The diagnostic is the only part of the advisory that yields a number, and it yields a coarse one. Ten confirmed incidents means each case carries 10 percentage points of the reported gap [15]. One miscoded ticket moves the answer from four in ten to five. That is enough to establish whether the problem exists in a given program. It is not enough to trend quarter over quarter, and whoever presents it should say the sample size out loud.
The pull has a prerequisite the advisory states without dwelling on it: sending domain age and URL reputation as of the delivery timestamp, not as of now [10]. Reputation is a moving value. A domain that reads as malicious today may have been unrated when the message landed, which is the whole point. If the incident record never captured the rating at delivery, the audit cannot be reconstructed afterwards, and that missing field is itself a finding about the program.
Failure pattern one is worth walking step by step, because no step is exotic. A domain registered three days earlier hosts a credential harvesting page built from a legitimate phishing kit, sends from a dedicated IP with no prior sending history, and reaches a finance team member carrying a contextually accurate invoice reference. The IP is clean, the URL is unrated, and the content matches no known campaign pattern in any threat feed, so every element clears reputation filtering [8]. The tradecraft here is freshness rather than the kit itself, and freshness is the one attribute a database built from previously observed activity cannot price [7].
The consequence the advisory draws is a measurement one. When the recipient interacts, the post-mortem finds that the message passed all filtering checks, and the metrics log the message as compliant, because the program only ever measured what filtering can evaluate [9]. The compromise reads as an anomalous outcome rather than a coverage gap [9].
Now the limits of what is public. The advisory declares three failure patterns in programs with functioning filtering [5], and the text supplied runs out mid-sentence inside pattern two's second sub-case [14]. So two failure patterns are documented here, and the third is not. Within pattern two, the advisory says three delayed weaponization techniques are in active operational use and names redirect activation, where a time window or geographic trigger swaps the destination after delivery, and legitimate infrastructure hosting [13]. The third technique is unnamed in this copy [14]. The documented two stand as documented, and the rest remains pending rather than implied.
