Rebuild is a hard workload for a drive-managed disk, and the reason sits in the design's own description. The firmware accepts random writes anywhere by staging them in a small conventional, unshingled cache region and reorganising shingled bands in the background [5]. A rebuild is a sustained write across the whole device rather than a burst. Once the staging region is full, a further write can only be accepted after the affected overlapping run has been rewritten in order [4], which is the operation shingling exists to avoid [3]. The source asserts the outage outcome up front [11] but the section that would document the failure path is cut off in the material supplied [14], so treat the mechanism above as inference from its own text rather than a measurement.
The controller has no window into any of this. Drive-managed parts are described as fully backwards compatible with any OS, controller or RAID card, and completely opaque about when and how hard the background reorganisation is working [5]. That compatibility comes at a cost for a RAID card: it gets service time but no field that explains it [2].
That opacity is what breaks the purchase-time check. On Linux, `lsblk -o NAME,ZONED` reports host-managed or none [9]. A drive-managed disk presents an ordinary block device with no zone interface [5], so it reports none, the same value as CMR [1]. `smartctl -a /dev/sdX` will usually surface the zoned device model [9], but a good number of drive-managed models never advertise SMR at all, and for any spinning drive shipped after roughly 2015 with no zoned marker the author's rule is to stop assuming [10]. The retail box, meanwhile, rarely says [9].
Host-managed removes the ambiguity, and it is decent engineering. Writes must land exactly at the zone's current write pointer, violations are rejected at the command level, there is no translation layer and no hidden cache, and the interface is ZBC on SAS, ZAC on SATA or ZNS on NVMe [7]. That is the right trade when you have software willing to own the write pointer, which is the case the source is making for a user-space daemon [13]. The hedge that would let you run an unmodified stack and still cooperate, host-aware, was positioned as transitional and is uncommon in current product lines [6].
So the capacity discount [12] transfers on two conditions. Either your write pattern never sustains long enough to drain the conventional cache, or the ordering obligation lives in your stack rather than in firmware you cannot observe. Neither condition is visible on the spec line that says SMR [8].
