Class MailboxKnowledgeSearch
- Namespace
- MailFathom.Application.Retrieval
- Assembly
- MailFathom.Application.dll
Answers a retrieval from the mailbox search this deployment already serves.
public sealed class MailboxKnowledgeSearch : IEmailKnowledgeSearch- Inheritance
-
MailboxKnowledgeSearch
- Implements
- Inherited Members
Remarks
The whole of the retrieval side, and deliberately thin: hybrid ranking, the structured filters, the account
restriction, and the extracts themselves all belong to MailboxSearchReader, which is the same
retrieval a caller reaches through search_emails. Answering a question therefore cannot see mail that
searching for it would not, cannot narrow in a way searching could not, and improving the ranking improves both.
Every filter is handed on unvalidated and unnormalized, which is the point rather than an omission: the use case below refuses an unusable address, an over-long subject fragment, and a date range that can select nothing, in the same words and under the same error code whoever wrote them. A second validation here would be a second set of rules to keep in agreement with the first.
What this adds is the shape a model's context needs rather than a reader's: one bounded extract per message instead of a summary and several highlighted fragments, carrying only the identity an answer is traced through. Everything a listing publishes and an answer does not need — the participants, the size, the flags, the attachment summary — is dropped here rather than sent to a provider.
Constructors
- MailboxKnowledgeSearch(MailboxSearchReader, EmailKnowledgeBounds)
Initializes the retrieval over the mailbox search and the bounds it hands its results over under.
Methods
- FindPassagesAsync(MailboxScope, EmailKnowledgeQuery, CancellationToken)
Finds the passages relevant to one query within one scope.