Why placement data needs a canonical domain layer

A practical model for comparing marketplace offers without flattening away the details that change the real price.

Every placement marketplace describes inventory differently. One source prices an article with writing included, another exposes the publisher payout, and a direct seller quotes a final amount in a spreadsheet. Treating those rows as equivalent produces a misleading comparison.

Start with the domain, keep every offer

The canonical entity is the website. Offers remain separate children carrying the seller, source, currency, platform fee, content policy, link attributes, turnaround and guarantee.

This lets the interface answer two different questions:

Is this website suitable for the campaign?
Which seller has the best acceptable offer for it?

Normalize the final cost

The visible price alone is not sufficient. Store writing, platform fees, taxes and optional extras separately, then compute a comparable total for the buyer’s currency.

Freshness is part of quality

Every metric and price needs an observation time. A slightly weaker metric updated yesterday is often more useful than an impressive value imported six months ago.

This is also why raw source records should be retained: when a normalization rule changes, historical input can be replayed without fetching the entire source again.