# Current SMI Setup A connected-bench record

Status: Controller A runs the exact scrub-qualified bundle; F1-A1 is HPT initialized by evidence-backed false-negative adjudication
Type: inventory
Updated: 2026-08-05
Next Action: review the built terminal lane-mask correction, then separately qualify the bounded HPT2 stop/recovery gate before any HPT2 start
Campaign Reason: Historical record from campaign 009-complete-clean-contract-bound-a-b-qualification; retained independently after campaign closure and requires no separate owner-queue execution.
Campaign: standalone

Canonical Truth: E:/dev/high_point/7520553_SMI_FW/README.md

## Identity

### 2026-08-05 HPT-initialized current state

Owner-authorized operation `d909a40c-7cf4-46d7-b93f-c6b3873b32c5`
initialized F1-A1 without replay. Its original Pi terminal result remains
`FAILED` because the firmware terminal snapshot published a reset transient
lane mask. Independent read-only observation
`d1c51d83-043a-40ce-b78a-14652f670cab` proves exact installed flash,
`RAW_HPT/HPT_READY`, `LastLaneMask=7`, clean health, CID-valid and readable
media, and unchanged CID `05000C61736466672200001C8901A34F`. Pi Evidence `40`
adjudicates the physical state `HPT_INITIALIZED`; the original result is
preserved. The terminal publication correction is built but not installed.
HPT2 has not started. See
`work/evidence/controller-a-f1-a1-hpt-initialization-adjudicated-pass-20260805.md`.

### 2026-08-05 exact-candidate installed state

Controller A now runs bundle
`smi-00-05-00-b845d4b9bcfe-4f37e97877d8-release`, complete flash
`aed8fc3f289094c4e7d4c3313268687a31d822d117833527fd6c8e0a7b4dd242`.
Programming preserved VEEPROM and option memory, and independent observations
before and after the new baseline scrub passed. F1-A1 remains uninitialized at
raw CID `05000C61736466672200001C8901A34F`; CID-valid and
representative-sector-read status pass. Full-address operation
`abc4be6b-6cde-447b-8b98-1bf7021e8586` passed all `15,728,640` sectors with
zero failures or recoveries. Both protected registry roles are disabled and no
tool container remains. See
`work/evidence/controller-a-b845d4b9-program-verify-baseline-scrub-pass-20260805.md`.

### 2026-08-05 initialization adapter software-only state

Later the same day, the diagnostic-corrected bundle
`smi-00-05-00-b845d4b9bcfe-4f37e97877d8-release` superseded the earlier
initialization bundle for future programming. It is imported valid but
unadmitted; Controller A was not changed. The Pi now runs fixture-only release
`hpt-initialization-ui-d7547a4` / image `sha256:a999e305...388243` with the
protected trust-store reader and initialization envelope/status UI. See
`work/evidence/hpt-initialization-readiness-software-continuation-20260805.md`.

Controller A was not contacted or changed. It still runs scrub-qualified bundle
`smi-00-05-00-3861b4a2c040-e75e6df95e01-release`, and `F1-A1` remains
uninitialized with last accepted raw CID
`05000C61736466672200001C8901A34F`.

The new firmware bundle
`smi-00-05-00-c9ada4d5122b-e51ae77e9fab-release` is imported valid but
unadmitted on the Pi. Pi commit `1572ed39...`, release
`hpt-initialization-adapter-1572ed3`, and image
`sha256:491a0708cb81a9e1789db3309fe678b75d414d5e54dcd30a75e9faa81de865ad`
are deployed fixture-only with initialization roles/endpoints empty and live
admission false. Because the controller candidate would change, a fresh
post-program complete-flash/CID observation and full-address scrub are required
before initialization. Evidence:
`work/evidence/hpt-initialization-mailbox-adapter-software-deployment-20260805.md`.

### Canonical 2026-08-05 reconciled baseline state

Existing operation `6279f027-759c-46be-b3ce-ad70766c5707` reconciled read-only
to `COMPLETED/FULL_ADDRESS_SCRUB_PASS`; it completed all `15,728,640` sectors
with zero failed or retry-recovered sectors, zero identity mismatch, and no
command replay. Post-result observation
`c59cb0db-1c00-4226-bd6b-8ed60c5c2f67` again matched complete flash
`b1abe4776388b0978a2a1f93473e25a0091af8da535aa54582bb270b2dfcabb3`,
assembly `1751136`, variant `0`, J-Link `69664082`, and raw CID
`05000C61736466672200001C8901A34F`. Protocol detection remains
`NEEDS_INITIALIZATION`; CID-valid and representative-sector-read status pass.

Pi Evidence `26`/`27`, Evidence `28`, CID capture `6`, and event `51` form the
durable closeout. The card is still uninitialized and unwritten. The deployed
qualifier has no admitted initialization adapter or machine-intent execution
path, so manual initialization is prohibited. See
`work/evidence/controller-a-baseline-scrub-readonly-reconciliation-20260805.md`.

### Canonical 2026-08-04 cadence-promoted current state

Controller A remains assembly `1751136`, hardware variant `0`, J-Link
`69664082`. Programming operation `f38afdd2-45ee-4cb5-9915-593cb018cd88`
installed bundle `smi-00-05-00-3861b4a2c040-e75e6df95e01-release`; complete
flash is exactly
`b1abe4776388b0978a2a1f93473e25a0091af8da535aa54582bb270b2dfcabb3`,
with VEEPROM and option memory preserved. Independent observation
`e1d03af3-12e0-4398-a85d-1da6e79ea9e9` passed running/mapping/state checks
and recaptured `F1-A1` CID `05000C61736466672200001C8901A34F`.

Read-only scrub `3b04cc35-ce17-4201-88f9-7a885a9300ab` completed every one of
`15,728,640` sectors with zero failures or recovered retries. The Pi result is
immutable `FAILED` only because its adapter incorrectly requires the
event-driven HMI thread to wake; periodic controller scheduling progress was
positive. The card remains uninitialized. See
`work/evidence/controller-a-cadence-promotion-and-scrub-liveness-policy-block-20260804.md`.

### Canonical 2026-08-04 restored topology and current verified state

The owner reported that canonical Controller A was reconnected after the
display-test detour. Pi read-only observation
`55333d3c-d9ab-45a7-b78f-151ae9439481` independently reconfirmed assembly
`1751136`, hardware variant `0`, J-Link `69664082`, and a running CPU at
`3.283 V` VTref. Expected and observed complete code flash both equal
`68029e19efcc409f878b2f43371ee48d823ffff2d7a4f4c0b349176b6077bb28`,
the prior fixed-mailbox bundle
`smi-00-05-00-bc0b6ce4d066-release`.

The installed KEXIN `F1-A1` identity is valid and representative-sector
readable; raw CID is `05000C61736466672200001C8901A34F`, detection state is
`NEEDS_INITIALIZATION`, and driver status is `0`. This is pre-program
reconnection evidence only, not the required post-program CID admission
capture. See
`work/evidence/hpt-preinitialization-scrub-bundle-pi-import-campaign-binding-20260804.md`.

### Temporary 2026-08-04 topology history

The owner reported that a different controller was temporarily connected to
J-Link A. Read-only target access proved the MCU and running state, then the
owner explicitly authorized loading display-test firmware
`7520708_DispTest_FW_01-00`. Exact programming/readback and preservation
evidence is recorded in
`work/evidence/display-test-controller-on-jlink-a-load-20260804.md`.

This temporary controller's assembly identity was not independently
established. Do not infer that the historical Controller A assembly, hardware
variant, firmware, SD card, or CID below is currently attached. J-Link serial
`69664082` and port `19120` identify the probe channel, not the controller.

### Canonical historical mapping

- Bench label: Controller A / Setup A
- Controller assembly: `1751136`
- Running hardware variant: `0`
- J-Link serial: `69664082`
- Generic connector endpoint: `srv-rpi00` port `19120` through SSH forwarding
- Mapping method: read `gHardwareVersion` from the running target through the
  serial-bound J-Link endpoint; reconfirm after cabling changes

## Current installed firmware and read-only state

On 2026-08-02, authorized Pi operation
`741d370e-2083-4a8d-a325-332caabef70b` installed and complete-readback verified
bundle `smi-00-05-00-248c1e7b7aab-release`. The installed complete 4 MiB flash
SHA-256 is
`D1BABFA7A65610544741C0B329212B0EADFC92B32A8E95B513E2E2EC9AE70556`,
exactly matching source commit
`248c1e7b7aabe1d0da306e65b03baa5a3e05e67f`. Complete pre/post VEEPROM and
option-memory hashes were unchanged; CPU reset/run proof passed.

Independent read-only operation `19a603d0-d08d-4079-8357-0d88513ea157`
matched the same complete flash and stable mapped state. Hardware variant and
system errors are `0/0`; SD is `RAW_HPT` with valid CID and representative
sector read; HPT is `HPT_READY` with zero HPT/SSP error, lane mask `7`, no
retired/mismatched/lost records, and no replacement request. Card identity
remains SanDisk `SS08G`, raw CID `035344535330384780976F0084018559`, capacity
`15,523,840` sectors. See
`work/evidence/controller-a-candidate-program-verify-and-observation-20260802.md`.

Before that authorized installation, two independent read-only paths had
superseded a stale current-image claim. A complete 4 MiB flash readback exactly
matched the tracked Release S-record from source commit
`ca6f573a8d30c264ecd6de084786b12696c63d81`:

- installed complete-flash SHA-256:
  `23428A890890CA40172483A683F2B985C5BFC640650E7B33B977EAD15DB754CA`;
- installed ELF SHA-256:
  `68AC029578BE8EDB7CD3F5B39254C8A4B5351A5E9161E87DDE7A1D2DC6ECF0A9`;
- installed S-record file SHA-256:
  `6907F88FB91D5DC58EBF15A918C3993FDE3A076AFAE9DAEFB099899F55932581`.

Two exact-length RAM passes using that installed image's mapping were
byte-identical. Hardware variant and system errors were `0/0`; SD was
`RAW_HPT` with valid CID and representative-sector read; HPT was `HPT_READY`
with zero HPT/SSP error, lane mask `7`, no retired/mismatched/lost records, and
no replacement request. Card identity remains SanDisk `SS08G`, raw CID
`035344535330384780976F0084018559`, capacity `15,523,840` sectors.

The deployed Pi observer then compared that prior complete flash against the
reviewed candidate bundle and published
`BLOCKED/FIRMWARE_IDENTITY_MISMATCH`, operation
`ce3c5f3c-47b8-4b37-9beb-61925a58bbad`. CPU state was running before and
after, no mapped RAM read followed the mismatch, and no target-changing
operation occurred. See
`work/evidence/controller-a-readonly-bundle-identity-and-snapshot-20260802.md`.

## Historical final-candidate firmware and target state

### Prior 32 KiB buffered USB-publication image

- Debug S-record SHA-256:
  `35B020FDBDF2E02E0ACD96E83A261917A11F5EE48DB24058FB98450F6A956B4F`.
- Programming/verify/reset/run: PASS on J-Link `69664082`.
- CID unchanged at `035344535330384780976F0084018559`.
- Real retained-range export published 1,828 records with zero gaps,
  fallbacks, dropped logs, system errors, or HPT/USB status.
- Timing: `2.33 s` HPT staging, `0.45 s` USB publication through final
  flush/rename, `2.82 s` total; 3 bounded writes.
- Boot cache ready in `2.96 s` total (`2.11 s` traversal).
- Final Graph is active and following newest with current rendering and zero
  viewport lag; newest was `20 s` behind RTC at capture.
- Evidence:
  `work/evidence/hpt-usb-csv-buffer-window-tuning-20260726.md`.

### Prior graph real-event timer image

- Debug S-record SHA-256:
  `D3D3615586FD522A9472504EAB55BC8A6F4955956AAE3229CCD397A0AD1A1EAF`.
- Programming/verify/reset/run: PASS on J-Link `69664082`.
- CID unchanged at `035344535330384780976F0084018559`.
- Cache ready in `2.71 s` total (`1.87 s` traversal), `720` chart records.
- Real `GX_SIGNAL` task-button paths advance on a later persisted record at
  24, 12, 6, 3, and 1 hour zoom; Pan Left hold and Pan Right reattachment pass.
- Owner observed advancement while zooming in and accepted the correction on
  2026-07-26.
- Evidence:
  `work/evidence/graph-real-event-refresh-timer-correction-20260726.md`.

### Current graph-follow review image

- Debug S-record SHA-256:
  `143F1941FCF43BD31A5084CF21E0BC8595880F8E58C2CBA22B98D594759C3124`.
- Programming/verify/reset/run: PASS on J-Link `69664082`.
- CID unchanged at `035344535330384780976F0084018559`.
- Cache ready in `2.68 s` total (`1.81 s` traversal), `720` chart records.
- Active Graph followed new records with zero edge lag at 1, 3, 6, 12, and
  24 hour zoom; pan-away hold and exact Pan Right reattachment passed.
- Evidence: `work/evidence/graph-follow-newest-correction-20260726.md`.

### Prior warning-truth review image

- Debug S-record SHA-256:
  `944BEA28BFD6DE5604A5BE1B03F4F5690CA84213B2BD923940ED27D193844501`
- Programming/verify/reset/run: PASS on J-Link `69664082`.
- Fresh CID/read/HPT baseline: PASS; raw CID unchanged at
  `035344535330384780976F0084018559`.
- Exact setpoint exercise: `25.000067 C` to `25.100067 C` and back to the
  original bits; both saves recovered from first-attempt
  `SSP_ERR_CMD_LOCKED` on attempt 2, warning transaction remained zero, and
  RAM plus both VEEPROM blocks match.
- Graph cache: ready with real data; `1,498` scanned, `1,497` selected;
  `1.75 s` traversal and `2.60 s` total; `291` read commands, `9,090` blocks,
  zero bulk fallbacks.
- Evidence:
  `work/evidence/operator-warning-truth-and-target-proof-20260726.md`.

### Prior retained-start review image

- Debug S-record SHA-256:
  `91D05B3593D58DD525FFA2EDA74E165D25967652EBE3D0B885CC8368F180CCB5`
- Programming/verify/reset/run: PASS on J-Link `69664082`
- Fresh CID/read/HPT baseline: PASS; raw CID unchanged at
  `035344535330384780976F0084018559`
- Graph cache: ready, complete, retained data present, `720/720`, zero system
  errors, zero dropped logs
- Bulk-16 timing: `1058` scanned, `1057` selected; traversal
  `124 ticks / 1.24 s`; total `209 ticks / 2.09 s`; `210` read commands,
  `6,450` blocks, maximum `32` blocks/command, `68` refills, zero fallbacks
- Published retained customer bound: sequence `2`, timestamp `301739760`
  (`2026-07-25 08:36 UTC`)
- Current lane state after normal logging: `38/38/38` active records with
  first sequences `1024/1024/1024`
- Evidence:
  `work/evidence/hpt-export-start-adjustment-20260726.md`
- Performance plan:
  `work/wp/completed/wp-optimize-hpt-triple-mirror-read-performance.md`

### Previous frozen-candidate qualification

- Firmware implementation commit:
  `391c7b4d053ee146b823c505987897d3f88c0022`
- Artifact freeze commit:
  `c0901312e0fe850d8a0b2452eb20d88c86ce07bd`
- Debug S-record SHA-256:
  `D386F03460E225B683D5146FFC7AD949CCFEF71A3EC8169C944EFA82B5992F9F`
- Programming/verify/reset: PASS on J-Link `69664082`
- Card protocol status, CID validity, and representative-sector read: PASS
- Raw CID: `035344535330384780976F0084018559`
- Full-address read-only scrub: PASS; all `15,523,840` sectors accounted for,
  zero failures, retry recoveries, identity mismatches, or HPT errors
- Integrated HPT/chart/logging gate: PASS; zero system errors, raw HPT open,
  representative read good, and chart rebuild completed `720/720`
- Evidence:
  `work/evidence/post-freeze-integrated-qualification-20260725.md`

The owner reported PASS on 2026-07-25 for visual wording/placement/legibility,
touchscreen navigation, and alarm audio. Removal, reinsertion,
injected-failure, physical interruption, endurance, and multi-sample work
remains separately scoped and is not implied by this pass.

## Pre-final historical firmware and target state

On 2026-07-25 Controller A was programmed/verified/reset with the complete
pre-freeze development integration image, Debug S-record SHA-256
`74D00768F9A66BD7081E28FD4539C26A4AA7EE380E954E765D47728E14A82F73`.
This image comes from a dirty working tree at source HEAD `4896fe86`; the
S-record hash, not HEAD alone, identifies it. Same-run J-Link evidence found
zero system errors, `RAW_HPT`, valid/readable CID
`035344535330384780976F0084018559`, an open healthy triple store, and a
successful active 30-day chart rebuild. Two production records completed
during the observation and drops remained zero. See
`work/evidence/hpt-log-store-circular-mirror/pre-freeze-integration-20260725.md`.
This is retained as pre-final historical evidence and is not the current
next-operation record.

On 2026-07-24 Controller A was programmed through J-Link `69664082` with the
frozen Debug 00.05.00 artifact from commit `c7a20522`, SHA-256
`BA2CF3DB7A52C1487A28C1F274D7DBF0D60F17F3A2E014DE015ADE4508FAF15C`.
Program & Verify passed. VEEPROM and option memory were byte-identical across
the load; configuration/calibration, CID, and HPT state were recaptured.
System/configuration errors were zero and HPT reported all lanes with no
retirements, mismatches, losses, or replacement request. Evidence:
`work/evidence/integrated-00-05-00-target-qualification-20260724/controller-a-load-and-readonly-gate.md`.

The owner confirmed the complete service-467 screen and service 514's displayed
`[-10.0,+10.0] C` range on 2026-07-24; service 514 was exited with the stored
offset still `+0.3 C`. The owner then established that the decrement reaches
`-10.0 C` while the increment stops at `+9.9 C`. That secondary endpoint
defect is ticketed; the primary safety path is the reference-device calibration
screen under Temperature Setup. Remaining interactive cases are in
`work/evidence/integrated-00-05-00-target-qualification-20260724/controller-a-calibration-matrix.md`.
No physical power interruption was performed.

Later on 2026-07-24, Controller A alone was programmed and verified with the
replacement calibration candidate from source commit `23c6cc7f`, artifact
commit `2616e275`, Debug S-record SHA-256
`00F6430B7EDD65B222065807B4D8D95EA15DFA029F08DAFD8826103096ECB93A`.
The owner confirmed reference-calibration single-tap and press/hold/release
behavior in both directions. An isolated one-step Cancel test then passed:
VEEPROM was byte-identical, the stored offset remained effectively `+0.3 C`,
the temporary copy reflected the unsaved step, and no save request was pending.

The negative repeated-save customer-calibration case passed on 2026-07-24.
After save/reopen and another decrement attempt, read-only J-Link state showed
both stored and temporary offsets exactly `-10.0 C`, raw temperature about
`22.3 C`, and displayed/adjusted temperature about `12.3 C`.

The earlier candidate was restored to the intended offset through service 514,
with same-run J-Link readback at approximately `+0.300004 C`.

The replacement candidate subsequently passed reference-calibration
tap/hold/release, isolated Cancel, upper-boundary clamp, service-514 positive
endpoint with the former stale state injected, and bounded `+10.0 C` Save.
The owner then restored displayed `+0.3 C`. Same-run J-Link readback confirmed
`gRamConfig.MainTemp_Offset=0x3E99986D` (`+0.2999910 C`), zero pending save
request, zero system errors, and identical redundant VEEPROM copies with
SHA-256
`1010B47AB5E07E58F3389D7A1D6E0E0A4489DE2244FF23D711FBE71D2EEB157E`.
The card remained ready with valid CID and representative read; raw CID
`035344535330384780976F0084018559`. Controller B was not accessed.

A following requested negative-endpoint Save attempt did not produce a
negative record. Immediate readback found a clean redundant
`0x411FFFF0` (`+9.999985 C`) record at generation `212`, VEEPROM SHA-256
`524D0FAA537B6FBCA0FC0747BF7C120F2AE29C19A3168E5A0191243EBF3A38FA`.
The pre-Save display was not reported, so the attempt is inconclusive rather
than a firmware failure. Controller A must be restored to displayed `+0.3 C`
before unrelated testing.

On the deliberate retry, the owner confirmed that service 514 stopped at
displayed `-9.9 C` and saved it. Readback found `0xC11E667E`
(`-9.9000225 C`) in identical generation-214 VEEPROM copies, SHA-256
`9ED0AB60C79C7461AD0D912B47261820BCC518A0AD96294C2302C03FC50265FE`.
The root cause is accumulated binary-float error followed by rejection of the
slightly overshooting next step. The correction passes automated gates and is
frozen in source commit `b76c6fb2` and artifact commit `044022ba`.

The owner restored displayed `+0.3 C`. Readback confirmed
`gRamConfig.MainTemp_Offset=0x3E999735` (`+0.2999817 C`), identical redundant
generation-216 copies, VEEPROM SHA-256
`BED1340CBE6E770F5F20F27F0D5BBAEEA816E06642076DC38735EEC9C63F82C5`,
zero save request, zero system errors, and unchanged SD CID/status.

Three authorized programming attempts initially stopped before `loadfile`
because J-Link `69664082` could not be acquired through the Pi remote server.
The owner requested a Pi reboot and then authorized bouncing the exact Windows
Disable/Enable remote-route shortcuts. SSH recovered, the Pi reported a fresh
boot, and J-Link `69664082` passed its serial-bound service health check.

The hash-locked retry programmed and verified corrected Debug S-record
`1D047B144B6DFA88447212ED2670C3F7B636D24D29B5A20B100453A4806B2978`
on Controller A. Post-flash readback preserved `0x3E999735`
(`+0.2999817 C`), zero save request, zero system errors, ready SD status, valid
CID and representative read, and raw CID
`035344535330384780976F0084018559`. The 4 KiB VEEPROM SHA remains
`BED1340CBE6E770F5F20F27F0D5BBAEEA816E06642076DC38735EEC9C63F82C5`,
byte-identical to the safe pre-flash capture. Controller B was not operated.

The corrected candidate then passed service-514 lower endpoint, bounded Save,
safe restoration, and timeout-without-Save. The owner also completed the
reference-calibration held-decrement test: the temporary value reached exactly
`-10.0 C` and remained there after continued hold plus three extra taps while
stored configuration remained `+0.3 C`. Save produced identical CRC-valid
generation-222 VEEPROM copies. The final service-514 restoration produced
stored `0x3E999A27` (`+0.3000042 C`) in identical CRC-valid generation-224
copies, VEEPROM SHA-256
`D455DE6F843FFB5FB6F3095AADE83D053D6B2FE35B98D956FECBE4A77A0FA1CD`.
Save request and system errors were zero. SD status remained ready with valid
CID and representative read; raw CID
`035344535330384780976F0084018559`. Controller B was not operated and no
physical power interruption occurred.

## Historical evidenced firmware and target state

The lines below retain the 2026-07-20 `00.04.13` connected-bench baseline.
Later dated evidence under
`work/evidence/hpt-log-store-circular-mirror/controller-a/` shows Controller A
running `00.05.00` HPT media 2.0 development images through wrap/export and the
manual data-durable interruption test. Treat neither as a current live-state
claim; recapture exact commit/image hashes and CID before the next run.

- Engineering candidate: `00.04.13`
- Approved Debug S-record SHA-256 used for end-to-end connector proof:
  `8A3B43AF3BC35922655DB5216F4ED7C93AA829E03ACC2FEC65961B9232CDAA80`
- Programming/verify: pass
- Source-mapped hardware breakpoint/backtrace/step/detach: pass
- Final CPU state: running
- Nominal SD state after load: ready; mirrored periodic logging passed with
  zero system/SD errors
- Temperature alarm: owner reports re-enabled on 2026-07-20; electronic
  readback and alarm-function regression remain pending

## Service 474 snapshot

- Final state: 3 (Stopped), intentionally stopped by owner
- Run: `86275`
- Elapsed seconds: `5513`
- Completed records: `5432`
- Dropped/mirror/I/O/latched failures: all zero
- Reset count: `6`
- Live and persisted records matched after programming/debug qualification

## Open inventory and authority fields

- Connected sensor availability and scope (2026-07-24): RTD main temperature
  only. RTD is the only main-temperature sensor in the approved qualification
  scope. LM73 testing is explicitly excluded and no LM73 qualification claim
  will be made. Other unavailable non-temperature sensor cases remain deferred
  until the office setup is restored. RTD-compatible non-power cases remain
  executable.
- Installed SD identity: owner-identified SanDisk EDGE 8 GB; MID `0x03`, OEM
  `SD`, controller-reported product `SS08G`, revision `8.0`, serial
  `0x976F0084`, May 2024, raw CID
  `035344535330384780976F0084018559`; capacity `15,523,840` sectors
- Automated power: the owner received the remaining hardware on 2026-07-24 and
  is finishing external HPT Lab Power Control on `srv-rpi00.local`; no
  power-cut/endurance run is permitted until the canonical interface is
  reported ready and G-PWR passes
- Configuration and calibration backup/hash:
  `work/evidence/integrated-00-05-00-target-qualification-20260724/setup-a/preflash-veeprom.bin`,
  SHA-256 `E5AABD742D00C7E65B90D6C017553E2041A2947EC77A5F1C4AF3BE489DE8A186`
- Known-good recovery image/hash:
  `work/evidence/integrated-00-05-00-target-qualification-20260724/setup-a/preflash-program.bin`,
  SHA-256 `946C1F9D0E1BF631D457EBA08895F151F0BAF819924AD753B34F156F215A91E0`
- Heater/load/independent protection and emergency disconnect: pending
- Destructive media, power-cut, or long-duration authority: operation-specific;
  do not inherit authority from the historical campaign record

Historical Setup A evidence through 2026-07-16 used controller `1751219` and is
retained in `smi-setup-a.md`. It must not be relabeled as coverage for this
current connected controller.
