Sky Lux Plaza in Barcelona (Diagonal, business district) holds a 9.4 on a major platform — top 1% of the city by the aggregator’s own ranking. On RawStay the same hotel scores 38 out of 100, verdict “Be careful”. The difference isn’t in the data: the hotel hasn’t faked anything, the 9.4 is real. The difference is how we count it.
The hotel has 210 reviews averaging 9.4. Bayesian shrinkage toward the industry prior (7.0) barely moves the number at that volume — the resulting “platform score” comes out to 85 out of 100. If the index stopped there, Sky Lux Plaza would look brilliant, same as everywhere else online.
But then the rule kicks in: a score alone can’t buy the index. The hotel has 0 live guest photos — zero confirmation that the room looks like what’s promised. 📷 Without confirmation the index is cut by at least half: 85 × 0.45 ≈ 38. That’s not a penalty for a bad hotel — it’s an honest “we can’t confirm this yet”.
For a guest this doesn’t mean “don’t book”. The 9.4 isn’t lying — but it doesn’t prove there’s nothing to hide either. It might simply be an honest hotel that never posted room photos. The practical move: read 5–6 recent reviews and check whether guests describe the room as shown in the promo shots, or differently.
We keep this breakdown live permanently as a test for the formula itself: if a future weight change stops this exact 38 from following this exact logic, we broke something — and the formula needs fixing, not the article.