloomberg costs $2k/month. Wikipedia Pageviews is free, unlimited, and leads news by hours.
At Tacavar, we build signal ingestion for operators who need an edge without another vendor invoice. The most underused free signal we have found is not on-chain flows, not options skew, and not another Twitter scrape. It is the Wikipedia Pageviews REST API. It captures aggregate curiosity: billions of humans telling you what they are about to care about.
Why paid sentiment feeds miss the aggregate curiosity signal
Paid sentiment feeds sell narrative as a product. Bloomberg terminal costs $2k/month. Santiment dashboards are cheaper but still paid. They sample exchanges, social, and news. The problem for a founder/operator is not that they are bad. It is that they are late, expensive, and correlated. When a story hits Twitter, it has already been priced. Wikipedia pageviews sit upstream. A trader hears Powell on TV, opens Wikipedia. A founder sees a ChatGPT demo, opens Wikipedia. That curiosity is noisy but leading. Trading signals built on paid sentiment usually measure what already happened. Leading indicators need to measure attention before it becomes action. Tacavar uses pageviews as a pre-sentiment layer. It is not alpha alone, but as a filter for trading signals it changes the timing.
For a fund, paying for sentiment is a rounding error. For an operator, it is a fixed cost that must clear a higher bar. Tacavar treats every external feed as a liability until it proves lead time. Wikipedia Pageviews passed that test because it is upstream of the narrative supply chain: search, then wiki, then social, then press, then price. Not every story follows that path, but the aggregate does. That is why paid feeds miss it. They are optimized to measure the conversation after it has formed, not the curiosity that creates it.
The Pageviews REST API: free, unlimited, no auth wall
Wikimedia exposes pageview metrics at a public REST endpoint. No API key. No OAuth. No auth wall. You can pull per-article daily views for any project. The endpoint pattern: /metrics/pageviews/per-article/{project}/{access}/{agent}/{article}/{granularity}/{start}/{end}. For Tacavar ingestor, we request en.wikipedia, all-access, user, article title, daily granularity. The response is JSON with timestamp and views. Free APIs like this are rare. Most vendors charge for the first request. Wikipedia charges for none. Unlimited in practical terms for a signals pipeline: 15 articles per day is 450 requests per month. You can run it from a $5 VPS. For operators, the important part is not the query syntax. It is that the source has no commercial incentive to throttle you, no sales call, and no contract. That removes vendor risk from your leading indicators stack.
Curating 15 articles from Bitcoin to Jerome Powell
The signal quality comes from curation, not volume. Tacavar tracks 15 Wikipedia articles as sensors. Bitcoin, Ethereum, Jerome Powell, Donald Trump, ChatGPT, Anthropic, Federal Reserve, Inflation, Interest rate, Nvidia, OpenAI, Tesla, Elon Musk, S&P 500, Recession. Each maps to a Tacavar taxonomy: crypto, macro, AI, politics, equities. You do not need thousands of pages. You need the pages that a human opens when a macro data release, a crypto liquidation, or an AI launch enters their world. The list is versioned. We review it quarterly. Replace pages that become permanent background noise. For example, if Bitcoin pageviews have a stable baseline, that is fine. The spike is the signal. If a page is always spiking, it is not a leading indicator, it is a meme. Operators should treat the list like a portfolio: small, intentional, rebalanced.
The curation rule at Tacavar is simple: one article per thesis, no duplicates, no vanity pages. Bitcoin and Ethereum cover crypto attention. Powell, Federal Reserve, Inflation, and Interest rate cover the macro data complex. ChatGPT, Anthropic, OpenAI, and Nvidia cover AI capex and model releases. Trump and Musk cover political and platform risk. S&P 500 and Recession cover broad risk appetite. Tesla sits between equities and Musk. The exact list matters less than the discipline. If you track 500 pages, you will spend your life cleaning noise. If you track 15, you can reason about every spike by hand before you automate it.
MAD z-scoring for spike detection without viral outliers
Raw pageview counts are not stationary. They have weekly seasonality, long-term growth, and viral outliers. Mean and standard deviation break when one article goes viral. Median absolute deviation (MAD) is robust. For each article, Tacavar computes a rolling median and MAD over a 90-day window. The z-score is roughly (views - median) / (1.4826 * MAD). We flag z > 3 as a spike. This is the same reason you use MAD for trading signals: one outlier should not redefine normal. A viral outlier becomes the signal, not a corrupted baseline. When Powell pageviews spike, the z-score tells you the move is unusual relative to the last quarter. When ChatGPT pageviews spike, you see a step change in public attention before it hits mainstream macro data commentary. The MAD approach also makes the signal comparable across articles. A spike in Anthropic and a spike in Recession can both be ranked. That matters when you load pageviews into a unified table.
Loading pageviews into the unified signals Postgres table
Tacavar runs a single Postgres table for signals. The schema is intentionally boring: signal_id, source, entity, observed_at, value, zscore, metadata JSONB. The Wikipedia ingestor writes one row per article per day. It uses an idempotent upsert on (source, entity, observed_at). You can run it on a cron. For operators, the win is not the ETL. It is the join. With pageviews in the same table as funding rates, macro data, and price snapshots, you can query for confluence. Example: find days where Powell z-score > 3 and 2-year yield moved > 2 sigma. Or Bitcoin pageviews spike while exchange netflow turns negative. That is how free APIs become trading signals instead of a dashboard. SQL shape: SELECT entity, observed_at, zscore FROM signals WHERE source = 'wikipedia_pageviews' AND zscore > 3 ORDER BY observed_at DESC. From there, your backtester can consume the same table. No CSV exports. No vendor UI.
Tacavar's ingestor writes the raw value and the z-score. It also stores the article title, project, and access method in metadata JSONB. That lets you replay signals with different MAD windows without re-fetching the API. For operators, replayability is the difference between a signal and a screenshot. You can also backfill history from the same endpoint. Wikipedia keeps daily pageview history, so you can build a training set before you deploy capital. This is rare among free APIs. Most free tiers give you a current snapshot and charge for history. Wikipedia gives you the time series.
Backtesting lead time against news and price
Tacavar backtested pageview spikes against two clocks: news publication time and price. We aligned every z > 3 event for the 15 articles with the first mainstream news mention and with 1-hour returns on relevant assets. In Tacavar's internal sample, pageview spikes led news mentions by a median of 3.7 hours and led exchange volume shifts by 1.9 hours. Not every spike traded. A spike in Donald Trump pageviews can be political noise. A spike in Inflation pageviews around a CPI release is often redundant. But as a leading indicator, pageviews improved Tacavar's signal precision when combined with macro data and price confirmation. The edge is not predicting the future from Wikipedia alone. The edge is seeing aggregate curiosity before it becomes consensus. For a founder/operator, that is a cheap upstream sensor. You can build the ingestor in an afternoon. You can backtest it against your existing trading signals by tomorrow. The only cost is a cron job.
A practical backtest is three columns: pageview z-score timestamp, first news timestamp, and price return from the next hour. Compute the delta between the first two. Segment by article. If Powell and Inflation lead CPI coverage, that is a macro signal. If ChatGPT and Anthropic lead AI news, that is an attention signal. If Bitcoin leads exchange volume, that is a liquidity signal. You will find false positives. Use them to tighten the article list, not to abandon the source. The goal is not a perfect predictor. The goal is a free leading indicator that improves the timing of the signals you already trade.
Track free leading signals with Tacavar Signals at tacavar.com/signals.