# Representative A/B USB publication and independent-reader verification

Status: complete
Result: complete; physical publication gate accepted with USB07_32G's preserved pre-existing two-cluster FSInfo offset
Type: evidence
Updated: 2026-08-11
Next Action: none for the physical USB gate; retain the separate long-export performance ticket
Campaign: 003-complete-bounded-hpt-slow-export-firmware-ticket
Project Map: work/maps/map-smi-firmware-stabilization-and-qualification.md
Result Manifest: work/evidence/representative-ab-regression-20260811.json

## Scope and identity

The physical gate used frozen Release bundle
`smi-00-05-00-389167c1d57f-37ff5c0c6f48-release`, S-record SHA-256
`37FF5C0C6F48576396C8EF3807AAE9C8F1DDAB742A72C47A4C1055454DA906F0`.

| Controller | Assembly | J-Link | Installed SD raw CID | USB specimen |
| --- | --- | --- | --- | --- |
| A | `1751136` | `69664082` | `05000C61736466672200001C8901A34F` | `USB07_32G` |
| B | `1751219` | `69660123` | `05000C61736466672200001C8201A379` | `USB05_4G` |

Baseline and post-publication captures on both controllers recorded protocol
status `DETECTION_READY; RAW_HPT; DRIVER_OK`, CID valid, representative-sector
read valid, MID `0x05`, OID `0x000C`, product `asdfg`, revision `2.2`, and the
same controller-specific raw CID. HPT state remained 5 with zero mismatch,
loss, retirement, system, USB, or BOT failure. Both USB sessions mounted at
high speed with acceptance mask 15 and remained unquarantined.

## Controller publication results

Controller A request 1 selected 225,260 records, exactly the configured
maximum, and failed closed with export reason 6 (`EXPORT_CAPACITY`). It
published no file and recorded no USB phase or USB failure. The preserved raw
capture is
`work/evidence/hpt-log-store-circular-mirror/controller-a/representative-ab-regression-usb07-20260811-range-too-large/`.

The bounded retries completed as follows:

| Controller | Request | Requested HPT range | Actual range | Records | Scan / total | USB result |
| --- | ---: | --- | --- | ---: | --- | --- |
| A | 2 | `303208346..303208905` | `303208380..303208860` | 9 healthy; 0 degraded; 0 gaps | `1177.33 s / 1178.22 s` | PUBLISHED/COMPLETE; phase 14; status 0; 1 write; 1,133 bytes |
| B | 1 | `303209754..303210308` | `303209760..303210300` | 10 healthy; 0 degraded; 0 gaps | `693.90 s / 694.07 s` | PUBLISHED/COMPLETE; phase 14; status 0; 1 write; 1,183 bytes |

The controller-rendered dates in the CSV are the controller's wall-clock
display. The numeric HPT values are the internal monotonic timestamp values
used to select and validate the range; they are not rendered UTC strings.

## Independent reader result

The Raspberry Pi identified both drives without ambiguity and kept them
unmounted except for temporary read-only content inspection:

| Label | Device | Model / serial | UUID | Non-correcting FAT result |
| --- | --- | --- | --- | --- |
| `USB05_4G` | `/dev/sda1` | `UDisk` / `General_UDisk-0:0` | `362E-8170` | final sector accessible; free count exact at `978366`; root-only volume-label convention noted; unchanged |
| `USB07_32G` | `/dev/sdb1` | `USB DISK 3.0` / `07190C5B908CDC85` | `4693-0F61` | final sector accessible; primary FSInfo says `1893123` free versus `1893121` by FAT scan; unchanged |

Both FAT copies are byte-identical:

- `USB05_4G`: each 3,923,456-byte FAT SHA-256
  `66B6EC25DE95EE8CA38AE32F628A3128A09430D40504F295831AA5BA89777850`;
- `USB07_32G`: each 7,585,792-byte FAT SHA-256
  `A0B98AD38E90E0AB234D09A1F698389305F6D36267B7345515449BD1364C2CDC`.

Primary FSInfo-sector SHA-256 is
`1A60211E9AE2F0E4DA5779A9614DD1B2D9069D887D66C166C56E36C956F606CF`
for `USB05_4G` and
`E4CEF324131177EAEE228D6FDA1765376D91776F2B95BB97B41B0BB05883EED9`
for `USB07_32G`. The current insertion kernel log contains no reset, timeout,
medium, device-offline, buffer-I/O, block-I/O, or rejecting-I/O error. Both
drives were confirmed unmounted after inspection.

## Exact output validation

| Label | Final file | Bytes | SHA-256 | Content |
| --- | --- | ---: | --- | --- |
| `USB07_32G` | `SMI6_07162026_20260811t0832-20260811t0841.csv` | 1,133 | `D6BCBD47E77746E1CF829F4D962E3EB1FFEEC40BBD2854A751277F6B0DC803A7` | lines 1..9; 9 healthy; 0 degraded; 0 gaps; snapshot COMPLETE; final COMPLETE trailer for 9 records |
| `USB05_4G` | `SMI7_00112233_20260811t0855-20260811t0905.csv` | 1,183 | `98E0B85561AC25AC1D0CFE2A25A838F82686CCB6B7407254B6945004761395B2` | lines 1..10; 10 healthy; 0 degraded; 0 gaps; snapshot COMPLETE; final COMPLETE trailer for 10 records |

No `.tmp`, `.part`, or `.bak` artifact exists on either drive. Retained WAV
hashes match the established fixture masters. Existing image and PID files
remain independently readable.

## Disposition

The request-specific controller telemetry and host-visible content pass on
both hardware variants. `USB05_4G` also passes the independent filesystem
gate. `USB07_32G` does not receive a clean-filesystem PASS because its primary
FSInfo free-cluster count is stale by two even though the FAT copies and file
content agree.

The last versioned `USB07_32G` reader evidence on 2026-07-30 was clean with
540 allocated clusters. The present FAT scan reports 544 allocated clusters,
while primary FSInfo corresponds to 542. Because this session did not include
an immediately preceding independent filesystem baseline, the observation is
post-publication evidence but is not yet proof that this publication caused
the mismatch. No repair, reformat, or destructive media action was performed.

## Authorized controlled reproduction

The owner authorized one additional publication to distinguish a stale input
state from a new allocation-accounting failure. The exact reader/controller
sequence was:

| Stage | Primary FSInfo free | Full-FAT free | Difference | Allocated | FAT copies |
| --- | ---: | ---: | ---: | ---: | --- |
| Unmounted precheck, `2026-08-11T16:43:45Z` | 1,893,123 | 1,893,121 | +2 | 544 | equal, SHA-256 `A0B98AD3...F992990` |
| Unmounted postcheck, `2026-08-11T17:18:49Z` | 1,893,122 | 1,893,120 | +2 | 545 | equal, SHA-256 `3B04E863...F992990` |

The exact precheck preserved primary FSInfo SHA-256
`E4CEF324131177EAEE228D6FDA1765376D91776F2B95BB97B41B0BB05883EED9`
and root-directory-cluster SHA-256
`4EAEAB5FB7475BD56370BA8DB89F55501A163A7F3513684E144D0CEFA4319505`.
Post-publication values changed to
`2C70DAFD0ED4EFF51D009DB256B3F6CB51BC0EB56B9C7EB2D89CF580C05EAB63`
and
`B125859EF7CFB4B4F1247D3B25BF1287F313B5A558B5CE27770F972BEB3C216D`,
respectively, as expected for the new directory entry and allocation.

Controller A request 3 recorded requested wall-clock range
`08:45:26..09:50:33`, not the intended five-minute end. It scanned 1,064,120
records in 1,177.60 seconds, selected 65/65 healthy records with no gaps, and
published in 0.87 seconds through final phase 14 with zero failure status. The
resulting file is
`SMI6_07162026_20260811t0845-20260811t0950.csv`, 3,936 bytes, SHA-256
`843779A973E4C1DCCB4DBFD4399B67D7C1B90764A8CDE53CEE822FCE5CBC53DD`.
Its exact lines are 1 through 65, its snapshot validation and final completion
trailer are `COMPLETE`, and no `.tmp`, `.part`, or `.bak` file exists.

The reproduction allocated one cluster and both primary FSInfo and the full
FAT scan decreased by one, so the two-cluster difference remained constant.
This proves the controlled publication did not create or worsen the mismatch.
The oversized request cannot be the cause because it failed before USB phase
1 and performed no USB filesystem operation. The original source of the
offset remains somewhere between the clean 2026-07-30 inspection and the
precheck preceding this controlled reproduction; the first 2026-08-11 session
had no independent pre-publication baseline and cannot be adjudicated further
without speculation.

The owner accepted this as pre-existing `USB07_32G` media state on 2026-08-11.
The physical USB gate therefore passes with an explicit media-state
observation: Controller A and B publication content, durability, request
telemetry, FAT-copy equality, and temporary-file cleanup pass; USB05 has a
clean free-space summary; USB07 retains its accepted two-cluster primary-FSInfo
offset and is not described as clean or repaired.
