Hot Trend Lab: Real Data Trend and Cycle Decomposition

Hot topics usually have three time structures: a durable trend, a repeating cadence, and a residual shock that breaks the normal pattern. DeTime is useful here because it makes those structures explicit before the column turns them into a story.

Consider a model-release week. arXiv may show a category or query spike before the launch, Hugging Face may show a same-day snapshot jump, GitHub stars may accelerate after demos ship, and Wikimedia or market proxies may react later. Those signals are interesting together, but they answer different questions. The column keeps source coverage, freshness, query context, timestamp, and interpretation boundary next to each chart so a reader can tell what changed and what still needs outside evidence.

Public source arXiv, Hugging Face, GitHub, Wikimedia, DeFiLlama, market data
DeTime trend, season, residual, meta
Evidence table source card, coverage, trend score, residual event rank
Column judgment what moved, why it matters, what not to infer

What the reader gets

Each case study starts with a source protocol, then produces a table or figure that supports a specific editorial judgment:

Component What to read Example question
trend durable direction over the selected window Is a topic still building after the launch week?
season repeated cadence Is the pattern mostly deadline or weekday rhythm?
residual event-like deviation from the smooth baseline Which week deserves a closer read?
meta source, query, timestamp, coverage, limits Can this chart support the sentence being written?

The result is not a popularity scoreboard. A single source rarely measures quality, adoption, or economic value on its own.

Column map

Case-study notebooks

The pages below render notebook code, tables, figures, and captured outputs directly from examples/notebooks/hot_trends/. Read each notebook for three things: why the source is interesting, how to read the tables or axes, and what decision the output can support.

Rendered page Primary judgment Main boundary
01 arXiv category pulse which fields are structurally growing or spiking paper count is not paper quality
02 arXiv agent research pulse which agent topics deserve a weekly read query wording changes the sample
03 Hugging Face open-model pulse current public model attention and later retention one snapshot cannot show momentum
04 GitHub AI-agent star velocity developer attention during the covered window stars are not production adoption
05 Wikipedia attention hype decay public-attention decay after a burst pageviews are attention, not importance
06 crypto stablecoin liquidity pulse price residuals next to stablecoin context not a trading signal
07 AI infrastructure market pulse market-price proxy movement around AI infrastructure valuation needs financial statements

The rendered notebook index and the 00 overview notebook are appendix material for readers who want the full transcript list.

Install optional dependencies

python -m pip install -e .[dev,docs,notebook]
python -m pip install -r examples/hot_trends/requirements.txt

Then open:

jupyter lab examples/notebooks/hot_trends

Research scope

Use this column to analyze public attention, research activity, open-source activity, liquidity, and market-proxy time series. Adoption, quality, valuation, and investment decisions need separate evidence.