# Device paths

Every path here was read back off a running LivePremier before it was used.

**Firmware: 6.2.73.** Derived on the simulator (`NLC_CMAX`) and then
**re-verified leaf-by-leaf on a real Aquilon C (`NLC_C`) — 21 of 21 paths
resolved, zero failures.** Analog Way moved paths
at AWJ guide v4.0 and again before 6.2 — `$screenGroup` is gone on this
firmware, replaced by `$screenAuxGroup`, and the guide's own subscription
example would fail as printed. Treat this table as firmware-tagged and re-verify
against anything else.

## Two spellings, one model

The same object model is addressed differently by the two transports:

| | AWJ (TCP 10606) | Web RCS store (WebSocket, HTTP snapshot) |
|---|---|---|
| root | `DeviceObject` | `device` |
| collection | `$name` | `nameList` |
| item key | `@items/KEY` | `items`, `KEY` |
| property | `@props/name` | `pp`, `name` |

`src/lang/paths.ts` holds a path once and renders either. The `List` suffix
belongs to the store spelling only: AWJ answers E12 for `$screenList`.

## Memories

Note the asymmetry, which is the device's: **a load is addressed slot-first, a
save target-first.**

```
screen recall  presetBank/control/load/$slot/N/$screen/S<n>/$preset/{PREVIEW|PROGRAM}/xRequest
aux recall     presetBank/control/load/$slot/N/$auxiliary/A<n>/$preset/{…}/xRequest
screen store   presetBank/control/save/$screen/S<n>/$preset/{…}/$slot/N/xRequest
aux store      presetBank/control/save/$auxiliary/A<n>/$preset/{…}/$slot/N/xRequest

master recall  masterPresetBank/control/load/$slot/N/$preset/{…}/xRequest
master store   masterPresetBank/control/save/$slot/N/xRequest
               …preceded by the filter writes below

layer recall   layerBank/control/load/$slot/N/$screen/S<n>/$preset/{…}/$layer/L/xRequest
layer store    layerBank/control/save/$screen/S<n>/$preset/{…}/$layer/L/$slot/N/xRequest

mv recall      monitoringBank/control/load/$slot/N/$output/n/xRequest
mv store       monitoringBank/control/save/$output/n/$slot/N/xRequest

take           $screenAuxGroup/S<n>/control/xTake
```

## The master record mask

`Store Master` takes no screen argument. It takes filters, which persist on the
device until overwritten — so an unfiltered store must write them wide open
rather than leaving whatever was there.

```
masterPresetBank/control/save/@props/mode            SAVE_FROM_PGM | SAVE_FROM_PVW
masterPresetBank/control/save/@props/screenFilter    ["S1"…"S24"]
masterPresetBank/control/save/@props/auxFilter       ["A1"…"A96"]
masterPresetBank/control/save/@props/layerFilter     ["NATIVE","1"…"128"]
masterPresetBank/control/save/@props/categoryFilter  SOURCE POS SIZE OPACITY
                                                     CROPPING BORDER TRANSITIONS
                                                     EFFECTS FLYING_CURVE TIMING
                                                     SPEED CUT_AND_FILL MASK KEYER
```

Filters first, trigger last. Both transports preserve ordering on one
connection, so the whole thing goes out as a burst.

`presetBank` and `layerBank` have **no** `control/save/@props/mode` — those
banks carry no mask on this firmware, which is why `If` is refused on them.

## Memory metadata

```
<bank>/$bank/@items/N/control/@props/label      the memory's name
<bank>/$bank/@items/N/control/@props/xDelete    erase it
<bank>/$bank/@items/N/status/@props/isValid     whether it holds anything
```

## Ranges

Confirmed on a running device, not inferred from probing:

| Dimension | Range |
|---|---|
| `PEMEM_BANK_SLOT` — screen/aux memory | 1–1000 |
| `MASTERMEM_BANK_SLOT` | 1–500 |
| `LAYERMEM_BANK_SLOT` | 1–50 |
| `MONITORING_BANK_SLOT` | 1–50 |
| `SCREEN` | S1–S24 |
| `AUXILIARY` | A1–A96 |
| `SCREEN_LAYER` | NATIVE, 1–128 |
| `PRESET` | A, B, C |

Probing AWJ for `E12` finds the *model's* maximum, not what a chassis has
configured — and it disagrees with this table. Use the table.

## Feedback, and its traps

**Preview and Program are not buffers.** The device keeps three fixed buffers
per screen — `A`, `B`, `C` — and preview/program are names for whichever is
pending or live. A take swaps which is which. Control paths use
`PREVIEW`/`PROGRAM`; status paths report `A`/`B`/`C`:

```
presetBank/status/presetId/$screen/S<n>/$preset/{A|B|C}/@props/id
presetBank/status/presetId/$screen/S<n>/$preset/{A|B|C}/@props/isNotModified
```

**A recall of an empty memory produces silence.** Observed on 6.2.73, recalling
into S1 preview:

```
slot 1 (populated)          slot 5 (empty)
  xRequest = true             xRequest = true
  xRequest = true             xRequest = true
  presetId B id = 0           (nothing further)
  isLoading = true
  presetId B isNotModified = false
  presetId B id = 1
  presetId B isNotModified = true
  isLoading = false
```

So `isLoading` never appearing is the tell, and it is what Mynah reports as an
empty memory. There is no error and no negative acknowledgement.

**Reproduced identically on a real Aquilon C**, not just the simulator.

## The echo carries the value you wrote

A write is acknowledged by the device pushing the same path back. Match on the
*value you sent*, not on `true`: the triggers are booleans, but a label is a
string and the master-store filters are arrays, and a client that only accepts
`true` will report those as unconfirmed forever while they land perfectly well.
The device also pushes a trigger back to `false` afterwards, which is not an
acknowledgement of anything. Found on hardware — a `Label` that worked was
being reported as failed.

## The HTTP side

```
GET /api/device/snapshots/{inputs|images|outputs|multiviewers|timers}/<n>
```
Live thumbnails, unauthenticated, PNG.

```
GET /api/stores/device
```
The **entire** store as one document — 124 MB on a populated chassis. There is
no narrowing: sub-paths 404 and a `?path=` query is ignored. It also carries
**no `Access-Control-Allow-Origin`**, so a page served from anywhere but the
device itself cannot read it at all. The WebSocket has no such restriction and
accepts any origin.

That is why Mynah's memory index is opt-in and usually unavailable, and why the
empty-memory case is detected from the socket instead.