Every cell's Data sheet tracks the same 14 feed types (the list is derived from tbl_DataTypes on the Control sheet - never hard-coded). Learn them as a checklist; a cell cannot reach Data sign-off until every one of them is resolved:

Each feed carries one status from tbl_DataStatus. The happy path is Pending → Received → Validated, and the gap between the last two is where the real work sits: Received means a file landed; Validated means it survived the checks - right granularity, full history, consistent hierarchy (sub-brands re-sum to brand), no unexplained gaps or duplicates. A feed can sit at Received for weeks while issues are chased.

Three off-ramps exist, and they mean different things:

• Granularity Issue - data exists but at the wrong grain for this cell's read. The recurring example: brand-health (bgs) data is captured at brand level only and gets duplicated across sub-brands and variants. A one-year-old variant cannot really be "salient", but it inherits the brand's salience because that is the only brand-health repository there is. The client is informed and it goes in the caveats - but if the cell's read depends on sub-brand granularity that does not exist, the feed is flagged here.
• Blocked - data should exist but cannot be obtained: an access request stuck with a data owner, an agency that has not responded, a licensing gap.
• Not in Scope - a deliberate scoping decision that this cell does not need the feed. This is a recorded call, not a shrug.

Nielsen vs Kantar

Nielsen carries the overwhelming majority of sales coverage - roughly 98% Nielsen to ~2% Kantar in a typical market. Kantar's genuine contribution is the hard and soft discounter retailer sets that sit outside the Nielsen universe; they are summed together, never modelled separately. The judgement call: if the Kantar share is negligible, do not merge it - tell the client it adds difficulty rather than value and run on Nielsen alone. Merging tiny panel volumes into a census-scale read buys noise, not accuracy.

The TDP decomposition check

The distribution feed arrives as three series, and they must reconcile: Total TDP = Promo TDP + Non-Promo TDP. If the identity fails, the promo decomposition downstream (promo TDP index, promo price index) is built on sand - the feed stays at Received until the source explains the gap.

The e-commerce overlap trap

In markets where retailers run their own online and click-and-collect operations (the UK and US, for instance), the offline Nielsen read already contains e-commerce sales. You cannot add ecomm_sales to sales to get total brand sales - that double counts. The prescription is structural: a separate offline read and a separate e-com read, each with its own media tagged against it, never additive. This is why ecomm_sales and ecomm_media are their own feeds rather than columns in the main ones.

Data readiness is not an ambient state - it is a gate. When every one of the 14 feeds reaches Validated (or is formally parked at Not in Scope), the cell earns its Data sign-off. Alongside it sits the Design sign-off - agreement on scope, model granularity and competitor set. Only when both sign-offs are in place does the committed 8-week Build clock start. That is the whole point of the gate: the build window is a promise, and the program refuses to start the clock on a cell whose data can still surprise it. Chasing feeds happens in Preparation, on a variable timeline, precisely so the Build phase does not absorb the risk.