Muse MuseConcept

Independent concept built on Meta's Model API. Not an official Meta product.

Made-up shop inbox (D8 test)

Decision dataset · uploaded

Ready

4 labels · 274 train · 38 dev · 78 test

A person's labeled examples for a decision model. Their inputs stay private: this page shows only the labels and how many rows each has.

Question

Which team should answer this made-up customer message?

Every prompt asks this about one input, with the labels as lettered options and "None of the listed options matches" last. The model answers with one letter.

Labels

Rows per split
Each label with its description and its rows in train, dev and test
LabelTrainDevTest
BillingCharges, invoices, refunds of payments, promo codes701020
ShippingWhere an order is, delivery problems, addresses701020
ReturnsSending items back, swaps, return labels, refund for a return71920
AccountSigning in, passwords, email, deleting or suspended accounts63918

Test rows (a fifth of each label, at least 5) were held out first and are never trained on: both models are scored on them once. Dev rows (a tenth, at least 3) only set how confident the answers say they are. The rest train, and near-repeats always land in the same split.