# Isolated 32 KiB USB write-path matrix closeout

Status: complete
Result: complete; historical write-path matrix evidence, with enumeration disposition completed by its successor
Type: evidence
Updated: 2026-08-01
Next Action: none; bounded enumeration recovery and the final broad baseline are complete
Campaign: 003-complete-bounded-hpt-slow-export-firmware-ticket

Project Map: work/maps/map-smi-firmware-stabilization-and-qualification.md
Workpackage: work/wp/completed/wp-restore-broad-usb-compatibility-baseline.md
Checklist: work/checklists/checklist-stage-1b-four-drive-usb-compatibility-baseline.md

## Controlled candidate

The deployed candidate is source commit
`8fa1479b6b942f72f18ff4ed2cb1a5c73ddc2501`, Release S-record SHA-256
`82A277EEBBED0CC4FBDDCC848ACEFD43E706849803673C4920E7851FC2ACABB5`.
Relative to the restored compatibility baseline, its only firmware behavior
change is the HPT export FileX maximum write from 1 KiB to 32 KiB. It adds no
USBX/FileX transport change, VBUS action, lifecycle retry/replay, optional
SCSI command, or drive-specific behavior.

## Completed evidence

USB09 and all eight four-drive/controller observations pass the full 32 KiB
HPT publication path and independent no-repair Pi validation:

| Observation | Controller A | Controller B | End-to-end result |
| --- | --- | --- | --- |
| Round 1 | `USB01_2G`; first-insertion mount | `USB05_4G`; first-insertion mount | PASS |
| Round 2 | `USB05_4G`; one clean reseat after below-class miss | `USB01_2G`; first-insertion mount | PASS |
| Round 3 | `USB02_8G`; first-insertion mount | `USB03_32G`; one clean reseat after below-class miss | PASS |
| Round 4 | `USB03_32G`; one clean reseat after below-class miss | `USB02_8G`; first-insertion mount | PASS |
| Incident regression | `USB09`; exact prior failure range | — | PASS |

Every HPT export completed and published with zero USB status, zero first USB
failure status, zero snapshot/requested-range gaps, 14 FileX writes, and a
32,768-byte maximum write. Independent Pi gates found equal FAT copies, exact
continuous CSV sequences and COMPLETE trailers, exact PID/WAV outputs,
internally CRC-valid recovery images, no temporary artifacts, clean insertion
kernel state, and final unmounted media.

This evidence answers the write-size question: broad drive compatibility and
the 32 KiB HPT speed improvement do not require choosing one over the other.
The original/common-denominator USB lifecycle can be retained while using the
32 KiB HPT write cap.

## Remaining compatibility blocker

The matrix is not a zero-intervention first-insertion pass. Three 32 KiB
observations required one clean powered reseat after the controller detected
the drive electrically but did not reach USBX class/media handoff. An earlier
recovery cross-check supplies a fourth occurrence with `USB03_32G` on A.

All four misses occurred with `SYSSTS0=32769`, `DVSTCTR0=512`, no usable
FileX media, and before any HPT or image write. Successful reseats reached
`SYSSTS0=32864`, `DVSTCTR0=531`, a fresh media generation, application
session state/status `2/0`, and a valid FileX pointer. The pattern spans both
controllers and disproportionately follows `USB03_32G`; it is therefore not
caused by the 32 KiB FileX write cap and is not an A-only failure.

The next investigation must stay below FileX: USB attach/electrical timing,
HCD state progression, USBX enumeration/class publication, and the insertion
event path. Do not weaken standards checks, add drive identity branches,
restore mandatory optional-SCSI commands, or use controller-driven VBUS
cycling as a production workaround. Production promotion remains blocked
until the first-insertion reliability issue has an evidence-backed
disposition.
