We ship a generation every 9 months and our support department has not kept up. The figures below are the real ones, including three targets we currently miss. The incumbent is better at this than we are and we are not going to argue otherwise on our own support page.
01Where we actually are
A service index of 64 against a median of 100 is bad, and it is bad for a reason we can name: we hired engineers to ship G5 and did not hire enough people to answer for G3 and G4 at the same time. Eleven service staff started between May and August 2029. The figures on this page will be re-measured in December and republished whether they have moved or not.
02Targets and actuals
| Route | Target | Actual | Notes |
|---|---|---|---|
| Self-serve index | Always on | Always on | 1,410 articles, every one dated. |
| Written ticket | 12 working h | 14 working h | We miss our own target. Median, August 2029. |
| Bench callback | 2 working days | 2.4 days | Engineer on the line, not a script. |
| Dead-on-arrival swap | 72 h | 58 h | The one number we beat. |
| Out-of-warranty repair | 15 working days | 19 days | Parts for G1 and G2 are gone. We will say so. |
Written tickets are raised through your yard’s supply desk with a deck code and a serial. We do not operate a personal account system and there is nothing to register. A bench callback puts an engineer on the line, not a script — that is the route worth using when a fold ratio has gone strange and you need someone who can read a run file.
03Known issues
Every fault we have reproduced and not yet closed, plus the ones we closed recently enough that you may still be running into them. A fault we have decided not to fix says so, and says why.
Fold table can mis-evict a Ward-class resident under a cold second jack inside 90 seconds. Throughput drops to roughly 340 MB/s until re-seat. No cache loss observed in 2,100 reproductions.
Fan curve steps audibly at 71 °C rather than ramping. It is a firmware table and we have not fixed it yet because the safe version is louder, not quieter.
The bench board throttles hard above 60% duty cycle with no airflow. This is the board working correctly. It has no heatsink because you were told it has no heatsink.
Service bus enumeration could hang for 40 seconds on units built before serial 4T-06100. Firmware fix shipped January 2029; no return required.
04Firmware
Stable is what a deck ships on and what we recommend. Roughly one release every six weeks. Bench is the same build a week earlier, for people who would rather find our mistakes than wait for them. Both are signed; both install over the service bus; either can be rolled back to the previous two releases from the deck itself.
The changelog is published the day a release ships, lists the faults closed by reference number, and lists anything that regressed. A release that fixes one thing and slows another says so in the first line.
There is no automatic update. A deck does not change behaviour because we pushed something overnight; you install firmware when you decide to.
| Gen | Shipped | Firmware to | Parts |
|---|---|---|---|
| G1 | March 2026 | None — out of support | None |
| G2 | December 2026 | None — out of support | None |
| G3 | September 2027 | September 2031 | Boards only |
| G4 | July 2028 | July 2033 | Full |
| G5 | April 2029 | April 2034 | Full |
G1 and G2 are out of support and out of parts. If you send one in we will send it back unopened rather than take credits for a repair we cannot complete. Forty G1 units were built; all forty were replaced free in 2027 and we still have the list.
05Repairs
The honest summary: our hardware is faster than the incumbent’s at a typical resident set and our service department is worse than theirs. If downtime costs you more than throughput earns you, buy the Mk V. That is a real recommendation and it is on our own support page because you would find out anyway, later, and in a worse mood.