Request for Quote or Purchase (RFQ|RFP) with suspicious sender or recipient pattern

type.inbound
and (
  (
    (
      length(recipients.to) == 0
      or all(recipients.to,
             .display_name in (
               "Undisclosed recipients",
               "undisclosed-recipients"
             )
      )
    )
    and length(recipients.cc) == 0
  )
  or (
    sender.email.domain.root_domain in $free_email_providers
    and any(headers.reply_to, .email.email != sender.email.email)
    and any(headers.reply_to, .email.email not in $recipient_emails)
  )
  or (
    length(headers.reply_to) > 0
    and all(headers.reply_to,
            .email.domain.root_domain != sender.email.domain.root_domain
            and not .email.domain.root_domain in $org_domains
            // wetransfer includes user specific reply-to's & link display text which triggers NLU logic further within the rule
            and not sender.email.domain.root_domain == "wetransfer.com"
    )
  )
  or (
    length(recipients.to) == 1
    and all(recipients.to, .email.email == sender.email.email)
    and (length(recipients.cc) > 0 or length(recipients.bcc) > 0)
  )
)
and (
  // Group the keyword patterns that specifically indicate RFQ/RFP
  (
    1 of (
      // RFQ/RFP specific language patterns
      regex.icontains(body.current_thread.text,
                      '(discuss.{0,15}purchas(e|ing))'
      ),
      regex.icontains(body.current_thread.text,
                      '(sign(ed?)|view).{0,10}(purchase order)|Request for (a Quot(e|ation)|Proposal)'
      ),
      regex.icontains(body.current_thread.text,
                      '(please|kindly).{0,30}quot(e|ation)'
      ),
      regex.icontains(subject.subject,
                      '(request for (purchase|quot(e|ation))|\bRFQ\b|\bRFP\b|bid invit(e|ation))'
      ),
      any(attachments,
          regex.icontains(.file_name, "(purchase.?order|Quot(e|ation))")
      ),
      any(ml.nlu_classifier(body.current_thread.text).tags,
          .name == "purchase_order" and .confidence == "high"
      )
    )
    // Required: at least one RFQ/RFP keyword pattern

    // Optional: at least one additional indicator (can be another keyword pattern or a non-keyword indicator)
    and (
      2 of (
        // RFQ/RFP keyword patterns (same as above)
        regex.icontains(body.current_thread.text,
                        '(discuss.{0,15}purchas(e|ing))'
        ),
        regex.icontains(body.current_thread.text,
                        '(sign(ed?)|view).{0,10}(purchase order)|Request for a Quot(e|ation)'
        ),
        regex.icontains(body.current_thread.text,
                        '(please|kindly).{0,30}quot(e|ation)'
        ),
        regex.icontains(subject.subject,
                        '(request for (purchase|quot(e|ation))|\bRFQ\b|\bRFP\b|bid invit(e|ation))'
        ),
        any(attachments,
            regex.icontains(.file_name, "(purchase.?order|Quot(e|ation))")
        ),
        any(ml.nlu_classifier(body.current_thread.text).tags,
            .name == "purchase_order" and .confidence == "high"
        ),

        // Non-keyword indicators
        (
          any(ml.nlu_classifier(body.current_thread.text).entities,
              .name == "request"
          )
          and any(ml.nlu_classifier(body.current_thread.text).entities,
                  .name == "urgency"
          )
          and not any(beta.ml_topic(body.current_thread.text).topics,
                    .name == "Advertising and Promotions"
                    and .confidence == "high"
          )
        ),
        (
          0 < length(filter(body.links,
                            (
                              .href_url.domain.domain in $free_subdomain_hosts
                              or .href_url.domain.domain in $free_file_hosts
                              or network.whois(.href_url.domain).days_old < 30
                            )
                            and (
                              regex.match(.display_text, '[A-Z ]+')
                              or any(ml.nlu_classifier(.display_text).entities,
                                     .name in ("request", "urgency")
                              )
                              or any(ml.nlu_classifier(.display_text).intents,
                                     .name in ("cred_theft")
                              )
                            )
                     )
          ) < 3
        )
      )
    )
  )
  or (
    length(attachments) == 1
    and length(body.current_thread.text) < 100
    and all(attachments,
            .file_type in $file_types_images
            and any(file.explode(.),
                    2 of (
                      regex.icontains(.scan.ocr.raw,
                                      '(discuss.{0,15}purchas(e|ing))'
                      ),
                      regex.icontains(.scan.ocr.raw,
                                      '(sign(ed?)|view).{0,10}(purchase order)|Request for a Quot(e|ation)'
                      ),
                      regex.icontains(.scan.ocr.raw,
                                      '(please|kindly).{0,30}quote'
                      ),
                      (
                        any(ml.nlu_classifier(.scan.ocr.raw).entities,
                            .name == "request"
                        )
                        and any(ml.nlu_classifier(.scan.ocr.raw).entities,
                                .name == "urgency"
                        )
                      ),
                      any(ml.nlu_classifier(.scan.ocr.raw).tags,
                          .name == "purchase_order" and .confidence == "high"
                      )
                    )
            )
    )
  )
)

// negate highly trusted sender domains unless they fail DMARC authentication
and (
  (
    sender.email.domain.root_domain in $high_trust_sender_root_domains
    and not headers.auth_summary.dmarc.pass
  )
  or sender.email.domain.root_domain not in $high_trust_sender_root_domains
)
and (
  not profile.by_sender().solicited
  or profile.by_sender().days_since.last_contact > 30
)
and not profile.by_sender().any_messages_benign