Claude
Add Cadence as a connector with your key. Claude discovers the ten tools and calls them as it needs them.
Cadence Skill
Give your AI a read on Europe.
The intelligence behind Cadence, inside the assistant you already use. Connect it to Claude or ChatGPT and ask what is changing across eleven European markets, compare findings and investigate the evidence behind them.
Every supported numerical claim comes back with its source and its reporting period. Where the evidence cannot carry an answer, Cadence says so and tells you what it can establish instead.
One connection. Eleven markets. Evidence you can inspect.
The conversation
What it looks like in the assistant.
A question in your own words, the assistant calling Cadence, the evidence it gets back, and what it does and does not establish. Every figure below is this quarter's published evidence.
You Is online pulling away from the rest of retail anywhere in Europe?
Consulting Cadence Skill… europe_picture · market_read
Assistant
Cadence evidence
retail volume, total −0.6%
retail volume, online +7.3%
In 3 of 7 eligible markets, total retail volume and online volume moved in opposite directions this period: Germany, Spain and Italy.
Spain is the clearest of the three. Total retail volume −0.6% year on year while online +7.3%, both legs the same Eurostat series and the same month, so the gap is a measurement rather than two sources disagreeing.
The four markets outside that seven are not zeros: Austria, France and Poland publish their two legs for different months, and Ireland's online leg is not published this period.
Eurostat sts_trtu_m · VOL_SLS · 2026-08 · both legs Eurostat sts_trtu_m VOL_SLS, same month per country, opposite signs
What this does not establish — 6
- Any composite, index or score across these measurements.
- Any causal claim between them.
- Any ranking of the countries in the eligible set.
- Any brand attribution: the series contains no brands.
- Any inference about a country in `excluded_countries`: it is not a zero, not a negative observation and not neutral.
- Any prevalence stated against 11 when fewer countries were eligible.
Markets outside the eligible set, and why
- Austria — legs are different periods (2026-08 vs 2026-07)
- France — legs are different periods (2026-08 vs 2026-07)
- Ireland — online absent, stale or withheld
- Poland — legs are different periods (2026-08 vs 2026-07)
You Which brands are actually running advertising in France right now?
Consulting Cadence Skill… market_read · movers
Assistant
Cadence evidence
LeclercCarrefourIntermarchéAuchanLidlMonoprix BNP Paribas · none running
Presence, not volume. The creative count the collector returns is capped, so it is not published as a measurement.
Six of the French retail brands Cadence observes are running advertising in the last 45 days: Leclerc, Carrefour, Intermarché, Auchan, Lidl and Monoprix.
BNP Paribas is observed in the same basket with none running, which is an observation rather than an absence of data.
Cadence does not join this to attention. Per-brand search movement is withheld this quarter, because repeated measurement did not reproduce it at publication precision, so there is no league table and no advertising-against-attention comparison to give you.
Google Ads Transparency · FR · 45-day window · advertiser presence by domain
What this does not establish — 4
- Any advertising spend: presence is observed, cost is not.
- Any ranking of the brands by how much they advertise.
- Any join between advertising and attention: per-brand movement is withheld this quarter.
- Any inference about a brand outside this market’s observed basket.
You How much of the music chart is domestic, and where does that differ most?
Consulting Cadence Skill… league · market_read
Assistant
Cadence evidence
Italy 76.6%
France 58.9%
Netherlands 0.0%
Position-weighted over a trailing year, and only the part of the chart whose artists resolve to a citizenship. The resolved share is on the receipt.
Italy 76.6%, France 58.9%, Netherlands 0.0%. Those are the measured domestic-artist shares of the eligible chart.
They describe the chart measurement and nothing beyond it. They are not a reading of national taste, and they establish nothing about whether locally voiced creative would travel in one market and not another.
The measure is not published everywhere: 3 of the eleven markets withhold it this quarter because too little of the chart weight resolved to a citizenship.
Domestic share: Wikidata citizenship (P27) only, position-weighted · trailing 365 days to 2026-09-30 · 91.3% of chart weight resolved
What this does not establish — 4
- Any claim about national musical taste: this is a chart measurement, not a preference survey.
- Any claim about how creative work would perform in a market.
- Any ranking against the markets that withhold the measure this quarter.
- Any comparison with a market whose resolved chart weight is materially different without saying so.
How it connects
One endpoint, one key, no second platform.
Cadence runs as an MCP service. Your assistant discovers the ten tools, calls the ones a question needs, and composes the answer from what they return. There is no separate intelligence dashboard to learn, and nothing to install locally.
ChatGPT
Connected the same way, through the same MCP endpoint. We will set it up with you rather than hand you a configuration file.
Anything else that speaks MCP
One endpoint, one key. Cadence also answers over plain HTTP for a client that does not speak MCP at all.
A key and an entitled account are required. What there is not is another place to go: the intelligence arrives in the assistant your team is already in.
Intelligence with boundaries
An answer, or a reason there isn't one.
Every figure carries its receipt
The source, the series and the reporting period come back with the number, so a claim can be checked rather than trusted. Not every answer has a chart; every numerical answer has its evidence.
Eligibility travels with the claim
A cross-market statement carries the set it was measured across. A market outside that set is named as outside it, never counted as a zero.
It declines, and says why
Where a measurement did not reproduce, or a rule's evidence is absent, the Skill returns the withholding and its reason instead of an estimate. Market similarity is withheld this quarter, and asking for it returns that rather than a nearest-market list.
The current edition, and why there is no history yet
The Skill answers on the quarter now published, and that is the only quarter published: Q3 2026 is the first edition. A movement needs two readings of the same thing, so quarter-on-quarter comparison begins with the next edition. It is not being withheld from this one.
Access
A subscription of its own.
The premium Skill
Every market and every published edition, with an allowance built for real sessions, and the premium reports your account has bought opened in the chat alongside them. Monthly or annual, from the Q4 2026 edition.
Cadence Enterprise
Every market and every edition in a private instance, with the Skill scoped to whatever that instance covers.
A premium report on its own
Does not include the Skill. A report grants a document, permanently; the Skill is hosted compute and comes with a subscription. A subscriber who has bought a report gets to ask questions of it; a report on its own is a document.
Nothing is billed per query: the Skill is scoped to what you hold rather than charged by the question. A fair-use limit applies per hour, to stop one caller exhausting a shared service. Cadence receives the tool call your assistant makes — the market, the sector, the parameters of the question — and not your conversation with it. The service keeps no record of what was asked. What happens on the assistant's own side is governed by whoever provides it.