diff --git a/content/core/firmware-updates/compatibility.mdx b/content/core/firmware-updates/compatibility.mdx
index f7a4254..559cd6d 100644
--- a/content/core/firmware-updates/compatibility.mdx
+++ b/content/core/firmware-updates/compatibility.mdx
@@ -1,6 +1,6 @@
---
title: 'Compatibility and releases'
-description: 'Verified hub and cover pairing, firmware artifacts, and platform architecture.'
+description: 'Verified hub and cover pairing, firmware artifacts, sensor data paths by firmware, and platform architecture.'
---
import { Callout } from 'nextra/components'
@@ -68,6 +68,41 @@ Every dynamic dependency (`libnats.so.3.11`, `libsystemd`, `libudev`, `libstdc++
+## Sensor data by firmware
+
+Firmware updates have changed how sensor readings reach Core more than once. The Pod generation does not tell you which path a hub uses; the firmware build does. Find yours with `sp-status` (the `rat_version` line) and compare it with this table.
+
+| Firmware | How sensor readings arrive | NATS subjects |
+| :-------------------------------------------------------- | :---------------------------------------------------------------------------------------------- | :-------------------------------------------------------- |
+| `.RAW` firmware (for example `8df35cc8` on a Pod 5 J55) | `.RAW` files. No `nats-server` and no `/persistent/jetstream`. | None |
+| NATS firmware (April 2026 onward, for example `ca35aafa`) | Local NATS JetStream stream `raw`, stored in `/persistent/jetstream`. | `raw.sens.*`, `raw.frz.*`, and the `raw.log` firmware log |
+| EmmcWear firmware (`4b7552d5`, built 3 November 2026) | Local NATS under a new prefix. The update adds an `EmmcWear` flag to cut eMMC reads and writes. | `rawmem.sens.*`, `rawmem.frz.*` |
+
+
+
+**EmmcWear firmware (`4b7552d5`) leaves biometrics empty on older Core builds.**
+
+Core builds before [core#917](https://github.com/sleepypod/core/pull/917) read only `raw.*` subjects. On `4b7552d5` the readings move to `rawmem.*`, so Core receives nothing even though heating, cooling, and the web UI work normally. The signs are:
+
+- The Core journal logs `[natsSource] no frames received 60s after subscribing`.
+- **Settings → Device → Sensor data source** shows **No frames** or **Stalled**.
+- **Calibrate all** fails because piezo and bed-temperature samples stay at zero.
+
+Update Core to a build that includes core#917, then run **Calibrate all** with the bed empty once readings appear in **System → Sensors**.
+
+If a later firmware moves the readings to yet another prefix, you don't need a new Core build. With [core#918](https://github.com/sleepypod/core/pull/918), **Settings → Device → Sensor data source** lists the prefixes the firmware publishes sensor (`sens.*`) readings under, up to 16, and lets you pin one. Prefixes that carry only freezer (`frz.*`) readings are not listed; if only those readings are missing, inspect the NATS subjects on the Pod. An agent connected over [MCP](/core/mcp/#sensor-data-source) can do the same with `get_sensor_source` and `set_device_settings`.
+
+
+
+Core builds with core#917 subscribe to both prefixes and choose one per reading type (`sens.piezo`, `frz.temp`, and so on). `raw` always wins: `rawmem` feeds a reading type only when `raw` has not delivered it within 30 seconds of Core subscribing to NATS, or when `raw` has gone silent on it for 60 seconds. A hub that publishes `raw` keeps exactly the data path it had before, and a firmware that publishes both prefixes never feeds a reading in twice. The journal records each choice:
+
+```bash
+journalctl -u sleepypod -n 500 --no-pager | grep 'natsSource\] reading'
+# [natsSource] reading sens.piezo from rawmem.sens.piezo
+```
+
+The `rawmem` prefix was found on a community Pod 5 (J55) on 8 November 2026 and confirmed over SSH. It has not yet been checked on every hub generation. If your hub has empty biometrics on firmware not listed here, include the `rat_version` and the `natsSource` journal lines when you report it.
+
## Sources and verification scope
-**Source reference:** [`sp-status` output format](https://github.com/sleepypod/core/blob/9bbc7232924fd2b18c7176c259556151aecc6412/scripts/bin/sp-status) · [Root SSH key resolution in the Core installer](https://github.com/sleepypod/core/blob/9bbc7232924fd2b18c7176c259556151aecc6412/scripts/lib/ssh-helpers) · [free-sleep firmware reset and slot check](https://github.com/throwaway31265/free-sleep/blob/e5172139874a274d1ced12c8da052ab2cbaa286d/INSTALLATION.md) · [Archived Pod 3 `.swu` releases](http://web.archive.org/web/*/fw.8slp.net/pod3-updates/pod3/releases/*). Stream limits, subject rates, serial port mapping, and the Slot B `sshd` configuration were verified on an MT8365 Pod 3 hub (`pn 20500`, hardware revision `0x0504`) with a Pod 4 cover (`pn 20600-0001-I12`, hardware `0812`), using artifacts extracted from releases `77af88bd8b941fed81786c09a9ac3cdb1e97b126` (`v1.5.156`, `v5.5.9`) and `3dd72bdefafbcfc202d6715fc45f9c7bbb3e02ff` (`v1.6.114`, `v5.5.22.0`).
+**Source reference:** [`sp-status` output format](https://github.com/sleepypod/core/blob/9bbc7232924fd2b18c7176c259556151aecc6412/scripts/bin/sp-status) · [Root SSH key resolution in the Core installer](https://github.com/sleepypod/core/blob/9bbc7232924fd2b18c7176c259556151aecc6412/scripts/lib/ssh-helpers) · [free-sleep firmware reset and slot check](https://github.com/throwaway31265/free-sleep/blob/e5172139874a274d1ced12c8da052ab2cbaa286d/INSTALLATION.md) · [Archived Pod 3 `.swu` releases](http://web.archive.org/web/*/fw.8slp.net/pod3-updates/pod3/releases/*) · [`rawmem` subject support (core#917)](https://github.com/sleepypod/core/pull/917). Stream limits, subject rates, serial port mapping, and the Slot B `sshd` configuration were verified on an MT8365 Pod 3 hub (`pn 20500`, hardware revision `0x0504`) with a Pod 4 cover (`pn 20600-0001-I12`, hardware `0812`), using artifacts extracted from releases `77af88bd8b941fed81786c09a9ac3cdb1e97b126` (`v1.5.156`, `v5.5.9`) and `3dd72bdefafbcfc202d6715fc45f9c7bbb3e02ff` (`v1.6.114`, `v5.5.22.0`).
diff --git a/content/core/mcp.mdx b/content/core/mcp.mdx
index 4525bca..7f53292 100644
--- a/content/core/mcp.mdx
+++ b/content/core/mcp.mdx
@@ -93,11 +93,35 @@ The server is intent-shaped, not endpoint-shaped. Each tool describes something
| "What's controlling the bed tonight?" | `get_automations` |
| "Pause all automations" / "Make a rule that cools when HR > 60" | `manage_automation` |
| "Something's wrong" / "Is the bed actually cooling?" | `diagnose_pod` |
+| "Why are my biometrics empty?" / "Is sensor data arriving?" | `get_sensor_source` |
| "Show me the logs" | `get_logs` |
| "Clear the pump alert" / "Restart the piezo service" / "Update" | `pod_maintenance` |
Temperatures accept your unit (`F` or `C`, defaulting to the device setting) and are converted to the 55–110°F hardware setpoint. Errors come back as readable text with the Core error code, so the assistant can explain "pump stall guard tripped" instead of retrying.
+## Sensor data source
+
+`get_sensor_source` reports where Core reads raw sensor frames from. It returns three blocks:
+
+- `firmware`: the firmware generation.
+- `stream`: whether Core reads .RAW files or NATS, and the age of the last frame.
+- `nats`: only while Core is reading NATS, on builds that include [core#918](https://github.com/sleepypod/core/pull/918):
+ - `mode` (`auto` or `pinned`) and `configuredPrefix`
+ - `subscribedSubjects`
+ - `sources`, the prefix feeding each reading type, for example `{ "sens.piezo": "rawmem" }`
+ - message counters
+ - `observedPrefixes`: the prefixes seen publishing `.sens.*` (up to 16), with `subscribed: false` for one Core does not read
+
+This is how an agent fixes a firmware that moves its data to a new prefix, without anyone patching code:
+
+1. Biometrics are empty, or the stream shows no frames.
+2. `get_sensor_source` shows no frames arriving, and `observedPrefixes` lists a prefix with `subscribed: false`.
+3. `set_device_settings` with `natsSubjectPrefix` set to that prefix. `null` returns to auto.
+4. The live stream re-subscribes at once. The biometrics modules read the prefix at startup, so the reply tells the agent to restart them with `pod_maintenance restart_service`. The agent must get your confirmation before each restart: the tool does not ask by itself, although it is marked destructive, so most hosts show an approval prompt.
+5. `get_sensor_source` again to confirm frames are arriving.
+
+The server tells agents to check `get_sensor_source` before changing anything when live data or biometrics are empty. See [sensor data by firmware](/core/firmware-updates/compatibility/#sensor-data-by-firmware) for the prefixes known so far.
+
## Sleep-stage rules
| Request | Tool |
@@ -157,4 +181,4 @@ Developer and installer surfaces stay REST-only: raw hardware opcodes, raw senso
---
-**Source reference:** [MCP server README and intent catalogue](https://github.com/sleepypod/core/blob/f8948b04586542fab1817aadc72b65a9c2988f8c/src/mcp/README.md) · [Server and prompts](https://github.com/sleepypod/core/blob/f8948b04586542fab1817aadc72b65a9c2988f8c/src/mcp/server.ts) · [HTTP endpoint and origin check](https://github.com/sleepypod/core/blob/f8948b04586542fab1817aadc72b65a9c2988f8c/app/api/mcp/route.ts)
+**Source reference:** [MCP server README and intent catalogue](https://github.com/sleepypod/core/blob/f8948b04586542fab1817aadc72b65a9c2988f8c/src/mcp/README.md) · [Server and prompts](https://github.com/sleepypod/core/blob/f8948b04586542fab1817aadc72b65a9c2988f8c/src/mcp/server.ts) · [HTTP endpoint and origin check](https://github.com/sleepypod/core/blob/f8948b04586542fab1817aadc72b65a9c2988f8c/app/api/mcp/route.ts) · [Sensor source tool and subject prefix setting (core#918)](https://github.com/sleepypod/core/pull/918)
diff --git a/content/core/settings/device.mdx b/content/core/settings/device.mdx
index 4aeefa0..bc862ef 100644
--- a/content/core/settings/device.mdx
+++ b/content/core/settings/device.mdx
@@ -69,18 +69,30 @@ When protection turns a side off, the Temp screen shows a banner. **Re-enable**
## Sensor data source
-This card is read-only. It tells you which firmware pipeline the Pod is on and whether sensor frames are reaching Core, which is the first thing to check when biometrics are empty.
+This card tells you which firmware pipeline the Pod is on and whether sensor frames are reaching Core, which is the first thing to check when biometrics are empty. On NATS firmware it also shows which NATS subjects Core reads, and lets you change them.
-- **Firmware.** How the Pod's firmware hands sensor data to Core. Newer firmware (April 2026 and later) publishes to a local **NATS JetStream** server. Older firmware writes **.RAW** files, either on a tmpfs through a `frank.sh` shim or a `frank.service` drop-in, or straight to `/persistent`.
+- **Firmware.** How the Pod's firmware hands sensor data to Core. Newer firmware (April 2026 and later) publishes to a local **NATS JetStream** server. Older firmware writes **.RAW** files, either on a tmpfs through a `frank.sh` shim or a `frank.service` drop-in, or straight to `/persistent`. EmmcWear firmware (`4b7552d5`) still uses NATS but under new `rawmem` subjects. On Core builds without [core#917](https://github.com/sleepypod/core/pull/917) this card then shows **No frames**. See [sensor data by firmware](/core/firmware-updates/compatibility/#sensor-data-by-firmware).
- **Reading from.** The source Core chose when the service started: **NATS stream** or **.RAW files**. Core detects this itself, once per start, so after a firmware update it needs a restart to detect again. **Check for updates** in [Updates](/core/settings/updates/) rebuilds and restarts Core on your current channel. **Restart service** on this page also restarts it, but in Core v3.0 it runs the **main** channel updater, so skip it on **dev** (see the warning under Reconnect and Restart service).
- **Last frame.** How long ago the newest sensor frame arrived, and its type. **Live** means frames are flowing. **Stalled** means nothing has arrived for over a minute.
+Core builds that include [core#918](https://github.com/sleepypod/core/pull/918) add three rows. **Subjects** and **Publishing prefixes** appear while Core is reading NATS. **Subject prefix** also appears when Core reads .RAW files on NATS firmware, for example when an override forces .RAW.
+
+- **Subjects.** Whether Core picks the subject prefix itself (`auto`) or reads one you chose (`pinned`), and which prefix feeds each reading type, for example `raw: sens.capsense, sens.piezo · rawmem: frz.temp`.
+- **Publishing prefixes.** The prefixes the firmware publishes sensor (`sens.*`) readings under, up to 16, and how recently, for example `raw 2s ago · rawv2 (not read) 1s ago`. A prefix marked **(not read)** is one Core does not subscribe to.
+- **Subject prefix.** Choose **Auto**, **raw**, **rawmem**, or **Custom…**. **Auto** reads `raw` and `rawmem`, preferring `raw` for each reading type. Pin a prefix only if the firmware publishes somewhere else. A custom prefix takes letters, digits, `_` or `-`, up to 32 characters, with no dots or wildcards.
+
+If no frames arrive and the firmware publishes under a prefix Core doesn't read, the card says so: "Firmware publishes under rawv2, which this pod doesn't read. Set the subject prefix to rawv2."
+
+After you change the prefix, the live sensor stream re-subscribes straight away. The biometrics modules read the prefix when they start, so the card offers **Restart biometrics modules**. That restarts the piezo processor, sleep detector, and environment monitor. The calibrator picks up the change on its next restart.
+
A note appears under the rows when the chosen source does not match the firmware, or when an environment override (`PIEZO_SENSOR_SOURCE` or `PIEZO_NATS_DISABLED`) is forcing a source. On a development machine without a Pod the card shows no firmware.
- There is no switch here on purpose. Core picks one source per start and never runs both at once,
- so a toggle would need a restart to take effect. The same information is in the [System
- dashboard](/core/system/) under **Health**, and on the Pod in `sp-status`.
+ There is no switch between .RAW files and NATS. Core picks one source per start and never runs
+ both at once, so a toggle would need a restart to take effect. The subject prefix control only
+ changes which NATS subjects Core reads. The same information is in the [System
+ dashboard](/core/system/) under **Health**, and on the Pod in `sp-status`. An agent can read it
+ and change the prefix through [MCP](/core/mcp/#sensor-data-source).
## Daily maintenance
diff --git a/content/developers/sensor-pipeline.mdx b/content/developers/sensor-pipeline.mdx
index ffc06f2..25d60c7 100644
--- a/content/developers/sensor-pipeline.mdx
+++ b/content/developers/sensor-pipeline.mdx
@@ -10,7 +10,7 @@ Firmware can emit CBOR RAW files or publish sensor records through local NATS. D
The source selector probes for NATS with a bounded startup grace period. A successful protocol greeting selects NATS; missing traffic is a health signal, not a reason to run both sources. If NATS is unavailable, RAW processing remains available. After bounded reconnect failures the process exits so systemd can restart and repeat source selection.
-NATS sensor consumers subscribe to `raw.sens.>` and `raw.frz.>` and decode each message as CBOR. They use live subscriptions, not durable JetStream replay. Restart therefore does not backfill historical frames. Calibrator builds a bounded live buffer and waits for enough samples. Cover-button NATS events and durable backfill remain separate follow-up work in the pinned source.
+NATS sensor consumers subscribe to `raw.sens.>` and `raw.frz.>`, and since [core#917](https://github.com/sleepypod/core/pull/917) also to `rawmem.sens.>` and `rawmem.frz.>`, which EmmcWear firmware `4b7552d5` uses. Each reading type is fed by one prefix: `raw` always wins, and `rawmem` is used only for a reading type `raw` has not delivered within 30 seconds of subscribing, or has left silent for 60 seconds ([sensor data by firmware](/core/firmware-updates/compatibility/#sensor-data-by-firmware)). Consumers decode each message as CBOR. They use live subscriptions, not durable JetStream replay. Restart therefore does not backfill historical frames. Calibrator builds a bounded live buffer and waits for enough samples. Cover-button NATS events and durable backfill remain separate follow-up work in the pinned source.
## Processing stages
diff --git a/content/troubleshooting/core.mdx b/content/troubleshooting/core.mdx
index db1b34f..5b5f8b5 100644
--- a/content/troubleshooting/core.mdx
+++ b/content/troubleshooting/core.mdx
@@ -51,7 +51,7 @@ Never build on the Pod. See [Core development](/developers/core/#never-build-on-
1. Check the selected side and date. For live measurements, confirm the bed is occupied.
2. Open **Settings → Device → Sensor data source**. It names the firmware pipeline (NATS or .RAW), the source Core chose at startup, and the age of the last frame, and it explains a fallback or an override. For module state use **System → Health**, and for live input use **Sensors**.
-3. On NATS firmware, confirm fresh frames and output rows; an absent RAW archive is not an ingestion failure. Calibrator may need time to collect its initial live sample window.
+3. On NATS firmware, confirm fresh frames and output rows; an absent RAW archive is not an ingestion failure. Calibrator may need time to collect its initial live sample window. If no frames arrive at all and the Core journal logs `[natsSource] no frames received`, compare your firmware with [sensor data by firmware](/core/firmware-updates/compatibility/#sensor-data-by-firmware). EmmcWear firmware `4b7552d5` publishes on `rawmem` subjects, which older Core builds do not read. On newer builds, the card's **Publishing prefixes** row marks any prefix Core does not read, and **Subject prefix** lets you pin it.
4. On RAW firmware, check the hot directory and firmware journal:
```bash