Method sheet — The Protocol

1. Sources & licences

co2	NOAA Global Monitoring Laboratory (Mauna Loa)	Public Domain (U.S. Government)
seaice_north	NSIDC Sea Ice Index v4 (NOAA@NSIDC)	NOAA/NSIDC: frei nutzbar mit Quellenangabe
seaice_south	NSIDC Sea Ice Index v4 (NOAA@NSIDC)	NOAA/NSIDC: frei nutzbar mit Quellenangabe
sst	NOAA OISST v2.1 (via Climate Reanalyzer, University of Maine)	NOAA: Public Domain; Aufbereitung: Climate Reanalyzer
fires	NASA FIRMS (VIIRS S-NPP, Near-Real-Time)	NASA: frei nutzbar mit Quellenangabe
quakes	USGS Earthquake Hazards Program	Public Domain (U.S. Government)
population	UN DESA, World Population Prospects 2024 (Extrapolation)	CC BY 3.0 IGO
refugees	UNHCR Global Trends	UNHCR: frei nutzbar mit Quellenangabe
food	FAO Food Price Index (2014–2016 = 100)	FAO: frei nutzbar mit Quellenangabe (CC BY-NC-SA 3.0 IGO)
rates	Europäische Zentralbank, €STR	EZB: frei nutzbar mit Quellenangabe
oil	U.S. Energy Information Administration (Brent Spot)	Public Domain (U.S. Government)
conflict	GDELT v2 Events (BigQuery Public Dataset); erfasst Medienberichte	GDELT: frei nutzbar mit Quellenangabe
attention	Wikimedia Pageviews API (en.wikipedia.org)	CC0 (Messdaten)

2. Cadence

Daily sources are retrieved nightly (03:30 UTC). Sources with monthly or annual cadence never claim daily freshness: every finding states its measurement date ("As of: …"). UNHCR and FAO values are maintained editorially as versioned data files.

3. Processing

Every sentence is derived deterministically from data — templates, no LLM, register version 1.0.0, stated on every page. The canonical artefact is the daily JSON file in the repository (one commit per session, author "Protokollführung"); the prose is presentation. Code is public: pipelines/protokoll in the site repository. dbt lineage follows with the BigQuery experiments.

→ pipelines/protokoll

4. Limits of the method

Attention uses the English-language Wikipedia as a proxy for global attention — a linguistic and cultural narrowing.
Attendance is a linear extrapolation of the UN World Population Prospects, not a measurement.
Ocean reports absolute sea surface temperature (NOAA OISST v2.1, served via Climate Reanalyzer) with a year-on-year comparison, not an anomaly against a baseline.
Fire counts satellite detections (VIIRS), not individual fires.
Conflict counts media reports of conflict events captured by GDELT, not the events themselves; the date is the date reports were captured, not the date of the events.
Comparison values for interest and oil refer to the previous observation day — across weekends and holidays this lies further back.
Displacement and Food are maintained manually; their measurement date is always stated.
Unreachable sources appear as "finding omitted", implausible values "under reserve" — failure is never silently bridged.

5. Compute footprint

One Cloud Run execution per day (< 60 s, 1 vCPU). One partitioned BigQuery query per day with a hard cost ceiling of 5 GB (≈ 0.5–1 GB processed, ≈ 25 GB/month — within the 1 TB/month free tier). The site itself is static; no cloud service is read at runtime.

6. Change log

→ To the work