1
0
Files
h8-536-decoder/docs/pt2-panel-atlas.md
2026-05-28 10:12:08 +10:00

339 lines
17 KiB
Markdown

# PT2 Panel Atlas
This note tracks bench tests that intentionally drive visible panel outputs from
PT2/command-0 table writes.
## Scenario Files
Current compact atlas scenarios:
```text
scenarios/panel-atlas-operator-lamps-v1.json
scenarios/panel-atlas-readout-status-v1.json
scenarios/panel-atlas-right-stack-isolation-v1.json
scenarios/panel-atlas-right-stack-fresh-latch-v1.json
scenarios/panel-atlas-standard-master-bit-sweep-v1.json
scenarios/panel-atlas-standard-master-neighbor-sweep-v2.json
scenarios/panel-atlas-standard-master-lower-neighbor-sweep-v3.json
```
Both are designed for webcam capture with the calibrated bench settings:
```text
--parity E --camera-index 4 --snapshot-delays 0.5
```
## Run: 2026-05-27
Logs:
```text
captures/panel-atlas-operator-lamps-v1-webcam.txt
captures/panel-atlas-readout-status-v1-webcam.txt
captures/panel-atlas-right-stack-isolation-v1-webcam.txt
captures/panel-atlas-right-stack-fresh-latch-v1-webcam.txt
captures/panel-atlas-standard-master-bit-sweep-v1-webcam.txt
captures/panel-atlas-standard-master-neighbor-sweep-v2-webcam.txt
captures/panel-atlas-standard-master-lower-neighbor-sweep-v3-webcam.txt
```
Snapshots:
```text
captures/panel-atlas-operator-lamps-v1-webcam-shots/
captures/panel-atlas-readout-status-v1-webcam-shots/
captures/panel-atlas-right-stack-isolation-v1-webcam-shots/
captures/panel-atlas-right-stack-fresh-latch-v1-webcam-shots/
captures/panel-atlas-standard-master-bit-sweep-v1-webcam-shots/
captures/panel-atlas-standard-master-neighbor-sweep-v2-webcam-shots/
captures/panel-atlas-standard-master-lower-neighbor-sweep-v3-webcam-shots/
```
Serial health:
| Run | RX frames | TX frames | Resync | Dropped bytes |
| --- | ---: | ---: | ---: | ---: |
| operator lamps | 82 | 19 | 0 | 0 |
| readout/status | 76 | 16 | 0 | 0 |
| right-stack isolation | 115 | 26 | 0 | 0 |
| right-stack fresh latch | 20 | 12 | 0 | 0 |
| standard/master bit sweep v1 | 144 | 34 | 0 | 0 |
| standard/master neighbor sweep v2 | 188 | 47 | 0 | 0 |
| standard/master lower-neighbor sweep v3 | 188 | 47 | 0 | 0 |
The compact and isolation runs stayed in the expected table-readback plus
`02 00 02 00 00 5A` CONNECT-OK response rhythm.
The fresh-latch run ended with one trailing unframed byte after the final send.
There were no resync events or dropped bytes; this is consistent with the run
ending while the next response frame was just beginning.
## Confirmed Visible Effects
## Far-Right Stack Reference
The physical far-right stack, top to bottom, is:
```text
TALLY light, with camera number
STANDARD
MASTER
SLAVE
CAM POWER
BARS
```
Use these names instead of generic "right-side status" labels when reviewing
webcam crops.
Readout/status run:
| Frame | Selector/value | Visible effect |
| --- | --- | --- |
| `00 01 0F 08 00 5C` | `E000[0x008F]=0x0800` | SHUTTER display shows `EUS`/likely `EVS`; iris AUTO lamp is lit |
| `00 01 0F 10 00 44` | `E000[0x008F]=0x1000` | SHUTTER display shows `OFF`; iris AUTO lamp remains lit |
| `00 01 13 80 00 C8` | `E000[0x0093]=0x8000` | white-balance / black-flare lamp cluster changes, consistent with prior preset/manual observation |
| `00 01 13 90 20 F8` | `E000[0x0093]=0x9020` | black-flare manual-context candidate remains visible |
| `00 01 13 90 FF 27` | `E000[0x0093]=0x90FF` | black-flare auto-context candidate remains visible |
| `00 01 90 80 00 4B` | `E000[0x0110]=0x8000` | KNEE AUTO lamp lights |
The `0x00B9` gate writes in this compact run did not light KNEE AUTO by
themselves at the 0.5 s snapshot point. That matches the ROM model where
`0x00B9.13` is more of a report-path gate, while `0x0110.15` is the stronger
visible KNEE AUTO source.
Operator-lamp run:
| Frame | Selector/value | Visible effect |
| --- | --- | --- |
| `00 00 13 80 00 C9` | `E000[0x0013]=0x8000` | far-right SLAVE lamp lights, based on stack order and crop position |
| `00 00 17 80 00 CD` | `E000[0x0017]=0x8000` | far-right bottom white BARS lamp lights |
| `00 00 1A 80 00 C0` | `E000[0x001A]=0x8000` | lower right white lamp appeared lit in the compact run, later refined as likely `0x0017` carryover |
Right-stack isolation/fresh-latch refinement:
| Frame | Selector/value | Visible effect |
| --- | --- | --- |
| `00 00 15 80 00 CF` | `E000[0x0015]=0x8000` | CALL lamp lights |
| `00 00 15 00 00 4F` | `E000[0x0015]=0x0000` | CALL lamp clears |
| `00 00 13 80 00 C9` | `E000[0x0013]=0x8000` | far-right SLAVE lamp lights |
| `00 00 13 00 00 49` | `E000[0x0013]=0x0000` | SLAVE lamp clears |
| `00 00 17 80 00 CD` | `E000[0x0017]=0x8000` | far-right bottom white BARS lamp lights |
| `00 00 17 00 00 4D` | `E000[0x0017]=0x0000` | lamp remains lit; low write does not clear it |
| `00 00 1A 80 00 C0` | `E000[0x001A]=0x8000` | no independent BARS-lamp effect from fresh boot |
| `00 00 07 80 00 DD` / `00 00 07 00 00 5D` | `E000[0x0007]` high/low | no clear visible delta from CONNECT-OK baseline |
So `0x0017` is currently the strongest BARS lamp-on/latch selector.
`0x001A` should not be labeled as the same lamp source from the prior compact
run; that was likely carryover after `0x0017` had latched the white lamp on.
`0x0007` is still protocol-relevant because the real panel emits the matching
CAM POWER event frame, but the host-write visible lamp mapping is not separated
from the CONNECT-OK baseline yet.
ROM trace refinement: selector `0x0013` dispatches to `H'2E06`, reads current
table word `E800[0x0013]` at `H'E826`, and maps bit 15 to SLAVE while bit 14
maps to `IRIS/M.BLACK LINK`. See `docs/pt2-iris-mblack-link-rom-trace.md`.
STANDARD/MASTER hunt:
| Run | Tested selector/value pocket | Result |
| --- | --- | --- |
| `panel-atlas-standard-master-bit-sweep-v1` | `0x0012`, `0x0013`, `0x0014` with high-bit/high-nibble candidates | no clean STANDARD or MASTER lamp trigger; positive SLAVE control worked |
| `panel-atlas-standard-master-neighbor-sweep-v2` | `0x0010`, `0x0011`, `0x0015`, `0x0016`, `0x0017`, `0x0018`, `0x0019`, `0x001A` high-nibble candidates | no clean STANDARD or MASTER trigger |
| `panel-atlas-standard-master-lower-neighbor-sweep-v3` | `0x0008` through `0x000F` high-nibble candidates | no clean STANDARD or MASTER trigger |
The `v2` sweep did show the far-right bottom BARS lamp/latch from the
`0x0017` family when using non-`0x8000` high-nibble values. The clearest
transition is `00 00 17 40 00 0D` (`E000[0x0017]=0x4000`) lighting the same
bottom white BARS lamp that `0x0017=0x8000` can light. The lamp persisted after
`0x0017=0x0000`, matching the existing BARS latch behavior. Later `0x0018`
rows are contaminated by that latch and need a fresh-boot isolation run before
labeling `0x0018` as a separate BARS source.
## Next Atlas Step
The next useful run is still a clear/ack/state-transition probe for the
`0x0017` BARS latch, now including `0x0017=0x4000`:
- test likely sibling selectors around `0x0016`, `0x0018`, `0x0019`, and `0x001A`
from a fresh boot after `0x0017` has latched on,
- try command-7 repeat/ack and selector-zero refreshes to see whether the lamp
clears through a state-machine transition rather than a simple low write,
- test `0x0007` with an alternate baseline, because CONNECT OK already lights
CAM POWER and masks any host-write lamp delta.
- do not re-run `0x0008` through `0x0014` high-nibble values for
STANDARD/MASTER unless the ROM trace points back there; the webcam sweeps did
not show those lamps in that pocket.
## ROM-Derived Button Output Sweep
To skip the physical RCP button-press side and directly test likely "on" states
from ROM handlers, use:
```powershell
.\.venv\Scripts\python.exe scripts\build_rom_button_output_sweep.py
.\.venv\Scripts\python.exe scripts\serial_scenario.py scenarios\panel-atlas-rom-button-output-candidates-v1.json --parity E --quiet-console --log captures\panel-atlas-rom-button-output-candidates-v1-webcam.txt --result-json captures\panel-atlas-rom-button-output-candidates-v1-webcam-result.json --snapshot-dir captures\panel-atlas-rom-button-output-candidates-v1-webcam-shots --camera-index 4 --snapshot-delays 0.5
```
This scenario power-cycles before every candidate, seeds `CONNECT: OK`, sends
one command-0 selector/value from the ROM-derived candidate list, and captures a
single webcam image 0.5 s later. It is intentionally slower than a continuous
sweep but should avoid most latch/carryover ambiguity.
Run result notes from `captures/panel-atlas-rom-button-output-candidates-v1-webcam-shots/`:
| Case | Frame | Selector/value | Visible result |
| --- | --- | --- | --- |
| 001 | `00 00 13 40 00 09` | `0x0013 = 0x4000` | IRIS/M.BLACK LINK |
| 002 | `00 00 13 80 00 C9` | `0x0013 = 0x8000` | SLAVE |
| 003 | `00 00 15 80 00 CF` | `0x0015 = 0x8000` | CALL and red tally |
| 004 | `00 00 17 80 00 CD` | `0x0017 = 0x8000` | BARS |
| 005 | `00 01 90 80 00 4B` | `0x0110 = 0x8000` | KNEE AUTO |
| 006 | `00 00 1A 08 08 40` | `0x001A = 0x0808` | MONITOR ENC |
| 007 | `00 00 1A 20 20 40` | `0x001A = 0x2020` | MONITOR B |
| 008 | `00 00 1A 40 40 40` | `0x001A = 0x4040` | MONITOR G |
| 009 | `00 00 1A 80 80 40` | `0x001A = 0x8080` | MONITOR R |
| 010 | `00 00 6B 80 00 B1` | `0x006B = 0x8000` | STANDARD |
| 011 | `00 01 03 00 04 5C` | `0x0083 = 0x0004` | IRIS AUTO, SHUTTER OFF, MASTER GAIN HP |
| 012 | `00 01 03 40 00 18` | `0x0083 = 0x4000` | IRIS AUTO, SHUTTER OFF, MASTER GAIN 0 |
| 013 | `00 01 03 20 00 78` | `0x0083 = 0x2000` | IRIS AUTO, SHUTTER OFF, MASTER GAIN 3 |
| 014 | `00 01 0F 80 00 D4` | `0x008F = 0x8000` | IRIS AUTO, SHUTTER begins with `1...` |
| 015 | `00 01 0F 20 00 74` | `0x008F = 0x2000` | IRIS AUTO, SHUTTER `00.0` |
| 016 | `00 01 0F 08 00 5C` | `0x008F = 0x0800` | IRIS AUTO, SHUTTER EVS |
| 017 | `00 01 0F 10 00 44` | `0x008F = 0x1000` | IRIS AUTO, SHUTTER OFF |
| 018 | `00 01 13 10 20 78` | `0x0093 = 0x1020` | BLACK/FLARE MANUAL, white balance MANUAL |
| 019 | `00 01 13 40 40 48` | `0x0093 = 0x4040` | BLACK/FLARE AUTO, white balance AUTO |
| 020 | `00 01 13 80 40 88` | `0x0093 = 0x8040` | BLACK/FLARE AUTO, white balance PRESET |
| 021 | `00 01 13 00 20 68` | `0x0093 = 0x0020` | BLACK/FLARE MANUAL, white balance MANUAL |
| 022 | `00 01 13 00 40 08` | `0x0093 = 0x0040` | BLACK/FLARE AUTO, white balance MANUAL |
| 023 | `00 01 1A 08 00 49` | `0x009A = 0x0800` | no panel change observed |
| 024 | `00 01 37 20 00 4C` | `0x00B7 = 0x2000` | no panel change observed |
The run directory contains 28 candidate photos. The user-supplied ordered notes
covered the first 24, so cases 025-028 still need visual review before assigning
meanings:
| Case | Frame | Selector/value | Candidate |
| --- | --- | --- | --- |
| 025 | `00 01 39 40 00 22` | `0x00B9 = 0x4000` | F6DC.7 handler value candidate |
| 026 | `00 01 44 80 00 9F` | `0x00C4 = 0x8000` | F6D4.0 bundle selector candidate |
| 027 | `00 01 46 80 00 9D` | `0x00C6 = 0x8000` | F6D4.0 bundle selector candidate |
| 028 | `00 01 78 80 00 A3` | `0x00F8 = 0x8000` | F6D4.1 handler candidate |
Key refinements from this run:
- `0x001A` is now best labeled as the MONITOR selector cluster: `ENC`, `B`, `G`,
and `R` appeared cleanly from the four packed values.
- `0x006B = 0x8000` is the first clean STANDARD lamp trigger.
- `0x0083` is a MASTER GAIN/status display word, with values observed for `HP`,
`0`, and `3`, while also lighting IRIS AUTO and showing SHUTTER OFF.
- `0x008F` carries local shutter display/value states beyond the earlier
EVS/OFF bits.
- `0x0093` now has stronger white-balance plus BLACK/FLARE field mapping:
high/mid bit combinations select WB AUTO/PRESET/MANUAL and BLACK/FLARE
AUTO/MANUAL together.
## Broad Visual Sweep Workflow
For exploratory lamp/readout mining, use the generated big sweep rather than
hand-writing thousands of frames:
```powershell
.\.venv\Scripts\python.exe scripts\build_panel_visual_sweep.py scenarios\panel-atlas-big-visual-sweep-0001-017f-highbits.json --start 0x0001 --end 0x017F --values 0x8000,0x4000,0x2000,0x1000,0x0800 --power-cycle-every 32 --ok-every 8 --listen 0.65 --clear-listen 0.15 --ok-listen 0.30
```
Run the generated scenario with webcam snapshots:
```powershell
.\.venv\Scripts\python.exe scripts\serial_scenario.py scenarios\panel-atlas-big-visual-sweep-0001-017f-highbits.json --parity E --quiet-console --log captures\panel-atlas-big-visual-sweep-0001-017f-highbits-webcam.txt --result-json captures\panel-atlas-big-visual-sweep-0001-017f-highbits-webcam-result.json --snapshot-dir captures\panel-atlas-big-visual-sweep-0001-017f-highbits-webcam-shots --camera-index 4 --snapshot-delays 0.5
```
The generated high-bit sweep covers selectors `0x0001` through `0x017F` with
five candidate values per selector, for 1,915 candidate snapshots. It power
cycles every 32 selectors to reduce latch contamination. Selector zero is
omitted because it controls the CONNECT OK baseline.
After the run, create labeled review sheets:
```powershell
.\.venv\Scripts\python.exe scripts\make_panel_sweep_contact_sheets.py captures\panel-atlas-big-visual-sweep-0001-017f-highbits-webcam-shots --output-dir captures\panel-atlas-big-visual-sweep-0001-017f-highbits-sheets --only-candidates --crop panel --cols 4 --rows 5 --thumb-width 360
```
If a sheet shows a visible change, the image label has the exact trigger form:
`candidate_XXXX_YYYY` means command-0 wrote `E000[0xXXXX]=0xYYYY`. Use that
selector/value in a smaller fresh-boot isolation scenario.
## Broad Sweep Findings
Run:
```text
captures/panel-atlas-big-visual-sweep-0001-017f-highbits-webcam.txt
captures/panel-atlas-big-visual-sweep-0001-017f-highbits-webcam-result.json
captures/panel-atlas-big-visual-sweep-0001-017f-highbits-webcam-shots/
```
Serial health:
| RX frames | TX frames | Resync | Dropped bytes | Snapshots |
| ---: | ---: | ---: | ---: | ---: |
| 706 | 2372 | 0 | 0 | 1916 |
User-reviewed new visible hits from the broad sweep:
| Candidate label | Frame | Selector/value | Reported visible effect |
| --- | --- | --- | --- |
| `candidate_0013_4000` | `00 00 13 40 00 09` | `E000[0x0013]=0x4000` | `IRIS/M.BLACK LINK` area/lamp |
| `candidate_0024_8000` | `00 00 24 80 00 FE` | `E000[0x0024]=0x8000` | LCD selector button/lamp |
| `candidate_0082_8000` | `00 01 02 80 00 D9` | `E000[0x0082]=0x8000` | IRIS readout shows `OP` |
| `candidate_0082_4000` | `00 01 02 40 00 19` | `E000[0x0082]=0x4000` | IRIS readout shows `1.4` |
| `candidate_0083_8000` | `00 01 03 80 00 D8` | `E000[0x0083]=0x8000` | MASTER GAIN readout shows `-3` |
| `candidate_0093_8000` | `00 01 13 80 00 C8` | `E000[0x0093]=0x8000` | white-balance PRESET lamp |
| `candidate_0093_4000` | `00 01 13 40 00 08` | `E000[0x0093]=0x4000` | white-balance AUTO lamp |
| `candidate_0093_2000` | `00 01 13 20 00 68` | `E000[0x0093]=0x2000` | white-balance MANUAL lamp |
The serial log shows immediate command-4 table readback frames for these writes,
so the RCP accepted the selector updates.
Fresh-boot isolation scenario:
```powershell
.\.venv\Scripts\python.exe scripts\serial_scenario.py scenarios\panel-atlas-big-hits-isolation-v1.json --parity E --quiet-console --log captures\panel-atlas-big-hits-isolation-v1-webcam.txt --result-json captures\panel-atlas-big-hits-isolation-v1-webcam-result.json --snapshot-dir captures\panel-atlas-big-hits-isolation-v1-webcam-shots --camera-index 4 --snapshot-delays 0.5
```
Create review sheets for that isolation run:
```powershell
.\.venv\Scripts\python.exe scripts\make_panel_sweep_contact_sheets.py captures\panel-atlas-big-hits-isolation-v1-webcam-shots --output-dir captures\panel-atlas-big-hits-isolation-v1-sheets --crop panel --cols 3 --rows 4 --thumb-width 420
```
Fresh-boot isolation results:
| Frame | Selector/value | Confirmed visible effect |
| --- | --- | --- |
| `00 00 13 40 00 09` | `E000[0x0013]=0x4000` | `IRIS/M.BLACK LINK` lamp |
| `00 00 24 80 00 FE` | `E000[0x0024]=0x8000` | LCD selector-button lamp |
| `00 00 24 00 00 7E` | `E000[0x0024]=0x0000` | same LCD selector-button lamp remained visible at 0.5 s |
| `00 01 02 80 00 D9` | `E000[0x0082]=0x8000` | IRIS readout `OP` |
| `00 01 02 40 00 19` | `E000[0x0082]=0x4000` | IRIS readout `1.4` |
| `00 01 02 00 00 59` | `E000[0x0082]=0x0000` | IRIS readout blank |
| `00 01 03 80 00 D8` | `E000[0x0083]=0x8000` | IRIS AUTO lamp, SHUTTER `OFF`, MASTER GAIN `-3` |
| `00 01 03 00 00 58` | `E000[0x0083]=0x0000` | same IRIS AUTO / SHUTTER `OFF` / MASTER GAIN `-3` state remained visible at 0.5 s |
| `00 01 13 80 00 C8` | `E000[0x0093]=0x8000` | BLACK/FLARE MANUAL plus white-balance PRESET |
| `00 01 13 40 00 08` | `E000[0x0093]=0x4000` | BLACK/FLARE MANUAL plus white-balance AUTO |
| `00 01 13 20 00 68` | `E000[0x0093]=0x2000` | BLACK/FLARE MANUAL plus white-balance MANUAL |
| `00 01 13 00 00 48` | `E000[0x0093]=0x0000` | BLACK/FLARE MANUAL plus white-balance MANUAL |
Interpretation:
- `0x0082` is a direct IRIS display/status selector. Clearing it blanks the IRIS
readout, so this one behaves like a simple display source.
- `0x0083=0x8000` drives a combined state: MASTER GAIN `-3`, SHUTTER `OFF`, and
IRIS AUTO. Clearing the selector did not visibly clear that state in the
isolation run, so it may be latched or copied into another display bank.
- `0x0093` is now a confirmed white-balance mode selector with
`0x8000=PRESET`, `0x4000=AUTO`, and `0x0000/0x2000=MANUAL` under this test
context. BLACK/FLARE MANUAL was also present for all tested `0x0093` states.
- `0x0024=0x8000` lights an LCD selector-button lamp, but `0x0024=0x0000` did
not clear it in this timing window.