Hot Trend Lab Source Protocol

Every Hot Trend Lab chart starts with a source card. The card tells the reader what the series measures, when it was collected, how much coverage it has, and which claim it can support. This keeps the column readable without hiding the measurement limits.

Source card

Use this checklist before publishing a chart or table:

Field	What to record	Pass condition
Source	API, endpoint, vendor, or cached public snapshot	the source can be named and re-fetched or inspected
Query	category, keyword, ticker basket, article title, or repo list	another reader can reproduce the selection
Timestamp	access date plus source time field	the collection date is visible in the table
Coverage	first timestamp, last timestamp, observation count, missing ratio	gaps are small enough for the intended chart
Refresh rule	per run, daily, weekly, or event-driven	cadence matches the claim being made
Interpretation	what the source measures and what it does not measure	the article sentence stays inside the evidence

Source registry

The registry is the live list of public feeds used by the notebooks. Each row maps a source to its metric, notebook use, refresh rule, and interpretation limit.

Reliability matrix

Source	Measures	Best use	Main limit	Publish threshold
arXiv API	public preprint counts and metadata	research-output trend, deadline pulses, topic watchlists	preprints are not peer-reviewed; cross-listing and query wording change counts	query and date window shown; spike explained against deadline context
Hugging Face Hub API	model metadata, downloads, likes	current open-model attention and repeated snapshot retention	one snapshot is cross-sectional; downloads do not measure quality	snapshot date shown; repeated snapshots required for momentum language
GitHub REST API	repo metadata and stargazer timestamps	developer attention and launch-week velocity	token limits, page limits, bots, and repo selection bias	covered page/window shown; stars described as attention
Wikimedia Analytics API	article pageviews	public attention and hype-decay curves	article naming and media events can dominate the series	project, article, date range, and access type shown
DeFiLlama stablecoin API	stablecoin supply and chain distribution	liquidity context around crypto events	endpoint schema can change; supply is not execution liquidity	schema inspected and top-chain context shown
CoinGecko API	crypto price and market data	price residual context for BTC/ETH examples	public API rate limits and vendor methodology	market structure language; no trade recommendation
Yahoo Finance through yfinance	public market prices	AI infrastructure price-proxy examples	unofficial public wrapper; not licensed point-in-time market data	basket definition shown; valuation claims excluded

API, cache, and timestamp rules

Public APIs change shape, paginate results, and rate-limit unauthenticated requests. Each notebook records the endpoint or query context, then writes a source audit table with first timestamp, last timestamp, observation count, and missing ratio. Cached files are treated as dated public snapshots rather than anonymous local data.

For snapshot sources, wording changes with depth:

Snapshot depth	Allowed wording	Avoid
One dated snapshot	current public ranking or adoption proxy	momentum, acceleration, retention
Repeated snapshots	change in downloads, likes, stars, or attention over time	quality, production usage, valuation
Full source history or scheduled collection	trend, seasonality, residual events over the covered window	claims outside the source universe

Source snapshot rules

Cached source snapshots keep the same source-card fields as freshly fetched data: source, query, access date, time range, coverage, and interpretation scope. A cache can support a chart only when its date and query are visible.

Replacement sources

A notebook may use a better licensed or institutional source, but the result still needs the same source card. For production work, add vendor name, point-in time policy, revision policy, calendar, and redistribution limits.