How a spike is called

Windows

Series are ordered observations. The last 7 points are the recent window. The 90 points immediately before that are the baseline. The recent window is never part of the baseline. For Google Trends those points are days. A keyword needs at least 30 baseline points before it is scored. Cava in the sample set is deliberately short so this state is visible.

z = (recent_mean - baseline_mean) / stdev(baseline)
pct = (recent_mean - baseline_mean) / baseline_mean
combined_z = weighted mean of the z-scores that exist
score = clamp(50 + 12.5 * combined_z, 0, 100)

Weights are 65% Google Trends, 20% Reddit mentions, 15% Reddit upvotes. If a source is missing or too short, it drops out and the remaining weights are rescaled. With Reddit unconfigured, Google Trends is the whole score. A flat baseline makes the z-score undefined; a jump of 100% or more can still flag on percent change.

A keyword flags when the combined z-score is at least 2, when any one source reaches 2.5, or when any source’s recent mean is at least double its baseline. Score 50 is normal. Score 75 is a 2-sigma blend. Score 100 means the shock is off the scale of this formula.

What gets mapped

A keyword maps to the listed company that sells the product. HOKA and UGG both sit on Deckers (DECK). Ozempic and Wegovy both sit on Novo Nordisk (NVO). e.l.f. and rhode both sit on e.l.f. Beauty (ELF); the rhode acquisition closed on August 5, 2025. Labubu maps to Pop Mart (9992.HK). Buldak maps to Samyang Foods (003230.KS), not the Samyang Roundsquare holding company. Arc'teryx maps to Amer Sports (AS). Chili's maps to Brinker (EAT). Temu maps to PDD Holdings' Nasdaq listing. Hims is the telehealth platform, not the drug maker.

Stanley drinkware is not on the list. The brand is privately held. It is not Quanta Services (PWR), and this tool will not invent that link.

What the sources actually are

Google Trends, via pytrends, returns a 0–100 index for one keyword over the requested window. A different keyword or a different window rescales it. The collector stores each keyword from its own request, caches the response, waits between calls, and backs off when Google says to slow down. The latest day is often partial.

Reddit uses the official OAuth API. Mention counts and upvote totals come from the posts and comments the search listing actually returns, deduped, then filled with zeros only inside that coverage. It is not a census of the site. Without client id, secret, and user agent, the source reports “not configured” and is skipped.

Closes come from Yahoo Finance through yfinance and are drawn on the chart only. TikTok is registered on the same source interface and disabled. This app does not scrape TikTok.

Sample data

If the database is empty, startup loads a synthetic history so the board, the charts, and the spike rule can be read without credentials. Those rows are marked sample in the database and on screen. They are not measurements. A live collection writes a separate set and does not erase the sample.