§ Per-source detail
For each source — the ten live plus the two trade build-out (markets, auctions): its taxonomy (how the data divides), the attributes we actually hold (real columns, not aspirations), the key attributes worth pulling out, the insights they drive in a cultural report, the value of those insights, and — the part that matters most — why each becomes unique only in combination with the others. Quick jump: Spotify · Apple · Amazon · Netflix · Trends · Wikipedia · GDELT · Eurostat · Finance · Markets · Auctions · Weather.
From france_poc_v1.combined: chart_type, date,
position, track_id, artist_id, isrc, title,
release_date, and the six audio features — valence, energy,
danceability, tempo, acousticness, mode — plus origin
metadata p_line, c_line, label. Derived in analysis:
mode_major, catalogue_age, local_share, position-weight.
In music_charts (source = apple_music_top_songs): country,
chart_date, rank, track_id, track_title,
artist_name, artist_id, genres (JSON), release_date,
url. No audio features — Apple gives chart position and metadata, not valence/energy.
In music_charts (source = amazon_retail_dmusic_songs): country,
chart_date, rank, track_id (ASIN), track_title,
url. artist_name currently empty — resolved via a follow-up /dp/ lookup
in v2. The distinction that matters: this is purchase, not stream.
In netflix_top10: country, week, category
(= "Films" or "TV"), rank, show_title, season_title,
cumulative_weeks. The films/TV split is native; local-vs-international origin is not —
it needs TMDB enrichment (a flagged roadmap item). cumulative_weeks is an under-used staying-power metric.
In trends: term, country, week_start,
index_value (0–100). Deliberately brand/category-anchored — these are the names the report's
category playbooks are built on. The taxonomy is the term list; it is fully parameterisable per market.
In wikipedia_pageviews: project (= fr.wikipedia), article,
date, views (daily integer count). Absolute counts, not a normalised index —
unlike Trends, this is a true volume. Daily granularity makes it the most responsive attention signal we hold.
In gdelt_tone: query_id, country, date,
tone (position-weighted sentiment, roughly −10 to +10). One national tone series plus one per
tracked brand — so brand discourse and national discourse can be read separately. Daily, 2019–2025 complete.
In macro_series (source = eurostat): series_id (8 series incl. 3 CCI
variants), country, period, frequency (monthly), value,
unit (balance_pct, index_2021_100, pct, pct_yoy). The sub-indices are the prize — savings
intent and major-purchases are where the music proxy validated.
In macro_series (source = yfinance): CAC40_CLOSE (eur_points) and
EURUSD_CLOSE (ratio), daily, 2019–2025. Sits in the same table as Eurostat, so macro context is one
query. Primary role today is a control variable in the backtests, not a headline report figure.
The academic anchor (Edmans et al.) is literally about music sentiment predicting stock returns — stocks are the original downstream variable in the bridge thesis, not just a nuisance to regress out. Two cuts carry real cultural signal: sector rotation (consumer-discretionary vs staples = a clean risk-on / risk-off mood read) and brand-level equity for the listed brands we already track on search, tone and mood.
The wealth pyramid has three rungs and the stack reads two: Spotify is mass mood (18–35), Amazon is the older/wealthier purchaser — auctions are the very top. For luxury especially, "are collectors paying records?" is a direct read on high-net-worth confidence — exactly what a maison wants. The real data lives in the sibling ART / Salle.art project (Christie's/Sotheby's via Barnebys); it is not in the Cadence pipeline today — it was illustrative in the v5 mockup and is honestly a concept-stage integration, not a current source.
In weather_daily: country, city (Paris), date,
variable (6 met variables), value, unit. A pure control layer —
weather confounds mood (sunshine lifts everything), so it is regressed out, never reported. The discipline of
holding it is itself a quality signal.