Merge main into archy-hwconfig — reconcile probe/dedup/name work

Both sides independently fixed the serial-alias dedup and the ESP32
boot-reset races; kept the branch's defer-to-auto-detect for unpinned
preferred paths (single probe pass per cycle) on top of main's
advert-name threading, Reticulum name propagation and radio-first
routing. Modal keeps main's 'Set Recommended' naming + probe progress
bar alongside the branch's in-app firmware flasher step.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

core/Cargo.lock

@@ -84,6 +84,15 @@ version = "1.0.100"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "a23eb6b1614318a8071c9b2521f36b424b2c83db5eb3a0fead4a6c0809af6e61"
 
+[[package]]
+name = "arbitrary"
+version = "1.4.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c3d036a3c4ab069c7b410a2ce876bd74808d2d0888a82667669f8e783a898bf1"
+dependencies = [
+ "derive_arbitrary",
+]
+
 [[package]]
 name = "arc-swap"
 version = "1.9.1"
@@ -161,6 +170,7 @@ dependencies = [
  "uuid",
  "zbase32",
  "zeroize",
+ "zip",
 ]
 
 [[package]]
@@ -1175,6 +1185,17 @@ version = "0.5.8"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "7cd812cc2bc1d69d4764bd80df88b4317eaef9e773c75226407d9bc0876b211c"
 
+[[package]]
+name = "derive_arbitrary"
+version = "1.4.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1e567bd82dcff979e4b03460c307b3cdc9e96fde3d73bed1496d2bc75d9dd62a"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn 2.0.114",
+]
+
 [[package]]
 name = "derive_builder"
 version = "0.20.2"
@@ -6895,8 +6916,37 @@ dependencies = [
  "syn 2.0.114",
 ]
 
+[[package]]
+name = "zip"
+version = "2.4.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "fabe6324e908f85a1c52063ce7aa26b68dcb7eb6dbc83a2d148403c9bc3eba50"
+dependencies = [
+ "arbitrary",
+ "crc32fast",
+ "crossbeam-utils",
+ "displaydoc",
+ "flate2",
+ "indexmap",
+ "memchr",
+ "thiserror 2.0.18",
+ "zopfli",
+]
+
 [[package]]
 name = "zmij"
 version = "1.0.16"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "dfcd145825aace48cff44a8844de64bf75feec3080e0aa5cdbde72961ae51a65"
+
+[[package]]
+name = "zopfli"
+version = "0.8.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f05cd8797d63865425ff89b5c4a48804f35ba0ce8d125800027ad6017d2b5249"
+dependencies = [
+ "bumpalo",
+ "crc32fast",
+ "log",
+ "simd-adler32",
+]

core/archipelago/Cargo.toml

@@ -108,6 +108,10 @@ bytes = "1"
 # Mesh networking (Meshcore serial protocol over USB LoRa radios)
 serial2-tokio = "0.1"
 
+# LoRa radio firmware flashing: Meshtastic ships per-board images inside a
+# per-platform release zip (see mesh/flash.rs).
+zip = { version = "2", default-features = false, features = ["deflate"] }
+
 # Double Ratchet key derivation (Phase 3: encrypted mesh messaging)
 hkdf = "0.12.4"
 

core/archipelago/src/api/rpc/dispatcher.rs

@@ -391,7 +391,12 @@ impl RpcHandler {
             // Mesh networking (Meshcore LoRa)
             "mesh.status" => self.handle_mesh_status().await,
             "mesh.probe-device" => self.handle_mesh_probe_device(params).await,
+            "mesh.flash-list-firmware" => self.handle_mesh_flash_list_firmware(params).await,
+            "mesh.flash-device" => self.handle_mesh_flash_device(params).await,
+            "mesh.flash-status" => self.handle_mesh_flash_status().await,
+            "mesh.flash-cancel" => self.handle_mesh_flash_cancel().await,
             "mesh.peers" => self.handle_mesh_peers().await,
+            "mesh.refresh" => self.handle_mesh_refresh().await,
             "mesh.messages" => self.handle_mesh_messages(params).await,
             "mesh.debug-dump" => self.handle_mesh_debug_dump().await,
             "mesh.send" => self.handle_mesh_send(params).await,

core/archipelago/src/api/rpc/mesh/flash.rs

@@ -0,0 +1,132 @@
+use super::super::RpcHandler;
+use crate::mesh;
+use crate::mesh::flash::{self, FlashBoard, FlashJobStatus};
+use crate::mesh::types::DeviceType;
+use anyhow::Result;
+
+fn parse_family(s: &str) -> Result<DeviceType> {
+    match s.trim().to_lowercase().as_str() {
+        "meshcore" => Ok(DeviceType::Meshcore),
+        "meshtastic" => Ok(DeviceType::Meshtastic),
+        "reticulum" | "rnode" => Ok(DeviceType::Reticulum),
+        other => anyhow::bail!("Unknown firmware family: {other} (expected meshcore|meshtastic|reticulum)"),
+    }
+}
+
+fn parse_board(s: &str) -> Result<FlashBoard> {
+    match s.trim().to_lowercase().as_str() {
+        "heltec-v3" | "heltec_v3" | "heltecv3" => Ok(FlashBoard::HeltecV3),
+        "heltec-v4" | "heltec_v4" | "heltecv4" => Ok(FlashBoard::HeltecV4),
+        other => anyhow::bail!("Unknown board: {other} (expected heltec-v3|heltec-v4)"),
+    }
+}
+
+impl RpcHandler {
+    /// mesh.flash-list-firmware — resolve the available firmware version(s)
+    /// for a given family. v1 only ever surfaces "latest".
+    pub(in crate::api::rpc) async fn handle_mesh_flash_list_firmware(
+        &self,
+        params: Option<serde_json::Value>,
+    ) -> Result<serde_json::Value> {
+        let family = params
+            .as_ref()
+            .and_then(|p| p.get("family"))
+            .and_then(|v| v.as_str())
+            .ok_or_else(|| anyhow::anyhow!("Missing family"))?;
+        let family = parse_family(family)?;
+        let versions = flash::list_firmware(family).await?;
+        Ok(serde_json::json!({ "versions": versions }))
+    }
+
+    /// mesh.flash-device — erase and reflash a detected LoRa radio with the
+    /// latest firmware for the given family, defaulting to a full chip
+    /// erase before write. `board` is optional: if the port's USB vid:pid
+    /// unambiguously resolves to a known board, that's used; otherwise the
+    /// caller must supply it explicitly (see `flash::resolve_flash_board`'s
+    /// doc comment on why we refuse to guess).
+    pub(in crate::api::rpc) async fn handle_mesh_flash_device(
+        &self,
+        params: Option<serde_json::Value>,
+    ) -> Result<serde_json::Value> {
+        let path = params
+            .as_ref()
+            .and_then(|p| p.get("path"))
+            .and_then(|v| v.as_str())
+            .ok_or_else(|| anyhow::anyhow!("Missing path"))?
+            .to_string();
+        let family = params
+            .as_ref()
+            .and_then(|p| p.get("family"))
+            .and_then(|v| v.as_str())
+            .ok_or_else(|| anyhow::anyhow!("Missing family"))?;
+        let family = parse_family(family)?;
+
+        let detected = mesh::detect_devices().await;
+        anyhow::ensure!(
+            detected.iter().any(|d| d == &path),
+            "{path} is not a detected mesh-radio candidate port"
+        );
+
+        let board = match params
+            .as_ref()
+            .and_then(|p| p.get("board"))
+            .and_then(|v| v.as_str())
+        {
+            Some(explicit) => parse_board(explicit)?,
+            None => {
+                let info = mesh::detect_devices_info()
+                    .await
+                    .into_iter()
+                    .find(|d| d.path == path);
+                info.as_ref()
+                    .and_then(flash::resolve_flash_board)
+                    .ok_or_else(|| {
+                        anyhow::anyhow!(
+                            "Could not auto-detect the board on {path} — specify board explicitly"
+                        )
+                    })?
+            }
+        };
+
+        flash::start_flash_job(
+            &self.flash_job,
+            &self.mesh_service_arc(),
+            self.config.data_dir.clone(),
+            path,
+            board,
+            family,
+        )
+        .await?;
+
+        Ok(serde_json::json!({ "started": true }))
+    }
+
+    /// mesh.flash-status — poll the current (or most recent) flash job.
+    pub(in crate::api::rpc) async fn handle_mesh_flash_status(&self) -> Result<serde_json::Value> {
+        let job = self.flash_job.read().await;
+        match job.as_ref() {
+            Some(j) => {
+                let status: FlashJobStatus = j.snapshot().await;
+                let mut value = serde_json::to_value(&status)?;
+                if let Some(obj) = value.as_object_mut() {
+                    obj.insert("active".into(), (!status.done).into());
+                }
+                Ok(value)
+            }
+            None => Ok(serde_json::json!({ "active": false })),
+        }
+    }
+
+    /// mesh.flash-cancel — best-effort; only honored before erase/write has
+    /// started (see `FlashJob::cancel`'s doc comment).
+    pub(in crate::api::rpc) async fn handle_mesh_flash_cancel(&self) -> Result<serde_json::Value> {
+        let job = self.flash_job.read().await;
+        match job.as_ref() {
+            Some(j) => {
+                j.cancel().await?;
+                Ok(serde_json::json!({ "cancelled": true }))
+            }
+            None => anyhow::bail!("No flash job in progress"),
+        }
+    }
+}

core/archipelago/src/api/rpc/mesh/messaging.rs

@@ -1,6 +1,7 @@
 use super::super::RpcHandler;
 use crate::mesh;
 use anyhow::Result;
+use std::sync::Arc;
 use tracing::info;
 
 impl RpcHandler {
@@ -131,7 +132,14 @@ impl RpcHandler {
             config.broadcast_identity = broadcast;
         }
         if let Some(name) = params.get("advert_name").and_then(|v| v.as_str()) {
-            config.advert_name = Some(name.to_string());
+            // Empty clears the custom mesh name (falls back to the server
+            // name) — without this, a name could be set but never unset.
+            let trimmed = name.trim();
+            config.advert_name = if trimmed.is_empty() {
+                None
+            } else {
+                Some(trimmed.to_string())
+            };
         }
         if let Some(announce) = params
             .get("announce_block_headers")
@@ -202,10 +210,24 @@ impl RpcHandler {
 
         mesh::save_config(&self.config.data_dir, &config).await?;
 
-        // If we have a running service, update its config
-        let mut service = self.mesh_service.write().await;
-        if let Some(svc) = service.as_mut() {
-            svc.configure(config.clone()).await?;
+        // Apply to the running service in the background: configure() may
+        // stop+start the listener (config changes now restart the session so
+        // they actually take effect), and that can take seconds when the old
+        // session is mid-probe. Holding the service write-lock for that long
+        // inside this handler stalled every concurrent mesh.status/mesh.peers
+        // poll behind it — the UI froze and nginx surfaced 502s. The config is
+        // already persisted above; the UI observes progress via mesh.status.
+        {
+            let service_arc = Arc::clone(&self.mesh_service);
+            let config_for_apply = config.clone();
+            tokio::spawn(async move {
+                let mut service = service_arc.write().await;
+                if let Some(svc) = service.as_mut() {
+                    if let Err(e) = svc.configure(config_for_apply).await {
+                        tracing::error!("Applying mesh config to running service failed: {e:#}");
+                    }
+                }
+            });
         }
 
         info!("Mesh config updated");

core/archipelago/src/api/rpc/mesh/mod.rs

@@ -1,5 +1,6 @@
 mod assistant;
 mod bitcoin_ops;
+mod flash;
 mod messaging;
 mod safety;
 mod status;

core/archipelago/src/api/rpc/mesh/status.rs

@@ -101,15 +101,61 @@ impl RpcHandler {
             detected.iter().any(|d| d == &path),
             "{path} is not a detected mesh-radio candidate port"
         );
-        let service = self.mesh_service.read().await;
-        let probe = match service.as_ref() {
-            Some(svc) => svc.probe_device(&path).await?,
-            // No mesh service yet (radio never enabled) — probe directly.
-            None => mesh::listener::probe_device(&path).await?,
-        };
+        // Refuse to probe while a firmware flash is in flight. Confirmed
+        // live 2026-07-23: esptool ("multiple access on port?") and
+        // rnodeconf (OSError Errno 71 Protocol error on an RTS ioctl) both
+        // failed with symptoms consistent with a second process holding the
+        // same serial fd — the flash subprocess runs for minutes outside
+        // our own async runtime, so nothing previously stopped a concurrent
+        // `mesh.probe-device` call (e.g. the hot-swap modal's own re-probe)
+        // from opening the identical port at the same time and corrupting
+        // both operations' handshakes.
+        if let Some(job) = self.flash_job.read().await.as_ref() {
+            anyhow::ensure!(
+                job.snapshot().await.done,
+                "A firmware flash is in progress — refusing to probe the serial port until it finishes"
+            );
+        }
+        // Only hold the mesh_service lock long enough for the quick
+        // active-path guard check — NEVER across the actual probe, which
+        // can take 15-60s across its internal collision retries. Confirmed
+        // live 2026-07-23: holding this read lock for the full probe starved
+        // a concurrent firmware-flash job's MeshService::stop() (which needs
+        // the write lock) well past its own bounded timeout, surfacing as
+        // "Mesh listener did not release the serial port" even though
+        // stop() itself was fast.
+        {
+            let service = self.mesh_service.read().await;
+            if let Some(svc) = service.as_ref() {
+                svc.ensure_probe_allowed(&path).await?;
+            }
+        }
+        let probe = mesh::listener::probe_device(&path).await?;
         Ok(serde_json::to_value(probe)?)
     }
 
+    /// mesh.refresh — Actively refresh discovery state: re-query the radio's
+    /// contact table and re-announce ourselves so quiet-but-alive neighbours
+    /// answer. This is what the UI's Refresh button calls — before it existed
+    /// the button only re-read server caches and never touched the radio.
+    pub(in crate::api::rpc) async fn handle_mesh_refresh(&self) -> Result<serde_json::Value> {
+        let service = self.mesh_service.read().await;
+        let Some(svc) = service.as_ref() else {
+            return Ok(serde_json::json!({ "refreshed": false, "device_connected": false }));
+        };
+        let status = svc.status().await;
+        if status.device_connected {
+            let state = svc.shared_state();
+            let _ = state
+                .send_cmd(crate::mesh::listener::MeshCommand::RefreshContacts)
+                .await;
+        }
+        Ok(serde_json::json!({
+            "refreshed": status.device_connected,
+            "device_connected": status.device_connected,
+        }))
+    }
+
     /// mesh.peers — List discovered mesh peers.
     pub(in crate::api::rpc) async fn handle_mesh_peers(&self) -> Result<serde_json::Value> {
         let service = self.mesh_service.read().await;

core/archipelago/src/api/rpc/mod.rs

@@ -89,6 +89,9 @@ pub struct RpcHandler {
     endpoint_rate_limiter: EndpointRateLimiter,
     response_cache: ResponseCache,
     mesh_service: Arc<tokio::sync::RwLock<Option<crate::mesh::MeshService>>>,
+    /// LoRa radio firmware-flash job state, sibling to `mesh_service` — one
+    /// job at a time, since flashing needs exclusive access to the port.
+    flash_job: crate::mesh::flash::FlashJobHandle,
     transport_router: Arc<tokio::sync::RwLock<Option<Arc<crate::transport::TransportRouter>>>>,
     /// Shared content-addressed blob store. Set by ApiHandler after construction
     /// so mesh.send-content / mesh.fetch-content RPCs can reach it without a
@@ -160,6 +163,7 @@ impl RpcHandler {
             endpoint_rate_limiter,
             response_cache: ResponseCache::new(5),
             mesh_service: Arc::new(tokio::sync::RwLock::new(None)),
+            flash_job: crate::mesh::flash::new_job_handle(),
             transport_router: Arc::new(tokio::sync::RwLock::new(None)),
             blob_store: Arc::new(tokio::sync::RwLock::new(None)),
             self_pubkey_hex: Arc::new(tokio::sync::RwLock::new(None)),

core/archipelago/src/api/rpc/system/handlers.rs

@@ -61,6 +61,28 @@ impl RpcHandler {
 
         info!("Server name updated to: {}", name);
 
+        // Propagate to the mesh: the listener advertises the server name (when
+        // no explicit mesh advert_name overrides it), but it was only read at
+        // process startup — a rename never reached the radio/RNS until the
+        // next full restart. Push it into the service and bounce the listener
+        // in the background (the restart re-probes the radio, which can take
+        // seconds — don't block the rename response on it).
+        {
+            let mesh_arc = self.mesh_service_arc();
+            let name_for_mesh = name.clone();
+            tokio::spawn(async move {
+                let mut guard = mesh_arc.write().await;
+                if let Some(svc) = guard.as_mut() {
+                    svc.set_server_name(Some(name_for_mesh));
+                    if svc.config().advert_name.is_none() {
+                        if let Err(e) = svc.restart_listener_if_running().await {
+                            warn!("Mesh listener restart after rename failed: {}", e);
+                        }
+                    }
+                }
+            });
+        }
+
         // Push the new name to federation peers in background
         let data_dir = self.config.data_dir.clone();
         let state_manager = self.state_manager.clone();

core/archipelago/src/mesh/flash.rs

@@ -0,0 +1,975 @@
+// WIP mesh/transport protocol — suppress dead code warnings
+#![allow(dead_code)]
+//! Firmware flashing for LoRa mesh radios — Heltec V3/V4 in v1, across all
+//! three firmware families the mesh module already knows how to detect (see
+//! `mesh::types::DeviceType`). Firmware is always fetched from upstream at
+//! flash time (never bundled/pinned in the repo), and every flash defaults
+//! to a full chip erase before write.
+//!
+//! MeshCore and Meshtastic are flashed the same way: download a released
+//! image, `esptool erase_flash`, then `esptool write_flash 0x0 <image>`.
+//! Reticulum/RNode is different: `archy-rnodeconf --autoinstall` owns the
+//! whole fetch+erase+flash+EEPROM-bootstrap sequence itself (confirmed live
+//! via `archy-rnodeconf --help` — there is no raw esptool path exposed for
+//! this family, so we deliberately don't resolve a firmware URL ourselves
+//! for Reticulum; rnodeconf already knows how).
+
+use super::serial::DetectedDeviceInfo;
+use super::types::DeviceType;
+use super::MeshService;
+use anyhow::{Context, Result};
+use regex::Regex;
+use serde::Serialize;
+use std::path::{Path, PathBuf};
+use std::process::Stdio;
+use std::sync::{Arc, OnceLock};
+use tokio::io::{AsyncBufReadExt, AsyncWriteExt, BufReader};
+use tokio::process::Command;
+use tokio::sync::RwLock;
+use tracing::{info, warn};
+
+/// Boards supported for v1. Both are ESP32-S3 (a single `--chip esp32s3`
+/// esptool target covers both), but ship different USB identities and
+/// different per-board firmware assets upstream.
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize)]
+#[serde(rename_all = "lowercase")]
+pub enum FlashBoard {
+    HeltecV3,
+    HeltecV4,
+}
+
+impl FlashBoard {
+    /// Meshtastic's board id (matches the release manifest's `board` field
+    /// and its per-board asset naming, e.g. `firmware-heltec-v3-<ver>.factory.bin`).
+    fn meshtastic_id(self) -> &'static str {
+        match self {
+            Self::HeltecV3 => "heltec-v3",
+            Self::HeltecV4 => "heltec-v4",
+        }
+    }
+}
+
+/// Map a detected USB vid:pid to a known flashable board, using the same
+/// table as `image-recipe/configs/99-mesh-radio.rules`. CP2102 (10c4:ea60)
+/// is confirmed there as Heltec V3's USB-UART bridge chip, and is safe to
+/// auto-match since that vid:pid is bridge-chip-specific.
+///
+/// Heltec V4 is NOT auto-matchable and deliberately has no entry here: it
+/// was confirmed live (real hardware, 2026-07-23) to use the ESP32-S3's
+/// built-in native-USB JTAG/serial peripheral, reporting vid:pid 303a:1001
+/// with product string "USB JTAG/serial debug unit" — that descriptor is
+/// baked into the chip's ROM and is IDENTICAL across every ESP32-S3 board
+/// with native USB enabled, not just Heltec V4. Adding `303a:1001 =>
+/// HeltecV4` here would silently misidentify any other native-USB ESP32-S3
+/// board (a T3-S3, a bare devkit, etc.) as a V4 and risk writing the wrong
+/// board's image. Callers (the RPC layer / frontend) must let the user pick
+/// the board manually whenever this returns `None`.
+pub fn resolve_flash_board(info: &DetectedDeviceInfo) -> Option<FlashBoard> {
+    match (info.vid.as_deref(), info.pid.as_deref()) {
+        (Some("10c4"), Some("ea60")) => Some(FlashBoard::HeltecV3),
+        _ => None,
+    }
+}
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize)]
+#[serde(rename_all = "lowercase")]
+pub enum FlashStage {
+    Downloading,
+    Erasing,
+    Writing,
+    Autoinstalling,
+    Done,
+    Failed,
+}
+
+#[derive(Debug, Clone, Serialize)]
+pub struct FlashJobStatus {
+    pub board: FlashBoard,
+    pub family: DeviceType,
+    pub path: String,
+    pub stage: FlashStage,
+    pub percent: Option<u8>,
+    pub log_tail: Vec<String>,
+    pub done: bool,
+    pub error: Option<String>,
+}
+
+const LOG_TAIL_MAX: usize = 200;
+
+/// How long to wait after a successful flash before resuming the mesh
+/// listener, so the board finishes its own post-flash boot/reset before we
+/// start opening the port (which itself toggles DTR/RTS) again.
+const POST_FLASH_SETTLE_DELAY: std::time::Duration = std::time::Duration::from_secs(5);
+
+/// Absolute ceiling on a whole flash job (download + erase + write, or
+/// autoinstall), regardless of what it's doing internally. Last-resort
+/// safety net so a hang anywhere can't wedge the single-flash-job guard
+/// forever — generous enough to never trigger on a legitimately slow
+/// multi-hundred-MB transfer.
+const MAX_JOB_DURATION: std::time::Duration = std::time::Duration::from_secs(15 * 60);
+
+/// How long to wait for MeshService::stop() to release the serial port
+/// before giving up. Confirmed live 2026-07-23: the listener's own
+/// reconnect/multi-candidate-probe loop doesn't check its shutdown signal
+/// between candidates, so stop() can take a while (or, if the loop is
+/// wedged, never return) — 20s comfortably covers a normal handshake-probe
+/// cycle without leaving a flash request hanging indefinitely if the
+/// listener genuinely won't let go.
+const STOP_LISTENER_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(20);
+
+/// How long to keep retrying the port-free check before giving up.
+const PORT_FREE_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(10);
+
+/// Confirm nothing else has `path` open by actually opening (and immediately
+/// closing) it ourselves. Retries across the timeout since a just-stopped
+/// listener's fd can take a moment to actually release even after `stop()`
+/// returns (task abort is a request, not an instant guarantee the OS-level
+/// resource is gone yet).
+async fn wait_for_port_free(path: &str) -> Result<()> {
+    let deadline = tokio::time::Instant::now() + PORT_FREE_TIMEOUT;
+    let mut last_err = None;
+    loop {
+        match serial2_tokio::SerialPort::open(path, 115200) {
+            Ok(_) => return Ok(()),
+            Err(e) => last_err = Some(e),
+        }
+        if tokio::time::Instant::now() >= deadline {
+            break;
+        }
+        tokio::time::sleep(std::time::Duration::from_millis(500)).await;
+    }
+    Err(anyhow::anyhow!(
+        "{path} is still held open by something else after {}s (last error: {}) — refusing to start the flasher against a contended port",
+        PORT_FREE_TIMEOUT.as_secs(),
+        last_err.map(|e| e.to_string()).unwrap_or_default()
+    ))
+}
+
+/// Live state for the one flash job that can run at a time. A single global
+/// slot is sufficient because flashing needs exclusive serial access to the
+/// one port being flashed — there is no meaningful concept of two concurrent
+/// flash jobs on this node.
+pub struct FlashJob {
+    status: RwLock<FlashJobStatus>,
+    /// Set once the background task is spawned. Only used while `stage` is
+    /// still `Downloading` — an interrupted erase/write can leave the chip
+    /// in a worse state than either finished or unstarted, so cancellation
+    /// is refused once erase begins (see `cancel()`).
+    abort_handle: RwLock<Option<tokio::task::AbortHandle>>,
+}
+
+impl FlashJob {
+    fn new(board: FlashBoard, family: DeviceType, path: String) -> Arc<Self> {
+        Arc::new(Self {
+            abort_handle: RwLock::new(None),
+            status: RwLock::new(FlashJobStatus {
+                board,
+                family,
+                path,
+                stage: FlashStage::Downloading,
+                percent: None,
+                log_tail: Vec::new(),
+                done: false,
+                error: None,
+            }),
+        })
+    }
+
+    pub async fn snapshot(&self) -> FlashJobStatus {
+        self.status.read().await.clone()
+    }
+
+    async fn set_stage(&self, stage: FlashStage) {
+        let mut s = self.status.write().await;
+        s.stage = stage;
+        s.percent = None;
+    }
+
+    async fn set_percent(&self, percent: u8) {
+        self.status.write().await.percent = Some(percent.min(100));
+    }
+
+    async fn push_log(&self, line: impl Into<String>) {
+        let mut s = self.status.write().await;
+        s.log_tail.push(line.into());
+        let overflow = s.log_tail.len().saturating_sub(LOG_TAIL_MAX);
+        if overflow > 0 {
+            s.log_tail.drain(0..overflow);
+        }
+    }
+
+    async fn fail(&self, err: &anyhow::Error) {
+        let mut s = self.status.write().await;
+        s.stage = FlashStage::Failed;
+        s.error = Some(format!("{err:#}"));
+        s.done = true;
+    }
+
+    async fn finish(&self) {
+        let mut s = self.status.write().await;
+        s.stage = FlashStage::Done;
+        s.done = true;
+    }
+
+    /// Best-effort cancel: only honored before erase/write/autoinstall has
+    /// started (i.e. still in `Downloading`). Once a stage that touches the
+    /// chip begins, this refuses — interrupting an erase or write can leave
+    /// the flash in a state worse than either finished or unstarted.
+    pub async fn cancel(&self) -> Result<()> {
+        let mut s = self.status.write().await;
+        if s.done {
+            anyhow::bail!("Flash job already finished");
+        }
+        if s.stage != FlashStage::Downloading {
+            anyhow::bail!(
+                "Cannot cancel once {:?} has started — let it finish or fail on its own",
+                s.stage
+            );
+        }
+        if let Some(handle) = self.abort_handle.write().await.take() {
+            handle.abort();
+        }
+        s.stage = FlashStage::Failed;
+        s.error = Some("Cancelled by user".to_string());
+        s.done = true;
+        Ok(())
+    }
+}
+
+/// Shared handle held by `RpcHandler`, sibling to `mesh_service`.
+pub type FlashJobHandle = Arc<RwLock<Option<Arc<FlashJob>>>>;
+
+pub fn new_job_handle() -> FlashJobHandle {
+    Arc::new(RwLock::new(None))
+}
+
+fn firmware_cache_dir(data_dir: &Path) -> PathBuf {
+    data_dir.join("mesh").join("firmware-cache")
+}
+
+/// No blanket `.timeout()` here on purpose: reqwest's request timeout covers
+/// the *entire* request including streaming the response body, which would
+/// kill a legitimate large download partway through (Meshtastic's esp32s3
+/// zip is ~170MB) — not just a hung connection. `download_to_file` instead
+/// applies a per-chunk stall timeout, and metadata calls (small JSON
+/// responses) get their own short timeout at the call site.
+fn github_client() -> Result<reqwest::Client> {
+    reqwest::Client::builder()
+        .user_agent("archipelago-mesh-flash")
+        .connect_timeout(std::time::Duration::from_secs(10))
+        .build()
+        .context("Failed to build HTTP client")
+}
+
+/// Applied per-chunk while streaming a firmware download — if the transfer
+/// stalls (no bytes for this long) it's treated as a failure, but a slow
+/// download that's still making progress is never killed just for taking a
+/// while.
+const DOWNLOAD_STALL_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(30);
+
+/// Applied to metadata calls (GitHub release JSON) — these are small
+/// responses with no reason to ever take this long.
+const METADATA_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(20);
+
+/// Resolve what firmware is available for a board+family. v1 only ever
+/// offers "latest" — MeshCore/Meshtastic latest GitHub release, or, for
+/// Reticulum, "latest" meaning "whatever archy-rnodeconf --autoinstall
+/// resolves on its own" (it does its own version checking upstream).
+pub async fn list_firmware(family: DeviceType) -> Result<Vec<String>> {
+    match family {
+        DeviceType::Reticulum => Ok(vec!["latest".to_string()]),
+        DeviceType::Meshtastic => {
+            let client = github_client()?;
+            let release: GithubRelease = client
+                .get("https://api.github.com/repos/meshtastic/firmware/releases/latest")
+                .send()
+                .await
+                .context("Fetching Meshtastic release list")?
+                .error_for_status()
+                .context("Meshtastic releases API error")?
+                .json()
+                .await
+                .context("Parsing Meshtastic release JSON")?;
+            Ok(vec![release.tag_name])
+        }
+        DeviceType::Meshcore => {
+            let client = github_client()?;
+            let release: GithubRelease = client
+                .get("https://api.github.com/repos/meshcore-dev/MeshCore/releases/latest")
+                .send()
+                .await
+                .context("Fetching MeshCore release list")?
+                .error_for_status()
+                .context("MeshCore releases API error")?
+                .json()
+                .await
+                .context("Parsing MeshCore release JSON")?;
+            Ok(vec![release.tag_name])
+        }
+        DeviceType::Unknown => anyhow::bail!("Pick a firmware family before listing versions"),
+    }
+}
+
+#[derive(serde::Deserialize)]
+struct GithubAsset {
+    name: String,
+    browser_download_url: String,
+}
+
+#[derive(serde::Deserialize)]
+struct GithubRelease {
+    tag_name: String,
+    assets: Vec<GithubAsset>,
+}
+
+/// Start a flash job in the background. Returns as soon as the job has been
+/// registered and the listener released — callers poll `FlashJobHandle` via
+/// `mesh.flash-status` for progress. Only one job may be in flight at a time.
+pub async fn start_flash_job(
+    handle: &FlashJobHandle,
+    mesh_service: &Arc<RwLock<Option<MeshService>>>,
+    data_dir: PathBuf,
+    path: String,
+    board: FlashBoard,
+    family: DeviceType,
+) -> Result<()> {
+    {
+        let existing = handle.read().await;
+        if let Some(job) = existing.as_ref() {
+            if !job.snapshot().await.done {
+                anyhow::bail!("A firmware flash is already in progress on this node");
+            }
+        }
+    }
+
+    let job = FlashJob::new(board, family, path.clone());
+    *handle.write().await = Some(Arc::clone(&job));
+
+    let bg_job = Arc::clone(&job);
+    let bg_service = Arc::clone(mesh_service);
+    let task = tokio::spawn(async move {
+        // esptool/archy-rnodeconf need exclusive serial access — release
+        // the listener's hold on the port before touching it. This USED
+        // TO run synchronously in start_flash_job before the job was even
+        // spawned, blocking the RPC call itself on s.stop().await — a real
+        // 2026-07-23 incident: the mesh listener was mid a multi-candidate
+        // reconnect/probe sequence that doesn't check its shutdown signal
+        // between candidates, so stop() never returned. The HTTP request
+        // timed out client-side ("Operation failed"), while the job
+        // (already inserted into `handle`) was permanently wedged — nothing
+        // had been spawned yet to ever mark it done, so every later flash
+        // attempt failed with "already in progress" until a full restart.
+        // Now this runs inside the spawned task with its own bounded
+        // timeout, so the RPC call always returns immediately regardless,
+        // and a slow-to-stop listener fails the job cleanly instead of
+        // hanging everything downstream of it forever.
+        let stop_result = tokio::time::timeout(STOP_LISTENER_TIMEOUT, async {
+            let mut svc = bg_service.write().await;
+            if let Some(s) = svc.as_mut() {
+                s.stop().await;
+            }
+        })
+        .await;
+        if stop_result.is_err() {
+            let err = anyhow::anyhow!(
+                "Mesh listener did not release the serial port within {}s — it may still be mid a reconnect attempt. Try again once mesh.status shows the device idle, or restart the archipelago service if this persists.",
+                STOP_LISTENER_TIMEOUT.as_secs()
+            );
+            bg_job.push_log(format!("ERROR: {err:#}")).await;
+            bg_job.fail(&err).await;
+            return;
+        }
+
+        // Belt-and-suspenders port-free check. `stop()` above should have
+        // fully released the port, but esptool/rnodeconf run as external
+        // subprocesses for minutes outside our own async runtime — if
+        // ANYTHING else still has it open (a racing probe, a not-yet-dropped
+        // fd from an aborted task, anything we haven't anticipated), handing
+        // the port to the flasher anyway risks exactly the corruption
+        // confirmed live 2026-07-23: esptool's "device disconnected or
+        // multiple access on port?" and rnodeconf's raw `OSError: [Errno 71]
+        // Protocol error` on an RTS ioctl are both textbook two-openers-on-
+        // one-fd symptoms. Verify by actually opening it ourselves — cheap,
+        // and definitive — before ever starting the flasher.
+        if let Err(e) = wait_for_port_free(&path).await {
+            bg_job.push_log(format!("ERROR: {e:#}")).await;
+            bg_job.fail(&e).await;
+            return;
+        }
+
+        // Outer ceiling on top of run_flash's own internal timeouts —
+        // belt-and-suspenders so that no future hang (network, subprocess,
+        // anything) can ever wedge the single-flash-job guard permanently
+        // again the way a stuck download did on 2026-07-23 (every
+        // subsequent mesh.flash-device call failed with "already in
+        // progress" until the service was restarted). Generous enough that
+        // a legitimately slow multi-hundred-MB transfer still completes.
+        let result = match tokio::time::timeout(
+            MAX_JOB_DURATION,
+            run_flash(board, family, &data_dir, &path, &bg_job),
+        )
+        .await
+        {
+            Ok(inner) => inner,
+            Err(_) => Err(anyhow::anyhow!(
+                "Flash job exceeded the {}-minute ceiling — aborted",
+                MAX_JOB_DURATION.as_secs() / 60
+            )),
+        };
+        let succeeded = result.is_ok();
+
+        match &result {
+            Ok(()) => {
+                bg_job.push_log("Flash completed successfully".to_string()).await;
+                bg_job.finish().await;
+                info!(path = %path, board = ?board, family = %family, "LoRa firmware flash succeeded");
+            }
+            Err(e) => {
+                // {:#} (alternate Display) walks the full anyhow context
+                // chain — plain {} / %e only prints the outermost .context()
+                // message, which made a real 2026-07-23 esptool failure
+                // undiagnosable from journalctl alone (just "esptool
+                // erase_flash failed", no actual esptool stderr).
+                warn!(path = %path, error = %format!("{e:#}"), "LoRa firmware flash failed");
+                bg_job.push_log(format!("ERROR: {e:#}")).await;
+                bg_job.fail(e).await;
+            }
+        }
+
+        // The board's firmware may now differ from whatever was pinned
+        // before — clear the pin either way so a later reconnect's strict
+        // auto-detect order picks up reality instead of getting wedged
+        // trying the old protocol first.
+        if let Ok(mut config) = super::load_config(&data_dir).await {
+            config.device_kind = None;
+            if let Err(e) = super::save_config(&data_dir, &config).await {
+                warn!(error = %e, "Failed to clear device_kind pin after flash");
+            }
+        }
+
+        if !succeeded {
+            // Deliberately do NOT auto-restart the listener here. A failed
+            // flash means we can't vouch for the board's state — reopening
+            // the port immediately (esptool/rnodeconf's own reset sequence
+            // plus our open() toggling DTR/RTS again right after) risks
+            // hammering a marginal device with reconnect attempts. Confirmed
+            // live 2026-07-23: exactly this sequence left a real Heltec V3
+            // boot-looping for 5+ minutes after a failed flash. Leave mesh
+            // stopped; the user reconnects explicitly via the hot-swap
+            // modal/Mesh page once they've confirmed the board is alive.
+            warn!(
+                path = %path,
+                "Leaving mesh listener stopped after failed flash — reconnect manually once the board is confirmed responsive"
+            );
+            return;
+        }
+
+        // On success, give the board a moment to finish booting after the
+        // flash tool's own reset sequence before we start hammering it with
+        // connection attempts — same reasoning as above, just the
+        // lower-risk (successful-flash) side of it.
+        tokio::time::sleep(POST_FLASH_SETTLE_DELAY).await;
+
+        let mut svc = bg_service.write().await;
+        if let Some(s) = svc.as_mut() {
+            match super::load_config(&data_dir).await {
+                Ok(config) => {
+                    // Only resume if mesh is actually still enabled per the
+                    // CURRENT persisted config — confirmed live 2026-07-23:
+                    // unconditionally forcing a restart here, regardless of
+                    // `enabled`, overrode a user's own concurrent "disable
+                    // mesh" toggle and left the listener running while
+                    // config said disabled. That inconsistent state is what
+                    // made a later legitimate "Keep As Is" click (which
+                    // correctly tries to start on a false→true transition)
+                    // fail with "already running" — the listener had already
+                    // been force-started behind the config's back.
+                    let should_run = config.enabled;
+                    if let Err(e) = s.configure(config).await {
+                        warn!(error = %e, "Failed to resume mesh listener after flash");
+                    }
+                    if should_run {
+                        if let Err(e) = s.start() {
+                            warn!(error = %e, "Failed to restart mesh listener after flash");
+                        }
+                    }
+                }
+                Err(e) => warn!(error = %e, "Failed to load mesh config after flash"),
+            }
+        }
+    });
+    *job.abort_handle.write().await = Some(task.abort_handle());
+
+    Ok(())
+}
+
+async fn run_flash(
+    board: FlashBoard,
+    family: DeviceType,
+    data_dir: &Path,
+    path: &str,
+    job: &Arc<FlashJob>,
+) -> Result<()> {
+    match family {
+        DeviceType::Meshtastic | DeviceType::Meshcore => {
+            let image = fetch_esptool_image(board, family, data_dir, job).await?;
+            esptool_erase_and_write(path, &image, job).await
+        }
+        DeviceType::Reticulum => {
+            let lora_region = super::load_config(data_dir)
+                .await
+                .ok()
+                .and_then(|c| c.lora_region);
+            rnodeconf_autoinstall(path, board, lora_region.as_deref(), job).await
+        }
+        DeviceType::Unknown => anyhow::bail!("Pick a firmware family before flashing"),
+    }
+}
+
+// ─── MeshCore / Meshtastic: esptool ─────────────────────────────────────
+
+async fn fetch_esptool_image(
+    board: FlashBoard,
+    family: DeviceType,
+    data_dir: &Path,
+    job: &Arc<FlashJob>,
+) -> Result<PathBuf> {
+    let cache = firmware_cache_dir(data_dir);
+    tokio::fs::create_dir_all(&cache)
+        .await
+        .context("Creating firmware cache dir")?;
+    let client = github_client()?;
+
+    match family {
+        DeviceType::Meshtastic => fetch_meshtastic_image(&client, board, &cache, job).await,
+        DeviceType::Meshcore => fetch_meshcore_image(&client, board, &cache, job).await,
+        _ => anyhow::bail!("{family} is not flashed via esptool"),
+    }
+}
+
+async fn fetch_meshtastic_image(
+    client: &reqwest::Client,
+    board: FlashBoard,
+    cache: &Path,
+    job: &Arc<FlashJob>,
+) -> Result<PathBuf> {
+    let release: GithubRelease = client
+        .get("https://api.github.com/repos/meshtastic/firmware/releases/latest")
+        .timeout(METADATA_TIMEOUT)
+        .send()
+        .await
+        .context("Fetching Meshtastic release list")?
+        .error_for_status()
+        .context("Meshtastic releases API error")?
+        .json()
+        .await
+        .context("Parsing Meshtastic release JSON")?;
+
+    // Meshtastic bundles all esp32s3 boards' images inside one per-platform
+    // zip rather than shipping per-board top-level assets — both Heltec V3
+    // and V4 are esp32s3, so this is the right zip for both (confirmed live
+    // against v2.7.26.54e0d8d).
+    let zip_asset = release
+        .assets
+        .iter()
+        .find(|a| a.name.starts_with("firmware-esp32s3-") && a.name.ends_with(".zip"))
+        .ok_or_else(|| anyhow::anyhow!("No esp32s3 firmware zip in latest Meshtastic release"))?;
+
+    let version = zip_asset
+        .name
+        .strip_prefix("firmware-esp32s3-")
+        .and_then(|s| s.strip_suffix(".zip"))
+        .ok_or_else(|| anyhow::anyhow!("Unexpected Meshtastic asset name: {}", zip_asset.name))?
+        .to_string();
+
+    let zip_path = cache.join(&zip_asset.name);
+    if tokio::fs::metadata(&zip_path).await.is_err() {
+        download_to_file(client, &zip_asset.browser_download_url, &zip_path, job).await?;
+    } else {
+        job.push_log(format!("Using cached {}", zip_asset.name)).await;
+    }
+
+    // "*.factory.bin" is Meshtastic's full merged image (bootloader +
+    // partition table + app) meant to be written at offset 0x0 on a freshly
+    // erased chip — confirmed by inspecting the real zip's contents, as
+    // opposed to the plain "*.bin" OTA-update image which assumes an
+    // existing bootloader/partition table already on the chip.
+    let entry_name = format!(
+        "firmware-{}-{}.factory.bin",
+        board.meshtastic_id(),
+        version
+    );
+    let out_path = cache.join(&entry_name);
+    if tokio::fs::metadata(&out_path).await.is_ok() {
+        return Ok(out_path);
+    }
+
+    job.push_log(format!(
+        "Extracting {entry_name} from {}",
+        zip_asset.name
+    ))
+    .await;
+    let zip_path_owned = zip_path.clone();
+    let entry_name_owned = entry_name.clone();
+    let out_path_owned = out_path.clone();
+    tokio::task::spawn_blocking(move || -> Result<()> {
+        let file = std::fs::File::open(&zip_path_owned).context("Opening downloaded firmware zip")?;
+        let mut archive = zip::ZipArchive::new(file).context("Reading firmware zip")?;
+        let mut entry = archive
+            .by_name(&entry_name_owned)
+            .with_context(|| format!("{entry_name_owned} not found in firmware zip"))?;
+        let mut out =
+            std::fs::File::create(&out_path_owned).context("Creating extracted firmware file")?;
+        std::io::copy(&mut entry, &mut out).context("Extracting firmware image")?;
+        Ok(())
+    })
+    .await
+    .context("Firmware extraction task panicked")??;
+
+    Ok(out_path)
+}
+
+async fn fetch_meshcore_image(
+    client: &reqwest::Client,
+    board: FlashBoard,
+    cache: &Path,
+    job: &Arc<FlashJob>,
+) -> Result<PathBuf> {
+    let release: GithubRelease = client
+        .get("https://api.github.com/repos/meshcore-dev/MeshCore/releases/latest")
+        .timeout(METADATA_TIMEOUT)
+        .send()
+        .await
+        .context("Fetching MeshCore release list")?
+        .error_for_status()
+        .context("MeshCore releases API error")?
+        .json()
+        .await
+        .context("Parsing MeshCore release JSON")?;
+
+    // Upstream's casing differs between boards (Heltec_v3_... vs
+    // heltec_v4_...) — match case-insensitively on the exact per-board
+    // substring so V4 isn't accidentally matched by "heltec_v4_tft_..."
+    // variants (there's a "_tft_" in between, so a straight substring match
+    // on "heltec_v4_companion_radio_usb" is already safe).
+    let needle = match board {
+        FlashBoard::HeltecV3 => "heltec_v3_companion_radio_usb",
+        FlashBoard::HeltecV4 => "heltec_v4_companion_radio_usb",
+    };
+    let asset = release
+        .assets
+        .iter()
+        .find(|a| {
+            let lower = a.name.to_lowercase();
+            lower.contains(needle) && lower.ends_with("-merged.bin")
+        })
+        .ok_or_else(|| {
+            anyhow::anyhow!("No matching MeshCore image in release {}", release.tag_name)
+        })?;
+
+    let out_path = cache.join(&asset.name);
+    if tokio::fs::metadata(&out_path).await.is_ok() {
+        job.push_log(format!("Using cached {}", asset.name)).await;
+        return Ok(out_path);
+    }
+    download_to_file(client, &asset.browser_download_url, &out_path, job).await?;
+    Ok(out_path)
+}
+
+async fn download_to_file(
+    client: &reqwest::Client,
+    url: &str,
+    dest: &Path,
+    job: &Arc<FlashJob>,
+) -> Result<()> {
+    job.set_stage(FlashStage::Downloading).await;
+    // Bound only the wait for the response to *start* (headers) — NOT a
+    // request-level `.timeout()`, which would cap the whole body transfer
+    // again (the bug this replaced: a blanket 30s client timeout killed
+    // large downloads mid-stream). If the server never responds at all,
+    // this is what stops the job from hanging forever; the per-chunk stall
+    // timeout below is what guards the body once streaming starts. Without
+    // this, a server that accepts the TCP connection but never sends
+    // headers back hangs this call indefinitely — confirmed live
+    // 2026-07-23: a stuck `.send()` here wedged the single-flash-job guard
+    // for good, permanently blocking every subsequent flash attempt with
+    // "already in progress" until the service was restarted.
+    let resp = tokio::time::timeout(METADATA_TIMEOUT, client.get(url).send())
+        .await
+        .context("Firmware download server did not respond")?
+        .context("Starting firmware download")?
+        .error_for_status()
+        .context("Firmware download returned an error status")?;
+    let total = resp.content_length();
+    let tmp = dest.with_extension("part");
+    let mut file = tokio::fs::File::create(&tmp)
+        .await
+        .context("Creating firmware download file")?;
+    let mut stream = resp.bytes_stream();
+    let mut downloaded: u64 = 0;
+    use futures_util::StreamExt;
+    loop {
+        let next = tokio::time::timeout(DOWNLOAD_STALL_TIMEOUT, stream.next())
+            .await
+            .context("Firmware download stalled")?;
+        let Some(chunk) = next else { break };
+        let chunk = chunk.context("Reading firmware download stream")?;
+        file.write_all(&chunk)
+            .await
+            .context("Writing firmware download")?;
+        downloaded += chunk.len() as u64;
+        if let Some(total) = total {
+            if total > 0 {
+                job.set_percent(((downloaded.saturating_mul(100)) / total) as u8)
+                    .await;
+            }
+        }
+    }
+    file.flush().await.ok();
+    tokio::fs::rename(&tmp, dest)
+        .await
+        .context("Finalizing firmware download")?;
+    job.push_log(format!(
+        "Downloaded {} ({downloaded} bytes)",
+        dest.display()
+    ))
+    .await;
+    Ok(())
+}
+
+/// Both Heltec V3 and V4 are ESP32-S3 boards.
+const ESPTOOL_CHIP: &str = "esp32s3";
+
+/// esptool's auto-reset-into-bootloader handshake (toggling DTR/RTS in a
+/// specific timed pattern) is well-known to be flaky on some CP2102/CH340
+/// board+adapter combinations — esptool's own docs recommend retrying at a
+/// lower baud rate when this happens. Rather than fail the whole job on the
+/// first hiccup, retry once at a conservative baud before giving up.
+const ESPTOOL_FALLBACK_BAUD: &str = "115200";
+
+/// `write_flash --erase-all` erases the whole chip before writing, in one
+/// esptool invocation. This needs the esp32s3 stub flasher loaded (see
+/// esptool_global_args' doc comment) — without it, --erase-all hits the
+/// exact same ROM limitation a standalone `erase_flash` does ("ESP32-S3 ROM
+/// does not support function erase_flash", confirmed live 2026-07-23), since
+/// esptool's --erase-all is implemented as the same full-chip-erase command,
+/// not a per-sector loop.
+async fn esptool_erase_and_write(path: &str, image: &Path, job: &Arc<FlashJob>) -> Result<()> {
+    job.set_stage(FlashStage::Writing).await;
+    let image_str = image.to_string_lossy().to_string();
+    esptool_with_retry(
+        path,
+        &["write_flash", "--erase-all", "0x0", &image_str],
+        job,
+    )
+    .await
+    .context("esptool write_flash failed")?;
+
+    Ok(())
+}
+
+/// esptool's global flags (--chip/--port/--baud) MUST precede the subcommand
+/// token (erase_flash/write_flash/...) — confirmed live 2026-07-23:
+/// appending `--baud 115200` after the subcommand on the retry path
+/// produced "esptool: error: unrecognized arguments: --baud 115200" every
+/// time, so the fallback-baud retry never actually got a chance to run.
+/// Building global args separately from subcommand args keeps this correct
+/// by construction instead of relying on call-site ordering.
+///
+/// Normal stub-loader mode (no --no-stub) needs the esp32s3 stub flasher
+/// blob at /usr/lib/python3/dist-packages/esptool/targets/stub_flasher/
+/// stub_flasher_32s3.json — Debian's `esptool` package (4.7.0+dfsg-0.1)
+/// ships without it (stripped for DFSG compliance: the prebuilt blob has no
+/// buildable-from-source path Debian could verify), so scripts/self-update.sh
+/// fetches the exact same file from the matching upstream esptool release
+/// tag and installs it alongside the apt package (see the esptool install
+/// step there). --no-stub (talk directly to the ROM bootloader, skip the
+/// stub) was tried first and works for connecting, but the ROM bootloader
+/// doesn't implement a full-chip-erase opcode at all — only the stub does —
+/// so --no-stub broke our "always erase before write" default outright
+/// rather than just being slower. Restoring the real stub file is the
+/// correct fix, not routing around its absence.
+fn esptool_global_args<'a>(path: &'a str, baud: Option<&'a str>) -> Vec<&'a str> {
+    let mut args = vec!["--chip", ESPTOOL_CHIP, "--port", path];
+    if let Some(b) = baud {
+        args.push("--baud");
+        args.push(b);
+    }
+    args
+}
+
+async fn esptool_with_retry(path: &str, subcommand: &[&str], job: &Arc<FlashJob>) -> Result<()> {
+    let mut cmd = Command::new("esptool");
+    cmd.args(esptool_global_args(path, None));
+    cmd.args(subcommand);
+    match run_streamed(cmd, None, job).await {
+        Ok(()) => Ok(()),
+        Err(first_err) => {
+            job.push_log(format!(
+                "First attempt failed ({first_err:#}); retrying once at {ESPTOOL_FALLBACK_BAUD} baud"
+            ))
+            .await;
+            let mut retry = Command::new("esptool");
+            retry.args(esptool_global_args(path, Some(ESPTOOL_FALLBACK_BAUD)));
+            retry.args(subcommand);
+            run_streamed(retry, None, job)
+                .await
+                .context(format!("retry also failed (first attempt: {first_err:#})"))
+        }
+    }
+}
+
+// ─── Reticulum/RNode: archy-rnodeconf ───────────────────────────────────
+
+fn rnodeconf_bin() -> String {
+    std::env::var("ARCHY_RNODECONF_BIN")
+        .unwrap_or_else(|_| "/usr/local/bin/archy-rnodeconf".to_string())
+}
+
+/// `--autoinstall`'s "which board is this" step is interactive by design —
+/// confirmed live against a real Heltec V4 (2026-07-23): even with a board
+/// given on the command line, rnodeconf can't always tell V3 from V4 apart
+/// (their bootstrap-time USB identity is often generic, same root cause as
+/// `resolve_flash_board`'s doc comment), so it always asks. The full prompt
+/// sequence observed for a Heltec board that already has *some* RNode
+/// firmware installed (the common case — a truly blank chip likely skips
+/// straight to the same "Device Selection" menu):
+///   1. numbered device-type menu  → answer with the menu number
+///   2. "Hit enter to continue"    → answer with a blank line
+///   3. numbered band menu         → answer with the menu number
+///   4. "Is the above correct? [y/N]" → answer "y"
+/// Feeding all four answers up front (rather than watching stdout for each
+/// prompt text) works because the menu is always asked in this fixed order
+/// for every board that needs (re)provisioning — verified by driving it
+/// through an unprovisioned real V4 end-to-end (erase → flash → EEPROM
+/// bootstrap → "Device signature validated" on the next probe).
+fn rnodeconf_device_menu_number(board: FlashBoard) -> &'static str {
+    match board {
+        FlashBoard::HeltecV3 => "8",
+        FlashBoard::HeltecV4 => "9",
+    }
+}
+
+/// rnodeconf's band choice is a coarse RF-frontend bootstrap parameter
+/// (868/915/923 MHz), not the final operating frequency — that's still
+/// configured later via the daemon's interface config, same as today. This
+/// is a best-effort mapping from the node's persisted Meshtastic-style
+/// region code (see `mesh::meshtastic::region_name_to_code`) down to
+/// rnodeconf's 3-way menu; regions with no exact 868/923 match fall back to
+/// 915 MHz as the broadest-compatibility default.
+fn rnodeconf_band_menu_number(lora_region: Option<&str>) -> &'static str {
+    match lora_region.map(|s| s.trim().to_uppercase()) {
+        Some(r) if r.contains("868") => "1",
+        Some(r) if r.contains("923") => "3",
+        _ => "2",
+    }
+}
+
+/// `--autoinstall` fetches, erases, flashes, and bootstraps the EEPROM for
+/// a detected board as one atomic step (confirmed via `archy-rnodeconf
+/// --help` AND a real end-to-end flash on real hardware) — this is the
+/// RNode-side equivalent of our "always erase before write" default, since
+/// autoinstall doesn't try to preserve any existing on-device state.
+async fn rnodeconf_autoinstall(
+    path: &str,
+    board: FlashBoard,
+    lora_region: Option<&str>,
+    job: &Arc<FlashJob>,
+) -> Result<()> {
+    job.set_stage(FlashStage::Autoinstalling).await;
+    let bin = rnodeconf_bin();
+    let mut cmd = if Path::new(&bin).exists() {
+        Command::new(bin)
+    } else {
+        // Dev fallback if only a plain venv/system rnodeconf is on PATH.
+        Command::new("rnodeconf")
+    };
+    cmd.args(["--autoinstall", path]);
+    let stdin = format!(
+        "{}\n\n{}\ny\n",
+        rnodeconf_device_menu_number(board),
+        rnodeconf_band_menu_number(lora_region)
+    );
+    run_streamed(cmd, Some(stdin.into_bytes()), job)
+        .await
+        .context("archy-rnodeconf --autoinstall failed")
+}
+
+// ─── Subprocess streaming ────────────────────────────────────────────────
+
+fn percent_regex() -> &'static Regex {
+    static RE: OnceLock<Regex> = OnceLock::new();
+    RE.get_or_init(|| Regex::new(r"\((\d{1,3})\s*%\)").expect("valid regex"))
+}
+
+async fn run_streamed(mut cmd: Command, stdin: Option<Vec<u8>>, job: &Arc<FlashJob>) -> Result<()> {
+    cmd.stdout(Stdio::piped());
+    cmd.stderr(Stdio::piped());
+    if stdin.is_some() {
+        cmd.stdin(Stdio::piped());
+    }
+    // Deliberately NOT kill_on_drop: an interrupted erase/write can leave
+    // the chip in a worse state than either finished or unstarted (see the
+    // cancellation-safety note in mesh flashing docs). The job is expected
+    // to run to completion or fail on its own.
+    let mut child = cmd.spawn().context("Failed to start subprocess")?;
+
+    if let Some(bytes) = stdin {
+        if let Some(mut child_stdin) = child.stdin.take() {
+            child_stdin
+                .write_all(&bytes)
+                .await
+                .context("Writing to subprocess stdin")?;
+        }
+    }
+
+    let mut tasks = Vec::new();
+    if let Some(stdout) = child.stdout.take() {
+        let job = Arc::clone(job);
+        tasks.push(tokio::spawn(async move {
+            let mut lines = BufReader::new(stdout).lines();
+            while let Ok(Some(line)) = lines.next_line().await {
+                if let Some(cap) = percent_regex().captures(&line) {
+                    if let Ok(pct) = cap[1].parse::<u8>() {
+                        job.set_percent(pct).await;
+                    }
+                }
+                job.push_log(line).await;
+            }
+        }));
+    }
+    if let Some(stderr) = child.stderr.take() {
+        let job = Arc::clone(job);
+        tasks.push(tokio::spawn(async move {
+            let mut lines = BufReader::new(stderr).lines();
+            while let Ok(Some(line)) = lines.next_line().await {
+                job.push_log(line).await;
+            }
+        }));
+    }
+
+    let status = child.wait().await.context("Waiting for subprocess")?;
+    for t in tasks {
+        let _ = t.await;
+    }
+    if !status.success() {
+        // Exit status alone isn't diagnosable — the actual esptool/rnodeconf
+        // stderr (already captured into job.log_tail by the reader tasks
+        // above) is what actually explains a failure. Confirmed live
+        // 2026-07-23: a bare "Command exited with exit status: 1" told us
+        // nothing when esptool's real error was sitting in the log tail the
+        // whole time, only visible via the UI's live poll, not journald.
+        let tail: Vec<String> = job
+            .snapshot()
+            .await
+            .log_tail
+            .iter()
+            .rev()
+            .take(10)
+            .rev()
+            .cloned()
+            .collect();
+        anyhow::bail!("Command exited with {status}\n{}", tail.join("\n"));
+    }
+    Ok(())
+}

core/archipelago/src/mesh/listener/decode.rs

@@ -578,7 +578,7 @@ pub(super) async fn handle_identity_received(
         .insert(contact_id, shared_secret);
 
     // Update peer record
-    let peer = MeshPeer {
+    let mut peer = MeshPeer {
         contact_id,
         // .get(): a malformed DID shorter than the "did:key:" prefix must
         // not panic the listener on a radio-supplied string.
@@ -607,6 +607,24 @@ pub(super) async fn handle_identity_received(
     let is_new = {
         let mut peers = state.peers.write().await;
         let is_new = !peers.contains_key(&contact_id);
+        if let Some(existing) = peers.get(&contact_id) {
+            // This id is shared with the federation-seeded row for the same
+            // node (that's the point — identity adverts MERGE, not duplicate).
+            // The wholesale insert below must not stomp the federation row's
+            // real node name with our synthetic "Archy-…" placeholder — with
+            // Reticulum re-emitting identity adverts every announce tick,
+            // that renamed every federated contact once a minute. Same for a
+            // known position: keep it rather than nulling it out.
+            if !existing.advert_name.trim().is_empty()
+                && !existing.advert_name.starts_with("Archy-")
+            {
+                peer.advert_name = existing.advert_name.clone();
+            }
+            if peer.lat.is_none() {
+                peer.lat = existing.lat;
+                peer.lon = existing.lon;
+            }
+        }
         peers.insert(contact_id, peer.clone());
         is_new
     };

core/archipelago/src/mesh/listener/mod.rs

@@ -557,6 +557,10 @@ pub fn spawn_mesh_listener(
         let mut shutdown = shutdown;
         let mut cmd_rx = cmd_rx;
         let mut reconnect_delay = RECONNECT_DELAY_INIT;
+        // Mutable so a successful auto-detect can pin the firmware kind for
+        // the rest of this listener's lifetime — see the pin-on-first-success
+        // block below for why.
+        let mut device_kind = device_kind;
         // Backlog #12 hot-swap re-binding: each run_mesh_session call already
         // builds a fresh device struct (contacts/current_region/etc. all
         // start empty), so per-device session state is naturally isolated
@@ -628,6 +632,45 @@ pub fn spawn_mesh_listener(
                 }
             }
 
+            // Pin the firmware kind after the first successful auto-detect.
+            // Confirmed live 2026-07-23: with device_kind left unpinned (e.g.
+            // after clearing a stale pin), EVERY reconnect re-runs the full
+            // Reticulum→Meshcore→Meshtastic auto-detect cascade — each
+            // candidate past the first does its own open() with the DTR/RTS
+            // reset both boards need, so a device correctly identified as
+            // Meshtastic still gets reset once for the failed Meshcore
+            // attempt before Meshtastic's own open() resets it again. That
+            // doubled the reset count on every single reconnect indefinitely,
+            // not just during initial detection. Once auto-detect has
+            // identified the device this listener is actually talking to,
+            // there's no reason to keep guessing on subsequent reconnects —
+            // pin it, both in this task's own loop (takes effect
+            // immediately) and on disk (survives a service restart). A
+            // genuine hot-swap to different firmware is still handled: the
+            // setup modal's `mesh.probe-device` always re-probes unpinned,
+            // and the flash flow already clears this pin on its own.
+            if device_kind.is_none() {
+                let detected = state.status.read().await.device_type;
+                if detected != super::types::DeviceType::Unknown {
+                    device_kind = Some(detected);
+                    match super::load_config(&data_dir).await {
+                        Ok(mut cfg) if cfg.device_kind.is_none() => {
+                            cfg.device_kind = Some(detected);
+                            if let Err(e) = super::save_config(&data_dir, &cfg).await {
+                                warn!("Failed to persist auto-detected device_kind: {}", e);
+                            } else {
+                                info!(
+                                    kind = %detected,
+                                    "Pinned auto-detected firmware kind to avoid repeated multi-protocol resets on reconnect"
+                                );
+                            }
+                        }
+                        Ok(_) => {}
+                        Err(e) => warn!("Failed to load mesh config to persist device_kind: {}", e),
+                    }
+                }
+            }
+
             // Update status to disconnected. device_type/firmware_version are
             // reset too — they were previously left holding the LAST radio's
             // identity, so after a hot-swap the UI showed the old firmware

core/archipelago/src/mesh/listener/session.rs

@@ -270,6 +270,7 @@ async fn auto_detect_and_open(
     our_x25519_pubkey_hex: &str,
     device_kind: Option<DeviceType>,
     skip_path: Option<&str>,
+    advert_name: Option<&str>,
 ) -> Result<(String, MeshRadioDevice, DeviceInfo)> {
     let mut paths = super::super::serial::detect_serial_devices().await;
     // When falling back from a just-failed preferred path, don't probe that
@@ -287,6 +288,7 @@ async fn auto_detect_and_open(
             None => "No serial devices found in /dev".to_string(),
         });
     }
+    info!(candidates = ?paths, "Auto-detect candidate ports for this attempt");
     for path in &paths {
         debug!(path = %path, "Probing for mesh radio device");
         // Tried FIRST: `ReticulumLink::open()` gates its expensive daemon
@@ -304,6 +306,7 @@ async fn auto_detect_and_open(
                 data_dir,
                 Some(our_ed_pubkey_hex),
                 Some(our_x25519_pubkey_hex),
+                advert_name,
             )
             .await
             {
@@ -465,6 +468,7 @@ async fn open_preferred_path(
     our_ed_pubkey_hex: &str,
     our_x25519_pubkey_hex: &str,
     device_kind: Option<DeviceType>,
+    advert_name: Option<&str>,
 ) -> Result<(MeshRadioDevice, DeviceInfo)> {
     // Pinned: try only the configured firmware and surface its own error —
     // never fall through to (and inject probe bytes into) another firmware's
@@ -497,6 +501,7 @@ async fn open_preferred_path(
                     data_dir,
                     Some(our_ed_pubkey_hex),
                     Some(our_x25519_pubkey_hex),
+                    advert_name,
                 )
                 .await
                 .context("Could not open preferred path as Reticulum")?;
@@ -510,40 +515,21 @@ async fn open_preferred_path(
         };
     }
 
-    // Reticulum first — see the matching comment on auto_detect_and_open:
-    // its cheap probe_rnode gate fails in ~1s for non-RNode firmware, while
-    // trying Meshcore/Meshtastic first was observed leaving a real RNode
-    // board unresponsive by the time Reticulum's turn came.
-    match ReticulumLink::open(
-        path,
-        data_dir,
-        Some(our_ed_pubkey_hex),
-        Some(our_x25519_pubkey_hex),
-    )
-    .await
-    {
-        Ok(mut dev) => match dev.initialize().await {
-            Ok(info) => return Ok((MeshRadioDevice::Reticulum(dev), info)),
-            Err(e) => {
-                debug!(path = %path, error = %e, "Preferred path is not a working Reticulum RNode")
-            }
-        },
-        Err(e) => debug!(path = %path, error = %e, "Could not open preferred path as Reticulum"),
-    }
-    match MeshcoreDevice::open(path).await {
-        Ok(mut dev) => match dev.initialize().await {
-            Ok(info) => return Ok((MeshRadioDevice::Meshcore(dev), info)),
-            Err(e) => debug!(path = %path, error = %e, "Preferred path is not Meshcore"),
-        },
-        Err(e) => debug!(path = %path, error = %e, "Could not open preferred path as Meshcore"),
-    }
-    match MeshtasticDevice::open(path).await {
-        Ok(mut dev) => match dev.initialize().await {
-            Ok(info) => Ok((MeshRadioDevice::Meshtastic(dev), info)),
-            Err(e) => Err(e).context("Preferred path is not a working Meshtastic device"),
-        },
-        Err(e) => Err(e).context("Could not open preferred path as Meshtastic"),
-    }
+    // Unpinned: don't probe this path ourselves at all. Confirmed live
+    // 2026-07-23 — this function used to run its own Reticulum→Meshcore→
+    // Meshtastic sequence here, and the caller (run_mesh_session) falls
+    // back to `auto_detect_and_open` on any error, which scans every
+    // candidate path (this one included) with the exact same three-protocol
+    // sequence. With a single physical radio — the overwhelmingly common
+    // case — `path` here IS the one candidate `auto_detect_and_open` is
+    // about to try, so every unpinned reconnect was resetting the board via
+    // Reticulum/Meshcore/Meshtastic's DTR/RTS toggle TWICE: once here, once
+    // again moments later in auto-detect. Bailing immediately (no port
+    // access at all) means auto-detect's single pass is the only one that
+    // ever touches the port when nothing is pinned yet. (auto_detect_and_open
+    // carries the advert_name threading from main, so nothing is lost.)
+    let _ = advert_name;
+    anyhow::bail!("No device_kind pin — deferring to auto-detect for {path}")
 }
 
 /// Bring up a Reticulum daemon over plain TCP — no physical RNode, no
@@ -556,6 +542,7 @@ async fn open_reticulum_tcp(
     data_dir: &Path,
     our_ed_pubkey_hex: &str,
     our_x25519_pubkey_hex: &str,
+    advert_name: Option<&str>,
 ) -> Result<(String, MeshRadioDevice, DeviceInfo)> {
     let mut dev = match cfg {
         ReticulumTcpConfig::Server { bind } => ReticulumLink::open_tcp_server(
@@ -563,6 +550,7 @@ async fn open_reticulum_tcp(
             data_dir,
             Some(our_ed_pubkey_hex),
             Some(our_x25519_pubkey_hex),
+            advert_name,
         )
         .await
         .context("Could not open Reticulum TCP server interface")?,
@@ -571,6 +559,7 @@ async fn open_reticulum_tcp(
             data_dir,
             Some(our_ed_pubkey_hex),
             Some(our_x25519_pubkey_hex),
+            advert_name,
         )
         .await
         .context("Could not open Reticulum TCP client interface")?,
@@ -987,49 +976,93 @@ pub(super) async fn run_mesh_session(
     // auto-detect. TCP mode is additive/dev-only; it never changes behavior
     // for existing serial/RNode deployments where `reticulum_tcp` is None.
     //
+    // The name we present on the mesh: the operator's configured mesh name /
+    // server name, falling back to a DID fragment. Computed BEFORE the open
+    // sequence because Reticulum needs it at daemon-spawn time — the RNS
+    // announce carries it from the very first announce. (Meshcore/Meshtastic
+    // still receive it via set_advert_name after connect, below.)
+    let desired_advert_name: String = match server_name {
+        // Meshcore firmware limits advert names — truncate to 20 chars.
+        Some(name) => name.chars().take(20).collect(),
+        None => format!(
+            "Archy-{}",
+            our_did.chars().skip(8).take(8).collect::<String>()
+        ),
+    };
+
     // The whole open sequence runs under PORT_OPEN_LOCK so an RPC probe
     // can't interleave its own handshakes on the same tty (see the lock's
     // doc comment). Held only until the device is opened, then released.
-    let port_guard = PORT_OPEN_LOCK.lock().await;
-    let (device_path, mut device, device_info) = if let Some(tcp_cfg) = &reticulum_tcp {
-        open_reticulum_tcp(tcp_cfg, data_dir, our_ed_pubkey_hex, our_x25519_pubkey_hex).await?
-    } else if let Some(path) = preferred_path {
-        match open_preferred_path(
-            path,
-            data_dir,
-            our_ed_pubkey_hex,
-            our_x25519_pubkey_hex,
-            device_kind,
-        )
-        .await
-        {
-            Ok((dev, info)) => (path.to_string(), dev, info),
-            Err(e) => {
-                warn!(
-                    "Preferred path {} probe failed: {} — trying auto-detect",
-                    path, e
-                );
-                auto_detect_and_open(
-                    data_dir,
-                    our_ed_pubkey_hex,
-                    our_x25519_pubkey_hex,
-                    device_kind,
-                    Some(path),
-                )
-                .await?
+    //
+    // The sequence is raced against the shutdown signal: probes/handshakes
+    // can take 10s+, and without this a stop() issued mid-probe (config
+    // change, disable, rename) always burned the full listener-shutdown
+    // timeout and ended in a hard abort — observed live on archi-dev-box
+    // 2026-07-28. Dropping the open future mid-probe is safe: it holds no
+    // session state yet and the port guard/serial handle close with it.
+    let open_fut = async {
+        let port_guard = PORT_OPEN_LOCK.lock().await;
+        let result = if let Some(tcp_cfg) = &reticulum_tcp {
+            open_reticulum_tcp(
+                tcp_cfg,
+                data_dir,
+                our_ed_pubkey_hex,
+                our_x25519_pubkey_hex,
+                Some(&desired_advert_name),
+            )
+            .await
+        } else if let Some(path) = preferred_path {
+            match open_preferred_path(
+                path,
+                data_dir,
+                our_ed_pubkey_hex,
+                our_x25519_pubkey_hex,
+                device_kind,
+                Some(&desired_advert_name),
+            )
+            .await
+            {
+                Ok((dev, info)) => Ok((path.to_string(), dev, info)),
+                Err(e) => {
+                    warn!(
+                        "Preferred path {} probe failed: {} — trying auto-detect",
+                        path, e
+                    );
+                    auto_detect_and_open(
+                        data_dir,
+                        our_ed_pubkey_hex,
+                        our_x25519_pubkey_hex,
+                        device_kind,
+                        Some(path),
+                        Some(&desired_advert_name),
+                    )
+                    .await
+                }
             }
+        } else {
+            auto_detect_and_open(
+                data_dir,
+                our_ed_pubkey_hex,
+                our_x25519_pubkey_hex,
+                device_kind,
+                None,
+                Some(&desired_advert_name),
+            )
+            .await
+        };
+        drop(port_guard);
+        result
+    };
+    let (device_path, mut device, device_info) = tokio::select! {
+        res = open_fut => res?,
+        _ = shutdown.changed() => {
+            if *shutdown.borrow() {
+                info!("Shutdown requested during device open — ending session");
+                return Ok(());
+            }
+            anyhow::bail!("shutdown signal changed during device open");
         }
-    } else {
-        auto_detect_and_open(
-            data_dir,
-            our_ed_pubkey_hex,
-            our_x25519_pubkey_hex,
-            device_kind,
-            None,
-        )
-        .await?
     };
-    drop(port_guard);
 
     // Update status
     {
@@ -1176,19 +1209,14 @@ pub(super) async fn run_mesh_session(
         }
     }
 
-    // Set advert name to the server's human-readable name (e.g. "ThinkPad"),
-    // falling back to the DID fragment if no name is configured. Skipped in
-    // keep-as-is mode — the radio keeps the name it came with (already
-    // reflected in status from the connect handshake).
-    if manage_radio {
-        let advert_name = if let Some(name) = server_name {
-            // Meshcore firmware limits advert names — truncate to 20 chars
-            name.chars().take(20).collect::<String>()
-        } else {
-            let short_did = our_did.chars().skip(8).take(8).collect::<String>();
-            format!("Archy-{}", short_did)
-        };
-        if let Err(e) = device.set_advert_name(&advert_name).await {
+    // Set advert name to the configured mesh/server name (computed above).
+    // Skipped in keep-as-is mode for radio-held names — the radio keeps the
+    // name it came with (already reflected in status from the connect
+    // handshake). Reticulum is exempt from keep-as-is: its display name
+    // lives in OUR daemon (the RNode holds no name), so "keep as is" has
+    // nothing to preserve and an unnamed node would be anonymous on RNS.
+    if manage_radio || matches!(device, MeshRadioDevice::Reticulum(_)) {
+        if let Err(e) = device.set_advert_name(&desired_advert_name).await {
             warn!("Failed to set advert name: {}", e);
         } else {
             // Reflect the post-set name in MeshStatus too so the UI can filter
@@ -1196,7 +1224,7 @@ pub(super) async fn run_mesh_session(
             // still carries whatever pre-set name the firmware reported and the
             // self-filter never matches.
             let mut status = state.status.write().await;
-            status.self_advert_name = Some(advert_name.clone());
+            status.self_advert_name = Some(desired_advert_name.clone());
         }
     }
 
@@ -1489,6 +1517,15 @@ async fn handle_send_command(
             } else {
                 *consecutive_write_failures = 0;
             }
+            // The self-advert alone is a no-op for discovery on Meshtastic
+            // (heartbeat + time carry no identity) — the NodeInfo broadcast
+            // is what makes peers learn/refresh us. want_response=true so
+            // neighbours answer with their own NodeInfo: the user pressed
+            // Broadcast to be seen AND to see who's out there. No-op on
+            // Meshcore/Reticulum, whose self-advert already carries identity.
+            if let Err(e) = device.send_nodeinfo_advert(true).await {
+                warn!("Failed to send NodeInfo advert: {}", e);
+            }
         }
         MeshCommand::RebootRadio { seconds } => {
             if let Err(e) = device.reboot(seconds).await {

core/archipelago/src/mesh/mod.rs

@@ -8,6 +8,7 @@
 pub mod alerts;
 pub mod bitcoin_relay;
 pub mod crypto;
+pub mod flash;
 pub mod listener;
 pub mod meshtastic;
 pub mod message_types;
@@ -477,6 +478,31 @@ impl Default for MeshConfig {
     }
 }
 
+/// Whether a mesh config file has ever been written for this node — lets the
+/// boot path distinguish "operator explicitly disabled mesh" (file exists,
+/// enabled=false) from "never configured" (no file), which is the only case
+/// radio auto-enable should touch.
+pub fn config_file_exists(data_dir: &Path) -> bool {
+    data_dir.join(MESH_CONFIG_FILE).exists()
+}
+
+/// True when `new` differs from `old` in any field a running mesh session
+/// captured by value at spawn (device path/kind, advert name, region, PHY
+/// params, channel, manage_radio, TCP interface) — i.e. when applying `new`
+/// to a live service requires a listener restart. Fields the session reads
+/// live from shared state (broadcast flags, assistant settings, steganography
+/// mode, …) deliberately don't trigger a restart.
+fn session_config_changed(old: &MeshConfig, new: &MeshConfig) -> bool {
+    old.device_path != new.device_path
+        || old.device_kind != new.device_kind
+        || old.advert_name != new.advert_name
+        || old.lora_region != new.lora_region
+        || old.lora_radio_params != new.lora_radio_params
+        || old.channel_name != new.channel_name
+        || old.manage_radio != new.manage_radio
+        || old.reticulum_tcp != new.reticulum_tcp
+}
+
 pub async fn load_config(data_dir: &Path) -> Result<MeshConfig> {
     let path = data_dir.join(MESH_CONFIG_FILE);
     if !path.exists() {
@@ -742,10 +768,18 @@ impl MeshService {
         self.server_name = name;
     }
 
-    /// Start the background mesh listener.
+    /// Start the background mesh listener. Idempotent: if the listener is
+    /// already running, this is a harmless no-op rather than an error —
+    /// confirmed live 2026-07-23, a real race between the flash job's own
+    /// post-flash restart and a concurrent user "Keep As Is" click (both
+    /// legitimately trying to ensure the listener is running) surfaced this
+    /// as a user-facing "Mesh listener already running" RPC error. Ensuring
+    /// the listener is running is the intent every caller actually has;
+    /// whichever caller's start() happens to win the race, the other
+    /// finding it already satisfied is success, not failure.
     pub fn start(&mut self) -> Result<()> {
         if self.listener_handle.is_some() {
-            anyhow::bail!("Mesh listener already running");
+            return Ok(());
         }
 
         let (shutdown_tx, shutdown_rx) = watch::channel(false);
@@ -764,7 +798,11 @@ impl MeshService {
             self.our_ed_pubkey_hex.clone(),
             self.our_x25519_secret,
             self.our_x25519_pubkey_hex.clone(),
-            self.server_name.clone(),
+            // The mesh-page "Name on the mesh" (config.advert_name) wins over
+            // the server name — it existed as write-only config with no reader
+            // until this line, which is why renaming on the Mesh page never
+            // changed anything on the air.
+            self.config.advert_name.clone().or_else(|| self.server_name.clone()),
             self.config.lora_region.clone(),
             self.config.lora_radio_params,
             self.config.channel_name.clone(),
@@ -1055,13 +1093,12 @@ impl MeshService {
         self.state.peers.read().await.values().cloned().collect()
     }
 
-    /// Probe a serial port for a mesh radio without provisioning or keeping
-    /// it — powers the hot-swap "device detected" modal's current-details
-    /// view. Refuses to probe the port the live session currently occupies
-    /// (the probe would steal the serial port from under the session); a
+    /// Refuse to probe the port the live session currently occupies (the
+    /// probe would steal the serial port from under the session); a
     /// detected-but-not-connected port is fair game, accepting a benign race
-    /// with the reconnect loop (whichever loses just retries).
-    pub async fn probe_device(&self, path: &str) -> Result<listener::DeviceProbe> {
+    /// with the reconnect loop (whichever loses just retries). Split out from
+    /// the actual probe on purpose — see `probe_device`'s doc comment.
+    pub async fn ensure_probe_allowed(&self, path: &str) -> Result<()> {
         let status = self.state.status.read().await;
         if status.device_connected {
             if let Some(active) = status.device_path.as_deref() {
@@ -1076,8 +1113,7 @@ impl MeshService {
                 }
             }
         }
-        drop(status);
-        listener::probe_device(path).await
+        Ok(())
     }
 
     /// Get message history.
@@ -1124,7 +1160,32 @@ impl MeshService {
         let peer = peers
             .get(&contact_id)
             .ok_or_else(|| anyhow::anyhow!("Peer not found"))?;
-        let pubkey_hex = peer
+        // Cross-transport twin resolution: callers frequently hold the
+        // FEDERATION twin's contact_id (the UI's merged conversation row),
+        // whose pubkey_hex is the Archipelago ed25519 key — NOT a radio
+        // routing key. Sending a Reticulum resource with that prefix fails
+        // with "Unknown Reticulum prefix" (observed live 2026-07-28,
+        // image-over-LoRa to a merged contact). Route via the radio twin —
+        // same arch identity, radio-range id — whose pubkey_hex is the
+        // actual over-the-air routing key (RNS dest hash / firmware key).
+        let radio_peer = if peer.contact_id >= FEDERATION_CONTACT_ID_BASE {
+            peer.arch_pubkey_hex
+                .as_deref()
+                .and_then(|arch| {
+                    peers.values().find(|p| {
+                        p.contact_id < FEDERATION_CONTACT_ID_BASE
+                            && p.arch_pubkey_hex.as_deref() == Some(arch)
+                    })
+                })
+                .ok_or_else(|| {
+                    anyhow::anyhow!(
+                        "Peer is federation-only (no radio twin) — not reachable over the radio"
+                    )
+                })?
+        } else {
+            peer
+        };
+        let pubkey_hex = radio_peer
             .pubkey_hex
             .as_ref()
             .ok_or_else(|| anyhow::anyhow!("Peer has no public key"))?;
@@ -1262,12 +1323,44 @@ impl MeshService {
                 .map(|p| !p.reachable && p.arch_pubkey_hex.is_some())
                 .unwrap_or(false)
         };
+        // Transport policy: LoRa first when it can actually carry the message,
+        // then FIPS, then Tor. A federation-synthetic id (what the UI's merged
+        // conversation holds) used to ALWAYS take the federation path, even
+        // when the very same node was sitting one LoRa hop away — so chats
+        // between two radio-equipped nodes silently rode FIPS/Tor. If the
+        // federation contact has a REACHABLE radio twin (same archipelago
+        // identity, radio-range id) and the payload fits the radio, skip the
+        // federation branch: the fall-through LoRa path twin-resolves the
+        // routing key via peer_dest_prefix.
+        let device_connected = self.state.status.read().await.device_connected;
+        let radio_twin_reachable = is_federation_synthetic && !exceeds_lora && device_connected && {
+            let peers = self.state.peers.read().await;
+            peers
+                .get(&contact_id)
+                .and_then(|p| p.arch_pubkey_hex.clone())
+                .map(|arch| {
+                    peers.values().any(|p| {
+                        p.contact_id < FEDERATION_CONTACT_ID_BASE
+                            && p.reachable
+                            && p.arch_pubkey_hex.as_deref() == Some(arch.as_str())
+                    })
+                })
+                .unwrap_or(false)
+        };
         let mesh_only_mode = load_config(&self.data_dir)
             .await
             .ok()
             .and_then(|cfg| cfg.mesh_only_mode)
             .unwrap_or(false);
+        if radio_twin_reachable && !mesh_only_mode {
+            tracing::info!(
+                contact_id,
+                bytes = wire.len(),
+                "Radio-first routing: federation contact has a reachable radio twin — sending over LoRa"
+            );
+        }
         if !mesh_only_mode
+            && !radio_twin_reachable
             && (is_federation_synthetic || exceeds_lora || radio_federated_unreachable)
         {
             // Resolve the peer's pubkey/did. Prefer the live mesh peer table,
@@ -2091,6 +2184,7 @@ impl MeshService {
         save_config(&self.data_dir, &config).await?;
 
         let was_enabled = self.config.enabled;
+        let needs_session_restart = session_config_changed(&self.config, &config);
         self.config = config.clone();
 
         // Update the status to reflect new config
@@ -2115,11 +2209,31 @@ impl MeshService {
             status.firmware_version = None;
             status.self_node_id = None;
             status.peer_count = 0;
+        } else if config.enabled && was_enabled && needs_session_restart {
+            info!("Mesh session config changed — restarting listener to apply");
+            self.stop().await;
+            self.start()?;
         }
 
         Ok(())
     }
 
+    /// The service's current (last-applied) config.
+    pub fn config(&self) -> &MeshConfig {
+        &self.config
+    }
+
+    /// Restart the listener (if running) so it picks up out-of-band state a
+    /// spawn captured by value — currently the server name pushed by
+    /// `server.set-name`.
+    pub async fn restart_listener_if_running(&mut self) -> Result<()> {
+        if self.listener_handle.is_some() {
+            self.stop().await;
+            self.start()?;
+        }
+        Ok(())
+    }
+
     /// Get a reference to shared state (for RPC handlers).
     pub fn shared_state(&self) -> Arc<MeshState> {
         Arc::clone(&self.state)
@@ -2241,6 +2355,46 @@ async fn bitcoin_rpc_getblockheader_by_height(
 mod tests {
     use super::*;
 
+    #[test]
+    fn session_config_change_detection() {
+        let base = MeshConfig::default();
+
+        // Same config → no restart.
+        assert!(!session_config_changed(&base, &base.clone()));
+
+        // Every session-captured field individually triggers a restart.
+        let mut c = base.clone();
+        c.device_kind = Some(types::DeviceType::Reticulum);
+        assert!(session_config_changed(&base, &c));
+
+        let mut c = base.clone();
+        c.device_path = Some("/dev/ttyUSB0".into());
+        assert!(session_config_changed(&base, &c));
+
+        let mut c = base.clone();
+        c.advert_name = Some("RNode Shaza".into());
+        assert!(session_config_changed(&base, &c));
+
+        let mut c = base.clone();
+        c.manage_radio = !base.manage_radio;
+        assert!(session_config_changed(&base, &c));
+
+        let mut c = base.clone();
+        c.lora_region = Some("EU_868".into());
+        assert!(session_config_changed(&base, &c));
+
+        let mut c = base.clone();
+        c.channel_name = Some("private-net".into());
+        assert!(session_config_changed(&base, &c));
+
+        // Live-read fields must NOT force a session restart.
+        let mut c = base.clone();
+        c.broadcast_identity = !base.broadcast_identity;
+        c.announce_block_headers = !base.announce_block_headers;
+        c.assistant_enabled = !base.assistant_enabled;
+        assert!(!session_config_changed(&base, &c));
+    }
+
     fn mk_peer(contact_id: u32, name: &str, arch: Option<&str>, reachable: bool) -> MeshPeer {
         MeshPeer {
             contact_id,

core/archipelago/src/mesh/reticulum.rs

@@ -121,6 +121,7 @@ fn daemon_command(
     identity_key: &Path,
     archy_ed_pubkey_hex: Option<&str>,
     archy_x25519_pubkey_hex: Option<&str>,
+    display_name: Option<&str>,
 ) -> Command {
     let bin = std::env::var("ARCHY_RETICULUM_DAEMON_BIN")
         .unwrap_or_else(|_| "/usr/local/bin/archy-reticulum-daemon".to_string());
@@ -159,6 +160,15 @@ fn daemon_command(
             .arg("--archy-x25519-pubkey-hex")
             .arg(x);
     }
+    // The RNS-visible display name (what Sideband/NomadNet/other archy nodes
+    // show for us). Without this the daemon falls back to its argparse
+    // default and every archy node announces the same anonymous name.
+    if let Some(name) = display_name {
+        let name = name.trim();
+        if !name.is_empty() {
+            cmd.arg("--display-name").arg(name);
+        }
+    }
     // Run the daemon as its own process-group leader. The packaged binary is
     // a PyInstaller one-file bootloader that forks the real Python process;
     // making it a group leader lets shutdown signal the WHOLE group so the
@@ -207,6 +217,10 @@ struct ReticulumPeer {
     /// `bind_federation_twins`, which those two transports rely on instead).
     arch_pubkey_hex: Option<String>,
     reachable: bool,
+    /// Unix time of the last announce heard from this peer over the air.
+    /// In-memory only (a persisted value would be stale by definition) —
+    /// `0` after a restart until the peer re-announces.
+    last_advert_at: u64,
 }
 
 /// On-disk shape of `ReticulumPeer` — `[u8; 16]` can't be a JSON object key,
@@ -245,6 +259,11 @@ pub struct ReticulumLink {
     /// matching `resource_progress`/`resource_sent`/`resource_failed` events
     /// back to a log line; sends are fire-and-forget (see `send_resource`).
     resource_id_counter: u64,
+    /// Set when the daemon's RPC socket closes or its process exits. Once
+    /// true, `try_recv_frame` returns an error so the session loop tears
+    /// down and the outer reconnect loop respawns the daemon — without this
+    /// a dead daemon was invisible until the 30-minute RX-stall watchdog.
+    daemon_gone: bool,
 }
 
 impl ReticulumLink {
@@ -269,6 +288,7 @@ impl ReticulumLink {
         data_dir: &Path,
         our_ed_pubkey_hex: Option<&str>,
         our_x25519_pubkey_hex: Option<&str>,
+        display_name: Option<&str>,
     ) -> Result<Self> {
         probe_rnode(path)
             .await
@@ -278,6 +298,7 @@ impl ReticulumLink {
             data_dir,
             our_ed_pubkey_hex,
             our_x25519_pubkey_hex,
+            display_name,
         )
         .await
     }
@@ -290,6 +311,7 @@ impl ReticulumLink {
         data_dir: &Path,
         our_ed_pubkey_hex: Option<&str>,
         our_x25519_pubkey_hex: Option<&str>,
+        display_name: Option<&str>,
     ) -> Result<Self> {
         let host = bind.rsplit_once(':').map(|(h, _)| h).unwrap_or(bind);
         anyhow::ensure!(
@@ -302,6 +324,7 @@ impl ReticulumLink {
             data_dir,
             our_ed_pubkey_hex,
             our_x25519_pubkey_hex,
+            display_name,
         )
         .await
     }
@@ -313,6 +336,7 @@ impl ReticulumLink {
         data_dir: &Path,
         our_ed_pubkey_hex: Option<&str>,
         our_x25519_pubkey_hex: Option<&str>,
+        display_name: Option<&str>,
     ) -> Result<Self> {
         anyhow::ensure!(
             !targets.is_empty(),
@@ -323,6 +347,7 @@ impl ReticulumLink {
             data_dir,
             our_ed_pubkey_hex,
             our_x25519_pubkey_hex,
+            display_name,
         )
         .await
     }
@@ -332,6 +357,7 @@ impl ReticulumLink {
         data_dir: &Path,
         our_ed_pubkey_hex: Option<&str>,
         our_x25519_pubkey_hex: Option<&str>,
+        display_name: Option<&str>,
     ) -> Result<Self> {
         // Keep the RPC socket under the archipelago-owned data dir (not the
         // shared system temp dir) so its access is bounded by the same
@@ -379,6 +405,7 @@ impl ReticulumLink {
             &identity_key,
             our_ed_pubkey_hex,
             our_x25519_pubkey_hex,
+            display_name,
         );
         cmd.env("TMPDIR", &tmp_dir);
         let child = cmd
@@ -450,6 +477,7 @@ impl ReticulumLink {
             peers_file: runtime_dir.join("peers.json"),
             inbound: std::collections::VecDeque::new(),
             resource_id_counter: 0,
+            daemon_gone: false,
         };
         link.load_persisted_peers();
         Ok(link)
@@ -472,15 +500,25 @@ impl ReticulumLink {
             };
             let prefix: [u8; 6] = hash[..6].try_into().unwrap();
             self.prefix_to_hash.insert(prefix, hash);
+            // Heal names persisted by pre-2026-07-28 builds, which could
+            // store a raw `ARCHY:…` identity blob as the display name (seen
+            // live on archi-dev-box). Blob-only announces assert no name, so
+            // nothing would ever overwrite it — swap in the placeholder.
+            let display_name = if p.display_name.starts_with("ARCHY:") {
+                format!("Reticulum {}", hex::encode(&hash[..4]))
+            } else {
+                p.display_name
+            };
             self.peers.insert(
                 hash,
                 ReticulumPeer {
                     dest_hash: hash,
-                    display_name: p.display_name,
+                    display_name,
                     arch_pubkey_hex: p.arch_pubkey_hex,
                     // Reachability is a live property, not a persisted fact —
                     // start conservative and let the first real event refresh it.
                     reachable: false,
+                    last_advert_at: 0,
                 },
             );
         }
@@ -533,10 +571,12 @@ impl ReticulumLink {
     }
 
     pub async fn set_advert_name(&mut self, name: &str) -> Result<()> {
-        // The daemon's display_name is fixed at spawn time (CLI arg); changing
-        // it live would require an RPC verb we haven't added. Track locally so
-        // `advert_name()` reflects the caller's intent even though the
-        // RNS-visible name doesn't change until the daemon restarts.
+        // Live rename: the daemon's `set_name` verb updates the LXMF delivery
+        // destination's display_name and re-announces, so peers pick the new
+        // name up on their next announce receipt. Also tracked locally so
+        // `advert_name()` reflects it immediately.
+        self.send_rpc(serde_json::json!({"cmd": "set_name", "name": name}))
+            .await?;
         self.display_name = Some(name.to_string());
         Ok(())
     }
@@ -685,7 +725,7 @@ impl ReticulumLink {
             .map(|p| ParsedContact {
                 public_key_hex: hex::encode(p.dest_hash),
                 advert_name: p.display_name.clone(),
-                last_advert: 0,
+                last_advert: p.last_advert_at as u32,
                 // Deliberately not 1 ("friend"/meshcore type), so the
                 // meshcore-only auto-heal `reset_contact_path` loop in
                 // `refresh_contacts` (session.rs) skips these — RNS does its
@@ -718,6 +758,12 @@ impl ReticulumLink {
 
     pub async fn try_recv_frame(&mut self) -> Result<Option<InboundFrame>> {
         self.drain_events().await;
+        if self.daemon_gone {
+            // Surface the dead daemon as a hard error so run_mesh_session
+            // bails and the outer reconnect loop respawns it, instead of
+            // idling on an empty queue until the RX-stall watchdog fires.
+            anyhow::bail!("reticulum-daemon is gone (process exited or RPC socket closed)");
+        }
         Ok(self.inbound.pop_front())
     }
 
@@ -743,6 +789,15 @@ impl ReticulumLink {
     /// Drain any buffered daemon events (non-blocking) and translate them into
     /// peer-table updates / synthetic InboundFrames.
     async fn drain_events(&mut self) {
+        // A daemon that died without closing the socket cleanly (SIGKILL,
+        // OOM) leaves the socket readable-with-EOF or just silent — poll the
+        // child's exit status too so death is never mistaken for quiet.
+        if !self.daemon_gone {
+            if let Ok(Some(status)) = self.child.try_wait() {
+                warn!(%status, "reticulum-daemon process exited");
+                self.daemon_gone = true;
+            }
+        }
         loop {
             let mut line = String::new();
             let read =
@@ -750,10 +805,16 @@ impl ReticulumLink {
                     .await;
             let n = match read {
                 Ok(Ok(n)) => n,
-                _ => break, // timeout (no data) or read error — stop draining
+                Ok(Err(e)) => {
+                    warn!("Reticulum daemon RPC read failed: {}", e);
+                    self.daemon_gone = true;
+                    break;
+                }
+                Err(_) => break, // timeout — no data buffered
             };
             if n == 0 {
                 warn!("Reticulum daemon RPC connection closed");
+                self.daemon_gone = true;
                 break;
             }
             let Ok(ev) = serde_json::from_str::<Value>(line.trim()) else {
@@ -775,6 +836,23 @@ impl ReticulumLink {
                 };
                 let prefix: [u8; 6] = hash[..6].try_into().unwrap();
                 self.prefix_to_hash.insert(prefix, hash);
+                // Current daemons decode the LXMF announce app_data themselves
+                // and hand us clean fields: `display_name` (LXMF-standard
+                // msgpack name, Sideband-interoperable) and `archy_blob` (the
+                // `ARCHY:n:` identity string, carried as an extra msgpack list
+                // element stock clients ignore). The raw `app_data` text path
+                // below remains for announces from pre-upgrade archy nodes,
+                // whose app_data was EITHER the blob OR a bare-utf8 name.
+                let explicit_name = ev
+                    .get("display_name")
+                    .and_then(Value::as_str)
+                    .map(|s| s.trim().to_string())
+                    .filter(|s| !s.is_empty());
+                let explicit_blob = ev
+                    .get("archy_blob")
+                    .and_then(Value::as_str)
+                    .map(str::to_string)
+                    .filter(|s| !s.is_empty());
                 let app_data_text = ev
                     .get("app_data")
                     .and_then(Value::as_str)
@@ -794,12 +872,17 @@ impl ReticulumLink {
                 // now carry the same `arch_pubkey_hex`, instead of relying on
                 // `bind_federation_twins`'s advert_name matching, which never
                 // matches here — see `display_name` below.
-                let parsed_identity = app_data_text
+                let legacy_identity = app_data_text
                     .as_deref()
                     .and_then(protocol::parse_identity_broadcast);
-                let is_identity_blob = parsed_identity.is_some();
-                if is_identity_blob {
-                    let text = app_data_text.clone().unwrap();
+                let is_legacy_blob = legacy_identity.is_some();
+                let identity_blob_text = explicit_blob.or_else(|| {
+                    app_data_text.clone().filter(|_| is_legacy_blob)
+                });
+                let parsed_identity = identity_blob_text
+                    .as_deref()
+                    .and_then(protocol::parse_identity_broadcast);
+                if let Some(text) = identity_blob_text.as_deref().filter(|_| parsed_identity.is_some()) {
                     let mut data = Vec::with_capacity(7 + text.len());
                     data.push(0); // channel index — unused by the identity path
                     data.extend_from_slice(&prefix);
@@ -812,23 +895,31 @@ impl ReticulumLink {
                 }
                 let arch_pubkey_hex = parsed_identity.map(|(_did, ed_pubkey, _x25519)| ed_pubkey);
 
-                let display_name = app_data_text
-                    .filter(|_| !is_identity_blob)
-                    .unwrap_or_else(|| format!("Reticulum {}", hex::encode(&hash[..4])));
+                let announced_name =
+                    pick_announced_name(explicit_name, app_data_text, is_legacy_blob);
+                let now = std::time::SystemTime::now()
+                    .duration_since(std::time::UNIX_EPOCH)
+                    .unwrap_or_default()
+                    .as_secs();
                 self.peers
                     .entry(hash)
                     .and_modify(|p| {
-                        p.display_name = display_name.clone();
+                        if let Some(name) = announced_name.clone() {
+                            p.display_name = name;
+                        }
                         p.reachable = true;
+                        p.last_advert_at = now;
                         if arch_pubkey_hex.is_some() {
                             p.arch_pubkey_hex = arch_pubkey_hex.clone();
                         }
                     })
-                    .or_insert(ReticulumPeer {
+                    .or_insert_with(|| ReticulumPeer {
                         dest_hash: hash,
-                        display_name,
+                        display_name: announced_name
+                            .unwrap_or_else(|| format!("Reticulum {}", hex::encode(&hash[..4]))),
                         arch_pubkey_hex,
                         reachable: true,
+                        last_advert_at: now,
                     });
                 self.persist_peers();
             }
@@ -844,16 +935,24 @@ impl ReticulumLink {
                 // A peer that messages us without ever announcing still needs
                 // to survive a restart — give it a placeholder name (the real
                 // one, if any, arrives via a later "announce" and overwrites
-                // this) so its routing entry alone doesn't get lost.
-                if let std::collections::hash_map::Entry::Vacant(e) = self.peers.entry(source_hash)
-                {
-                    e.insert(ReticulumPeer {
-                        dest_hash: source_hash,
-                        display_name: format!("Reticulum {}", hex::encode(&source_hash[..4])),
-                        arch_pubkey_hex: None,
-                        reachable: true,
-                    });
-                    self.persist_peers();
+                // this) so its routing entry alone doesn't get lost. An
+                // existing entry is proof of life too: mark it reachable so a
+                // restart-restored (reachable=false) peer that DMs us doesn't
+                // stay red-dotted until its next announce.
+                match self.peers.entry(source_hash) {
+                    std::collections::hash_map::Entry::Vacant(e) => {
+                        e.insert(ReticulumPeer {
+                            dest_hash: source_hash,
+                            display_name: format!("Reticulum {}", hex::encode(&source_hash[..4])),
+                            arch_pubkey_hex: None,
+                            reachable: true,
+                            last_advert_at: 0,
+                        });
+                        self.persist_peers();
+                    }
+                    std::collections::hash_map::Entry::Occupied(mut e) => {
+                        e.get_mut().reachable = true;
+                    }
                 }
 
                 // A stock LXMF client (Sideband/NomadNet — not an archy peer)
@@ -932,15 +1031,20 @@ impl ReticulumLink {
                 };
                 let prefix: [u8; 6] = source_hash[..6].try_into().unwrap();
                 self.prefix_to_hash.insert(prefix, source_hash);
-                if let std::collections::hash_map::Entry::Vacant(e) = self.peers.entry(source_hash)
-                {
-                    e.insert(ReticulumPeer {
-                        dest_hash: source_hash,
-                        display_name: format!("Reticulum {}", hex::encode(&source_hash[..4])),
-                        arch_pubkey_hex: None,
-                        reachable: true,
-                    });
-                    self.persist_peers();
+                match self.peers.entry(source_hash) {
+                    std::collections::hash_map::Entry::Vacant(e) => {
+                        e.insert(ReticulumPeer {
+                            dest_hash: source_hash,
+                            display_name: format!("Reticulum {}", hex::encode(&source_hash[..4])),
+                            arch_pubkey_hex: None,
+                            reachable: true,
+                            last_advert_at: 0,
+                        });
+                        self.persist_peers();
+                    }
+                    std::collections::hash_map::Entry::Occupied(mut e) => {
+                        e.get_mut().reachable = true;
+                    }
                 }
                 use base64::{engine::general_purpose::STANDARD as B64, Engine as _};
                 let Some(data) = ev
@@ -1089,8 +1193,10 @@ pub(crate) async fn probe_rnode(path: &str) -> Result<()> {
     // ESP32-S3 native-USB boards (Heltec V3/V4 etc. — no separate USB-UART
     // bridge chip) treat a DTR/RTS transition on open as a reset signal, the
     // same mechanism esptool uses to force bootloader entry. Deassert both
-    // and let the board settle before writing the probe, or the reboot eats
-    // the DETECT_RESP window below.
+    // before writing the probe. Boards behind a USB-UART bridge (CP2102 on
+    // the Heltec V3) get the reset pulse from the open() itself, before we
+    // can deassert anything — that case is handled by the boot-settle retry
+    // below.
     let _ = port.set_dtr(false);
     let _ = port.set_rts(false);
     tokio::time::sleep(Duration::from_millis(300)).await;
@@ -1109,26 +1215,81 @@ pub(crate) async fn probe_rnode(path: &str) -> Result<()> {
         0x00,
         KISS_FEND,
     ];
+    // Attempt 1: probe immediately. A board that did NOT reset on open (it
+    // was already up — e.g. a re-probe of a running RNode) answers in well
+    // under a second, so the fast path stays fast.
     tokio::time::timeout(Duration::from_millis(500), port.write_all(&probe))
         .await
         .context("RNode probe write timed out")?
         .context("RNode probe write failed")?;
-
     let mut buf = [0u8; 256];
     let mut seen = Vec::new();
-    let deadline = tokio::time::Instant::now() + PROBE_READ_TIMEOUT;
-    while tokio::time::Instant::now() < deadline {
+    if await_detect_resp(&port, &mut buf, &mut seen, PROBE_READ_TIMEOUT).await {
+        return Ok(());
+    }
+
+    // No DETECT_RESP. If the open() power-cycled the board (verified live on
+    // a Heltec V3 RNode behind a CP2102: the ESP32 spends ~2.5-3s in boot
+    // ROM + app init and silently eats anything written meanwhile, so the
+    // first probe lands in the void), wait for its boot chatter to go quiet
+    // and probe once more with a fresh response window.
+    const BOOT_QUIET_WINDOW: Duration = Duration::from_millis(800);
+    const BOOT_SETTLE_MAX: Duration = Duration::from_secs(6);
+    let settle_deadline = tokio::time::Instant::now() + BOOT_SETTLE_MAX;
+    let mut last_data = tokio::time::Instant::now();
+    while tokio::time::Instant::now() < settle_deadline {
         match tokio::time::timeout(Duration::from_millis(150), port.read(&mut buf)).await {
             Ok(Ok(n)) if n > 0 => {
                 seen.extend_from_slice(&buf[..n]);
+                // A late DETECT_RESP to the first write still counts.
                 if contains_detect_resp(&seen) {
                     return Ok(());
                 }
+                last_data = tokio::time::Instant::now();
+            }
+            _ => {
+                if last_data.elapsed() >= BOOT_QUIET_WINDOW {
+                    break;
+                }
+            }
+        }
+    }
+
+    tokio::time::timeout(Duration::from_millis(500), port.write_all(&probe))
+        .await
+        .context("RNode probe rewrite timed out")?
+        .context("RNode probe rewrite failed")?;
+    seen.clear();
+    if await_detect_resp(&port, &mut buf, &mut seen, PROBE_READ_TIMEOUT).await {
+        return Ok(());
+    }
+    anyhow::bail!(
+        "No RNode DETECT_RESP within {:?} (incl. post-boot-settle retry)",
+        PROBE_READ_TIMEOUT
+    )
+}
+
+/// Read from `port` for up to `window`, accumulating into `seen`; true once
+/// the KISS DETECT_RESP sequence shows up anywhere in the stream.
+async fn await_detect_resp(
+    port: &serial2_tokio::SerialPort,
+    buf: &mut [u8],
+    seen: &mut Vec<u8>,
+    window: Duration,
+) -> bool {
+    let deadline = tokio::time::Instant::now() + window;
+    while tokio::time::Instant::now() < deadline {
+        match tokio::time::timeout(Duration::from_millis(150), port.read(buf)).await {
+            Ok(Ok(n)) if n > 0 => {
+                seen.extend_from_slice(&buf[..n]);
+                if contains_detect_resp(seen) {
+                    return true;
+                }
             }
             _ => continue,
         }
     }
-    anyhow::bail!("No RNode DETECT_RESP within {:?}", PROBE_READ_TIMEOUT)
+    false
 }
 
 /// Look for the `[FEND, CMD_DETECT, DETECT_RESP]` sequence anywhere in the
@@ -1138,6 +1299,33 @@ fn contains_detect_resp(buf: &[u8]) -> bool {
         .any(|w| w == [KISS_FEND, KISS_CMD_DETECT, KISS_DETECT_RESP])
 }
 
+/// The display name an announce actually asserted, if any.
+///
+/// Precedence: the daemon-decoded LXMF display name (`display_name` event
+/// field), then — legacy peers only — bare-utf8 app_data that wasn't an
+/// identity blob. The bare-utf8 fallback must actually look like text:
+/// lossy-decoded msgpack (a new-format announce whose name the daemon failed
+/// to decode) is full of U+FFFD/control chars and would otherwise become a
+/// mojibake display name. `None` (e.g. a blob-only legacy announce) means
+/// "no name asserted" and must NOT stomp a previously-learned name.
+fn pick_announced_name(
+    explicit_name: Option<String>,
+    app_data_text: Option<String>,
+    is_legacy_blob: bool,
+) -> Option<String> {
+    explicit_name
+        // A legacy blob-only announce utf8-decodes cleanly, so LXMF's
+        // display_name_from_app_data hands the daemon the ENTIRE `ARCHY:…`
+        // string as a "name" — seen live from a pre-upgrade Framework PT.
+        // An identity blob is never a display name.
+        .filter(|s| !s.starts_with("ARCHY:"))
+        .or_else(|| {
+            app_data_text
+                .filter(|_| !is_legacy_blob)
+                .filter(|s| !s.chars().any(|c| c.is_control() || c == '\u{FFFD}'))
+        })
+}
+
 impl Drop for ReticulumLink {
     fn drop(&mut self) {
         // Group-wide SIGTERM with a delayed SIGKILL backstop (`terminate_group`).
@@ -1154,6 +1342,37 @@ impl Drop for ReticulumLink {
 mod tests {
     use super::*;
 
+    #[test]
+    fn announced_name_precedence() {
+        // Daemon-decoded LXMF name always wins.
+        assert_eq!(
+            pick_announced_name(
+                Some("RNode Shaza".into()),
+                Some("ARCHY:2:aa:bb".into()),
+                true
+            ),
+            Some("RNode Shaza".to_string())
+        );
+        // Legacy bare-utf8 name (old daemon, no explicit field).
+        assert_eq!(
+            pick_announced_name(None, Some("zaza".into()), false),
+            Some("zaza".to_string())
+        );
+        // Legacy blob-only announce asserts NO name (must not stomp).
+        assert_eq!(
+            pick_announced_name(None, Some("ARCHY:2:aa:bb".into()), true),
+            None
+        );
+        // Lossy-decoded msgpack must not become a mojibake name.
+        assert_eq!(
+            pick_announced_name(None, Some("\u{FFFD}\u{FFFD}Shaza\u{FFFD}".into()), false),
+            None
+        );
+        assert_eq!(pick_announced_name(None, Some("has\u{1}ctl".into()), false), None);
+        // Nothing at all.
+        assert_eq!(pick_announced_name(None, None, false), None);
+    }
+
     #[test]
     fn detect_resp_found_in_kiss_stream() {
         let stream = [

core/archipelago/src/mesh/serial.rs

@@ -557,29 +557,39 @@ fn likely_non_mesh_serial_device(path: &str) -> bool {
 /// Scan for serial devices that could be Meshcore radios.
 /// Returns paths to existing serial device files.
 ///
-/// Candidates are deduplicated by canonical path: /dev/mesh-radio is a udev
-/// symlink to a ttyUSB*/ttyACM* node that is ALSO in the candidate list, so
-/// without this one physical board shows up (and gets probed, and gets its
-/// DTR/RTS reset toggled) twice per cycle. The first candidate wins, which
-/// keeps the stable /dev/mesh-radio name when the symlink exists.
+/// Dedupes by canonical (symlink-resolved) target: `/dev/mesh-radio` is a
+/// stable udev symlink to whatever `/dev/ttyUSB*`/`/dev/ttyACM*` node the
+/// primary radio currently enumerates as, so both names always pointed at
+/// the same candidate list entry and both passed this scan — confirmed live
+/// 2026-07-23, this made an already-connected, working radio (connected via
+/// its `/dev/mesh-radio` alias) simultaneously appear as a second, separate
+/// "detected but unclaimed" device under its raw `/dev/ttyUSBn` name. The
+/// hot-swap UI's active-session guard compares path strings, so it didn't
+/// recognize the two aliases as the same port, showed the "device detected"
+/// modal for a radio that was already set up, and probing it there opened
+/// (and DTR/RTS-reset) the exact port the live session was mid-conversation
+/// with — a continuous, UI-driven reset loop that only ran while that view
+/// was open (matches the reported "stops when I leave, resumes when I come
+/// back"). SERIAL_CANDIDATES lists `/dev/mesh-radio` first, so it wins the
+/// dedup and is what's reported when both alias and target are present.
+/// (Independently re-discovered and fixed on main 2026-07-26 — both sides
+/// of the 2026-07-28 merge carried an equivalent implementation.)
 pub async fn detect_serial_devices() -> Vec<String> {
     let mut devices = Vec::new();
-    let mut seen_canonical: Vec<std::path::PathBuf> = Vec::new();
+    let mut seen_real_paths = std::collections::HashSet::new();
     for path in SERIAL_CANDIDATES {
         if tokio::fs::metadata(path).await.is_ok() {
             if likely_non_mesh_serial_device(path) {
                 debug!(path = %path, "Skipping known non-mesh serial device");
                 continue;
             }
-            let canonical = tokio::fs::canonicalize(path)
+            let real_path = tokio::fs::canonicalize(path)
                 .await
                 .unwrap_or_else(|_| std::path::PathBuf::from(path));
-            if seen_canonical.contains(&canonical) {
-                debug!(path = %path, canonical = %canonical.display(),
-                    "Skipping alias of an already-detected serial device");
+            if !seen_real_paths.insert(real_path.clone()) {
+                debug!(path = %path, real_path = %real_path.display(), "Skipping duplicate alias for an already-listed device");
                 continue;
             }
-            seen_canonical.push(canonical);
             devices.push(path.to_string());
         }
     }
@@ -610,12 +620,23 @@ pub async fn detect_serial_devices_info() -> Vec<DetectedDeviceInfo> {
     let mut out = Vec::new();
     for path in detect_serial_devices().await {
         let usb = usb_info_for_tty(&path).await;
-        let plugged_at = tokio::fs::metadata(&path)
-            .await
-            .ok()
-            .and_then(|m| m.modified().ok())
-            .and_then(|t| t.duration_since(std::time::UNIX_EPOCH).ok())
-            .map(|d| d.as_secs());
+        // Birth time (btime), falling back to inode-change time (ctime) —
+        // NOT mtime: a tty node's mtime bumps on every open()/write, so with
+        // mtime here each probe/session open minted a "new" plugged_at, the
+        // UI's (path, plugged_at) dismissal key never matched again, and the
+        // setup modal re-fired forever on a device that never left the port
+        // (observed live on archi-dev-box 2026-07-28). btime/ctime only
+        // change when udev (re)creates/chowns the node — i.e. on real plugs.
+        let plugged_at = tokio::fs::metadata(&path).await.ok().and_then(|m| {
+            m.created()
+                .ok()
+                .and_then(|t| t.duration_since(std::time::UNIX_EPOCH).ok())
+                .map(|d| d.as_secs())
+                .or_else(|| {
+                    use std::os::unix::fs::MetadataExt;
+                    u64::try_from(m.ctime()).ok()
+                })
+        });
         out.push(DetectedDeviceInfo {
             path,
             vid: usb.0,

core/archipelago/src/server.rs

@@ -294,8 +294,13 @@ impl Server {
                         .await
                         .unwrap_or_default();
 
-                    // Auto-enable mesh if a radio is detected and no config exists yet
-                    if !mesh_config.enabled {
+                    // Auto-enable mesh if a radio is detected and no config exists
+                    // yet. Only on a genuinely missing config: an existing file
+                    // with enabled=false is an explicit operator decision (e.g.
+                    // via mesh.configure) and force-re-enabling it on every boot
+                    // made "disable mesh" impossible on any node with a radio
+                    // plugged in.
+                    if !mesh_config.enabled && !crate::mesh::config_file_exists(&data_dir) {
                         let devices = crate::mesh::detect_devices().await;
                         if !devices.is_empty() {
                             info!("📡 Auto-detected mesh radio: {:?} — enabling mesh", devices);
@@ -1322,11 +1327,46 @@ async fn accept_loop(
                     }
                 };
                 let handler = handler.clone();
-                let permit = active_connections.clone().acquire_owned().await;
+                // NEVER park the accept loop on the connection budget.
+                // `acquire_owned().await` here froze accept() entirely when
+                // permits drained — and permits drained because half-open
+                // clients and hung upstreams held them forever (the .228
+                // session-flapping / CLOSE-WAIT `inode: 0` signature). Shed
+                // load instead: accept, answer 503, close.
+                let permit = match active_connections.clone().try_acquire_owned() {
+                    Ok(p) => p,
+                    Err(_) => {
+                        warn!(
+                            "{} connection budget exhausted — shedding {}",
+                            local_addr, peer_addr
+                        );
+                        tokio::spawn(async move {
+                            use tokio::io::AsyncWriteExt;
+                            let mut stream = stream;
+                            let _ = tokio::time::timeout(
+                                std::time::Duration::from_secs(5),
+                                stream.write_all(
+                                    b"HTTP/1.1 503 Service Unavailable\r\nConnection: close\r\nContent-Length: 0\r\n\r\n",
+                                ),
+                            )
+                            .await;
+                            let _ = stream.shutdown().await;
+                        });
+                        continue;
+                    }
+                };
                 tokio::spawn(async move {
                     let _permit = permit;
+                    // Set when a request carries an Upgrade header (websocket):
+                    // upgraded connections are legitimately long-lived and are
+                    // exempt from the non-upgraded connection deadline below.
+                    let upgraded = Arc::new(std::sync::atomic::AtomicBool::new(false));
+                    let upgraded_flag = upgraded.clone();
                     let service = service_fn(move |mut req: hyper::Request<hyper::Body>| {
                         let handler = handler.clone();
+                        if req.headers().contains_key(hyper::header::UPGRADE) {
+                            upgraded_flag.store(true, std::sync::atomic::Ordering::Relaxed);
+                        }
                         async move {
                             // Record the TCP peer so rate limiting only trusts
                             // forwarded headers on loopback (nginx) connections.
@@ -1345,13 +1385,48 @@ async fn accept_loop(
                                 .map_err(|e| std::io::Error::other(format!("{}", e)))
                         }
                     });
-                    if let Err(e) = Http::new()
+                    // header_read_timeout: a client that connects and never
+                    // sends a request (slowloris / half-open) is dropped
+                    // instead of holding a permit until the heat death of the
+                    // node. Long RPCs are safe — the clock only covers header
+                    // read.
+                    let conn = Http::new()
                         .http1_keep_alive(false)
+                        .http1_header_read_timeout(std::time::Duration::from_secs(30))
                         .serve_connection(stream, service)
-                        .with_upgrades()
-                        .await
-                    {
-                        error!("Error serving connection from {}: {}", peer_addr, e);
+                        .with_upgrades();
+                    tokio::pin!(conn);
+                    // Deadline watchdog for NON-upgraded connections. With
+                    // keep-alive off a plain connection serves one exchange;
+                    // 15 min bounds even the slowest legitimate RPC/stream
+                    // while guaranteeing a hung upstream can't hold a permit
+                    // forever. Upgraded (websocket) connections are exempt.
+                    const NON_UPGRADED_DEADLINE: std::time::Duration =
+                        std::time::Duration::from_secs(900);
+                    let started = std::time::Instant::now();
+                    let watchdog = async {
+                        loop {
+                            tokio::time::sleep(std::time::Duration::from_secs(30)).await;
+                            if !upgraded.load(std::sync::atomic::Ordering::Relaxed)
+                                && started.elapsed() >= NON_UPGRADED_DEADLINE
+                            {
+                                return;
+                            }
+                        }
+                    };
+                    tokio::select! {
+                        r = &mut conn => {
+                            if let Err(e) = r {
+                                error!("Error serving connection from {}: {}", peer_addr, e);
+                            }
+                        }
+                        _ = watchdog => {
+                            warn!(
+                                "connection from {} exceeded {}s without completing or upgrading — dropping",
+                                peer_addr,
+                                NON_UPGRADED_DEADLINE.as_secs()
+                            );
+                        }
                     }
                 });
             }

docs/RETICULUM-TRANSPORT-PROGRESS.md

@@ -12,6 +12,33 @@ Full plan: `.claude/plans/enchanted-strolling-rocket.md`. Memory pointer:
 avoiding `mesh/listener/session.rs` transport plumbing + `mesh/mod.rs` routing, which this work
 owns. Stay out of `meshtastic.rs`/`protocol.rs` to avoid collisions.
 
+## Checkpoint 2026-07-28 — RNode connect + names FIXED, live-verified E2E (read this first)
+
+The fleet reflash back to RNode firmware exposed a stack of bugs that made Reticulum
+unusable on CP2102-bridged boards (Heltec V3 etc.) and left every archy node nameless on
+RNS. All fixed in `a8c4694c` (backend) + `3f76b496` (UI), live-verified on archi-dev-box
+and archy-x250-dev with a real RNode-to-RNode LXMF message (`transport: "reticulum"` in
+mesh-messages) plus a cross-transport reply:
+
+1. **probe_rnode boot race** — serial open pulses DTR/RTS via the USB-UART bridge → ESP32
+   power-cycles → KISS DETECT written 300ms later is eaten during ~2.5-3s of boot. Fix:
+   immediate probe (fast path) + drain-until-quiet boot settle + second DETECT window.
+2. **configure() was a no-op on a running listener** (only enable/disable restarted it) —
+   the setup modal's apply/keep-as-is and every rename did nothing until process restart.
+3. **Name propagation** — `config.advert_name` had no reader; `server.set-name` never
+   reached mesh; daemon display name fixed at spawn to the "Archy" default; the ARCHY:2
+   announce blob REPLACED the LXMF name. Now: announces carry msgpack
+   `[name, stamp_cost, sf, ARCHY-blob]` (Sideband-compatible, blob invisible to stock
+   clients), daemon has a `set_name` verb, renames bounce the session live.
+4. **Daemon-death detection** (was invisible up to the 30-min RX-stall watchdog),
+   **modal re-trigger loop** (plugged_at used tty mtime → bumps on every open; now
+   btime/ctime), **ARCHY:2 federation-name clobber**, **mesh.refresh RPC** (Refresh button
+   now actually re-queries the radio), **Meshtastic mesh.broadcast now sends NodeInfo**.
+
+Still open here: legacy-format peers (old fleet builds) show as `Reticulum <hex4>` until
+they OTA; RNode RF params still daemon-hardcoded (EU-868 869.525/125k/SF8/CR5); Phase 4
+multi-radio; duty-cycle guard.
+
 ## Status at a glance
 
 | Phase | What | Status |

docs/combined-test-plan-2026-07-22.md

@@ -110,6 +110,61 @@ lands].
 2. ❑ Mobile Home: wallet card directly under My Apps (G5 from the voice epic).
 3. ❑ Mesh RF settings panel (Mesh → Device) still loads and saves.
 
+## H. LoRa radio firmware flashing (Heltec V3/V4, new — extends Section E)
+
+Full v1 scope is 3 firmware families × 2 boards (6 cells); mark each cell
+tested on real hardware vs. code-reviewed only as this is run.
+
+1. ❑ From the hot-swap modal's step 1 (device already probed), press
+   **Flash Firmware…** → new step shows firmware-family + board pickers and
+   the erase-confirmation checkbox; "Erase & Flash Now" stays disabled until
+   family, board, AND the checkbox are all set.
+2. ❑ Confirm what's currently on the test stick via the existing probe
+   BEFORE flashing it — don't flash the only known-good device without a
+   fallback board on hand.
+3. ❑ Prefer a spare Heltec V3/V4 for the first destructive erase+flash run;
+   only exercise a primary/in-use stick once the flow is proven safe.
+4. ❑ MeshCore → Heltec V3: erase + write completes, progress bar and log
+   tail update live, ends at "Flash complete".
+5. ❑ Meshtastic → Heltec V3: same, using the extracted `*.factory.bin` from
+   the esp32s3 release zip.
+6. ❑ Reticulum/RNode → Heltec V3: `archy-rnodeconf --autoinstall` path
+   completes (no raw esptool erase/write step for this family — see
+   `mesh/flash.rs` doc comment).
+7. ❑ Repeat 4-6 against a Heltec V4. Confirmed 2026-07-23 on real hardware:
+   V4 uses the ESP32-S3's native-USB JTAG/serial peripheral (vid:pid
+   303a:1001, generic to every native-USB ESP32-S3 board, not V4-specific)
+   — so unlike V3's CP2102 bridge chip, V4 is permanently NOT auto-matchable
+   by vid:pid. Board auto-detect should fail closed for it every time
+   (manual board selection required, "couldn't confirm automatically"
+   warning shown) — this is expected steady-state behavior, not a gap to
+   close later.
+8. ❑ After a successful flash, the modal automatically re-probes and shows
+   the NEW firmware's badge/details — same as unplugging and replugging
+   (Section E item 3), but without physically touching the cable.
+9. ❑ Deliberately test a failure path once (disconnect the board mid-write,
+   or point at a bad cached asset) — confirm the error surfaces in the
+   progress log AND that `docs/troubleshooting.md`'s "LoRa radio firmware
+   flash failed" recovery steps (BOOT+RST bootloader entry, manual esptool/
+   rnodeconf command) actually get the board back to a flashable state.
+10. ❑ Cancel button only appears (and only works) while still in the
+    "Downloading firmware…" stage — once erasing/writing starts, no cancel
+    affordance is offered.
+11. ❑ **Boot-loop regression (2026-07-23 incident)**: after a *failed* flash
+    (e.g. kill network access mid-download to force a failure), confirm the
+    mesh listener does NOT auto-resume — `journalctl -u archipelago` should
+    show a single `Leaving mesh listener stopped after failed flash` line
+    and then go quiet for that device, not a repeating `mesh::serial:
+    Opened serial port... Starting Meshcore handshake` cycle every few
+    seconds. Reconnect manually via the hot-swap modal afterward and confirm
+    it connects normally (the board itself should be untouched — the
+    download fails before esptool/rnodeconf ever runs).
+12. ❑ Separately, force a device to flap connected/disconnected a few times
+    in under 20s each (e.g. a marginal USB connection) and confirm
+    `reconnect_delay` in the logs actually escalates (5s → 10s → 20s → ...)
+    rather than resetting to 5s on every attempt — see
+    `STABLE_SESSION_THRESHOLD` in `mesh/listener/mod.rs`.
+
 ---
 
 After this passes: fold the batch + other agent's work into the next release

docs/troubleshooting.md

@@ -443,6 +443,86 @@ free -h
 - Check Nginx WebSocket proxy config: `/etc/nginx/sites-available/archipelago` must include `proxy_set_header Upgrade $http_upgrade`
 - If on WiFi, try wired Ethernet for more stable connectivity
 
+### 21. LoRa radio firmware flash failed / board unresponsive
+
+**Symptoms**: The "Erase & Flash Now" flow in the mesh hot-swap modal reports
+an error, or the radio no longer enumerates as a serial device after a flash
+attempt.
+
+**Diagnosis**:
+```bash
+# Poll the flash job's last-known stage/error directly
+curl -s http://localhost:5678/rpc/v1 \
+  -H 'Content-Type: application/json' \
+  -d '{"method":"mesh.flash-status","params":{}}'
+
+# Confirm the board is still enumerating at all
+ls -la /dev/ttyUSB* /dev/ttyACM* /dev/mesh-radio 2>&1
+
+# esptool/rnodeconf binaries present?
+which esptool; ls -la /usr/local/bin/archy-rnodeconf
+```
+
+**Solutions**:
+- A failure during `erasing`/`writing` (MeshCore/Meshtastic) or
+  `autoinstalling` (Reticulum) can leave the chip erased or half-written —
+  this is expected risk of the "always erase first" default, not a bug.
+- Heltec V3/V4 boards can be forced back into bootloader mode manually: hold
+  **BOOT**, tap **RST**, then release **BOOT** — this puts the chip in a
+  state esptool can always talk to, regardless of what firmware (if any) is
+  currently on it.
+- With the board in bootloader mode, a manual recovery flash can be run
+  directly over SSH without the UI:
+  ```bash
+  esptool --chip esp32s3 --port /dev/ttyACM0 erase_flash
+  esptool --chip esp32s3 --port /dev/ttyACM0 write_flash 0x0 <known-good-image.bin>
+  ```
+- For Reticulum/RNode boards, the equivalent manual recovery is
+  `archy-rnodeconf /dev/ttyACM0 --autoinstall` (or `/usr/local/bin/archy-rnodeconf`
+  if it's not on `PATH`) — it re-runs the same fetch+erase+flash+bootstrap
+  sequence the UI triggers.
+- If `esptool`/`archy-rnodeconf` are missing entirely, they should have been
+  installed by the last `self-update.sh` run — check
+  `sudo journalctl -u archipelago-update` for install failures, or install
+  `esptool` via `sudo apt-get install esptool` directly.
+- Once a fresh image is confirmed written, unplug/replug the radio (or wait
+  for the next detection poll) — the hot-swap modal re-probes automatically
+  and shows whatever firmware is actually on the board now.
+
+**Known incident (2026-07-23) — reconnect storm / device boot-loop after a
+failed flash**: a real Heltec V3 got stuck cycling "connect → partial
+handshake → drop" every 5-15s for 5+ minutes after a `mesh.flash-device`
+attempt failed with `Reading firmware download stream`. Root cause was two
+compounding issues, both now fixed:
+1. `spawn_mesh_listener`'s reconnect backoff (`core/archipelago/src/mesh/listener/mod.rs`)
+   reset to its 5s minimum any time the prior session had been `device_connected`
+   at all, even for under a second — so a device that connects-then-drops
+   repeatedly never actually backed off. Every retry's `open()` toggles
+   DTR/RTS, which resets many ESP32 boards' MCU (native-USB *and*
+   CP2102/CH340 auto-reset-circuit boards), so the aggressive retries were
+   themselves *causing* the boot loop, not just observing one. Fixed by only
+   resetting backoff when a session ran for at least `STABLE_SESSION_THRESHOLD`
+   (20s) — see that constant's doc comment.
+2. `mesh::flash::start_flash_job`'s post-completion handler auto-resumed the
+   listener unconditionally, even after a *failed* flash, immediately
+   re-entering the reconnect loop above with no cooldown. Fixed: on failure
+   the listener is now deliberately left stopped (reconnect manually via the
+   UI once the board is confirmed alive); on success there's a 5s settle
+   delay before resuming, so the board finishes booting from the flash
+   tool's own reset before Archipelago starts probing it again.
+3. Separately, the download itself was failing because `mesh::flash`'s HTTP
+   client had a blanket 30s request timeout that covered the *entire*
+   download (including streaming a 170MB Meshtastic zip), not just
+   connection setup — fixed with a per-chunk stall timeout instead of a
+   fixed total-transfer cap.
+
+If this symptom recurs (rapid repeating `mesh::serial: Opened serial
+port... Starting Meshcore handshake` lines in `journalctl -u archipelago`
+without a `LoRa firmware flash` job in progress), it's a NEW instance of the
+same class of bug, not the one above — check whether backoff is actually
+escalating (`Mesh session error: ... (retry in Xs)` — X should grow past 5s
+within a few cycles) before assuming it's flashing-related.
+
 ---
 
 ## General Maintenance

image-recipe/_archived/build-auto-installer-iso.sh

@@ -365,6 +365,11 @@ RUN apt-get update && apt-get -y full-upgrade && apt-get install -y --no-install
     ca-certificates \
     openssl \
     chrony \
+    iputils-ping \
+    esptool \
+    python3-venv \
+    binutils \
+    libpython3.13 \
     locales \
     console-setup \
     keyboard-configuration \

neode-ui/src/components/mesh/MeshDeviceSetupModal.vue

@@ -1,7 +1,7 @@
 <template>
   <BaseModal
     :show="show"
-    :title="step === 1 ? 'Mesh Radio Detected' : 'Apply Archipelago Settings'"
+    :title="step === 1 ? 'Mesh Radio Detected' : step === 2 ? 'Set Recommended' : 'Flash Firmware'"
     max-width="max-w-lg"
     content-class="max-h-[90vh] overflow-y-auto"
     @close="dismiss"
@@ -35,9 +35,17 @@
 
       <!-- What's currently flashed / configured on it -->
       <div class="mt-4 text-left rounded-xl bg-white/[0.05] border border-white/10 p-3">
-        <div v-if="probing" class="flex items-center gap-2 text-white/60 text-sm py-1">
-          <span class="inline-block w-3.5 h-3.5 rounded-full border-2 border-orange-300/70 border-t-transparent animate-spin"></span>
-          Reading what's on the radio…
+        <div v-if="probing" class="py-1">
+          <div class="flex items-center justify-between text-white/60 text-sm mb-1.5">
+            <span>{{ probeStage }}</span>
+            <span class="text-white/40 text-xs tabular-nums">{{ Math.round(probeProgress) }}%</span>
+          </div>
+          <div class="h-1.5 rounded-full bg-white/10 overflow-hidden">
+            <div
+              class="h-full rounded-full bg-orange-400/80 transition-[width] duration-500 ease-linear"
+              :style="{ width: probeProgress + '%' }"
+            ></div>
+          </div>
         </div>
         <template v-else-if="probe">
           <div class="flex items-center gap-2">
@@ -94,21 +102,123 @@
           :disabled="!!connecting"
           @click="step = 2"
         >
-          Set Up with Archipelago Settings
+          Set Recommended
         </button>
       </div>
       <p class="text-white/40 text-[11px] mt-3">
         "Keep As Is" uses the radio exactly as it is — nothing on it is changed,
         and you can hot-swap radios any time.
       </p>
+      <button
+        class="w-full text-center text-white/40 hover:text-white/70 text-[11px] mt-3 underline underline-offset-2"
+        :disabled="!!connecting"
+        @click="openFlashStep"
+      >
+        Flash Firmware…
+      </button>
       <p v-if="error" class="text-xs text-red-400 mt-2">{{ error }}</p>
     </div>
 
+    <!-- Step 3: erase + reflash — destructive, opt-in only -->
+    <div v-else-if="step === 'flash'">
+      <!-- Once a job exists (started via startFlash), ALWAYS show the
+           progress/result view below — including on failure. The old
+           condition (`!active && stage !== 'done'`) was also true for a
+           FAILED job (active:false, stage:'failed'), which silently sent
+           the user back to this picker instead of showing the error. -->
+      <template v-if="!flashJob">
+        <p class="text-white/60 text-xs mb-3">
+          Downloads the latest firmware from upstream and writes it to
+          <span class="font-mono text-orange-300">{{ devicePath }}</span>.
+        </p>
+
+        <div class="space-y-4">
+          <div>
+            <label class="block text-sm text-white/80 mb-1">Firmware family</label>
+            <select v-model="flashFamily" class="w-full rounded-lg bg-white/[0.06] border border-white/10 text-white px-3 py-2 text-sm focus:outline-none focus:border-orange-400/60">
+              <option value="">Choose…</option>
+              <option value="meshcore">MeshCore</option>
+              <option value="meshtastic">Meshtastic</option>
+              <option value="reticulum">Reticulum RNode</option>
+            </select>
+          </div>
+          <div>
+            <label class="block text-sm text-white/80 mb-1">Board</label>
+            <select v-model="flashBoard" class="w-full rounded-lg bg-white/[0.06] border border-white/10 text-white px-3 py-2 text-sm focus:outline-none focus:border-orange-400/60">
+              <option value="">Choose…</option>
+              <option value="heltec-v3">Heltec LoRa 32 V3</option>
+              <option value="heltec-v4">Heltec LoRa 32 V4</option>
+            </select>
+            <p v-if="!boardAutoDetected" class="text-[11px] text-amber-400/80 mt-1">
+              Couldn't confirm the board automatically — double check before flashing.
+              Flashing the wrong board's image can brick it.
+            </p>
+          </div>
+        </div>
+
+        <div class="rounded-xl bg-red-500/10 border border-red-500/30 p-3 mt-4">
+          <label class="flex items-start gap-2 text-xs text-red-300">
+            <input type="checkbox" v-model="flashConfirmed" class="mt-0.5" />
+            <span>
+              This <strong>erases the entire chip</strong>, including any existing
+              keys, identity, and contacts. This cannot be undone.
+            </span>
+          </label>
+        </div>
+
+        <p v-if="error" class="text-xs text-red-400 mt-3">{{ error }}</p>
+
+        <div class="flex gap-2 mt-6">
+          <button class="glass-button px-4 py-2 rounded-lg text-sm" @click="step = 1">Back</button>
+          <button
+            class="flex-1 glass-button px-4 py-2 rounded-lg text-sm font-medium bg-red-500/80 hover:bg-red-500 text-white disabled:opacity-50"
+            :disabled="!flashFamily || !flashBoard || !flashConfirmed || starting"
+            @click="startFlash"
+          >
+            {{ starting ? 'Starting…' : 'Erase & Flash Now' }}
+          </button>
+        </div>
+      </template>
+
+      <!-- Progress -->
+      <template v-else>
+        <div class="text-center py-2">
+          <p class="text-white text-sm font-medium">{{ flashStageLabel }}</p>
+          <div class="mt-3 h-2 rounded-full bg-white/10 overflow-hidden">
+            <div
+              class="h-full bg-orange-400 transition-all"
+              :style="{ width: (flashJob?.percent ?? (flashJob?.stage === 'done' ? 100 : 8)) + '%' }"
+            ></div>
+          </div>
+          <p v-if="flashJob?.error" class="text-xs text-red-400 mt-3">{{ flashJob.error }}</p>
+        </div>
+        <div class="mt-3 rounded-xl bg-black/30 border border-white/10 p-2 h-32 overflow-y-auto font-mono text-[10px] text-white/50 leading-relaxed">
+          <div v-for="(line, i) in (flashJob?.log_tail ?? []).slice(-40)" :key="i">{{ line }}</div>
+        </div>
+        <div class="flex gap-2 mt-4">
+          <button
+            v-if="flashJob?.stage === 'downloading' && flashJob?.active"
+            class="glass-button px-4 py-2 rounded-lg text-sm"
+            @click="cancelFlash"
+          >
+            Cancel
+          </button>
+          <button
+            v-if="!flashJob?.active"
+            class="flex-1 glass-button glass-button-warning px-4 py-2 rounded-lg text-sm font-medium"
+            @click="closeFlashStep"
+          >
+            Done
+          </button>
+        </div>
+      </template>
+    </div>
+
     <!-- Step 2: our latest parameters, shown before anything is written -->
     <div v-else>
       <p class="text-white/60 text-xs mb-3">
-        These are the latest Archipelago settings — nothing is written to the
-        radio until you confirm.
+        These are the recommended Archipelago settings — nothing is written to
+        the radio until you confirm.
       </p>
 
       <!-- Summary of what will be applied -->
@@ -190,7 +300,7 @@
 import { ref, computed, watch } from 'vue'
 import { useRouter } from 'vue-router'
 import BaseModal from '@/components/BaseModal.vue'
-import { useMeshStore, type MeshDeviceProbe, type MeshConfigureParams } from '@/stores/mesh'
+import { useMeshStore, type MeshDeviceProbe, type MeshConfigureParams, type FlashFirmwareFamily, type FlashBoard, type FlashJobStatus } from '@/stores/mesh'
 import { useAppStore } from '@/stores/app'
 import { LORA_REGIONS, regionByCode, suggestRegionFromLatLon, meshcorePlanFor } from '@/utils/loraRegions'
 import { resolveMeshDeviceImage } from '@/utils/meshDeviceImages'
@@ -199,13 +309,37 @@ const mesh = useMeshStore()
 const appStore = useAppStore()
 const router = useRouter()
 
-const step = ref<1 | 2>(1)
+const step = ref<1 | 2 | 'flash'>(1)
 const connecting = ref<false | 'keep' | 'setup'>(false)
 const error = ref('')
 const probing = ref(false)
 const probe = ref<MeshDeviceProbe | null>(null)
 const probeError = ref('')
 
+// Time-driven probe progress: the probe RPC is a single opaque call that can
+// take ~5-30s (boot settle + up to three firmware handshakes), so the bar
+// advances on a clock toward 92% and snaps to 100% when the result lands.
+const probeProgress = ref(0)
+const probeStage = ref('Waiting for the radio to boot…')
+let probeTicker: ReturnType<typeof setInterval> | null = null
+function startProbeProgress() {
+  stopProbeProgress()
+  probeProgress.value = 0
+  probeStage.value = 'Waiting for the radio to boot…'
+  const startedAt = Date.now()
+  probeTicker = setInterval(() => {
+    const elapsed = (Date.now() - startedAt) / 1000
+    // ~92% at 30s, decelerating — never looks stuck, never lies "done".
+    probeProgress.value = Math.min(92, 100 * (1 - Math.exp(-elapsed / 11)))
+    if (elapsed >= 4) probeStage.value = 'Detecting firmware…'
+    if (elapsed >= 18) probeStage.value = 'Still checking (radios can be slow to answer)…'
+  }, 400)
+}
+function stopProbeProgress(done = false) {
+  if (probeTicker) { clearInterval(probeTicker); probeTicker = null }
+  if (done) probeProgress.value = 100
+}
+
 const devicePath = computed(() => mesh.undismissedDetectedDevices[0] ?? '')
 const show = computed(() => !!devicePath.value)
 const imageFailed = ref(false)
@@ -264,7 +398,10 @@ const rfPreset = computed(() => {
 
 // (Re)probe + (re)apply presets each time a new device surfaces the modal
 watch([show, devicePath], async ([visible]) => {
-  if (!visible) return
+  if (!visible) {
+    stopFlashPoll()
+    return
+  }
   step.value = 1
   error.value = ''
   imageFailed.value = false
@@ -275,6 +412,7 @@ watch([show, devicePath], async ([visible]) => {
   probe.value = null
   probeError.value = ''
   probing.value = true
+  startProbeProgress()
   const path = devicePath.value
   try {
     const res = await mesh.probeDevice(path)
@@ -284,6 +422,7 @@ watch([show, devicePath], async ([visible]) => {
       probeError.value = e instanceof Error ? e.message : String(e)
     }
   } finally {
+    stopProbeProgress(true)
     if (devicePath.value === path) probing.value = false
   }
 }, { immediate: false })
@@ -349,6 +488,118 @@ async function applySetup() {
     connecting.value = false
   }
 }
+
+// ─── Step 3: erase + reflash ─────────────────────────────────────────────
+const flashFamily = ref<FlashFirmwareFamily | ''>('')
+const flashBoard = ref<FlashBoard | ''>('')
+const flashConfirmed = ref(false)
+const starting = ref(false)
+const flashJob = ref<FlashJobStatus | null>(null)
+let flashPollTimer: ReturnType<typeof setInterval> | null = null
+
+const detectedInfo = computed(() =>
+  mesh.status?.detected_device_info?.find(d => d.path === devicePath.value)
+)
+
+// Mirrors mesh::flash::resolve_flash_board (core/archipelago/src/mesh/flash.rs)
+// exactly — matching on the display label was wrong: a Heltec V3's CP2102
+// bridge chip reports "CP2102 USB to UART Bridge Controller" in its USB
+// strings, not "Heltec", so meshDeviceImages.ts falls back to a generic
+// "LoRa radio (CP2102 serial)" label that never matched /v3/i, showing the
+// "couldn't confirm automatically" warning even though the backend CAN
+// safely auto-detect V3 via vid:pid. Heltec V4 deliberately has no entry
+// here, same reasoning as the backend: its vid:pid (303a:1001) is the
+// ESP32-S3's generic native-USB descriptor, not V4-specific, so it can't be
+// safely auto-matched and always requires manual selection.
+const resolvedFlashBoard = computed<FlashBoard | ''>(() => {
+  const info = detectedInfo.value
+  if (info?.vid?.toLowerCase() === '10c4' && info?.pid?.toLowerCase() === 'ea60') return 'heltec-v3'
+  return ''
+})
+
+const boardAutoDetected = computed(() => !!resolvedFlashBoard.value)
+
+const flashStageLabel = computed(() => {
+  switch (flashJob.value?.stage) {
+    case 'downloading': return 'Downloading firmware…'
+    case 'erasing': return 'Erasing chip…'
+    case 'writing': return 'Writing firmware…'
+    case 'autoinstalling': return 'Installing (rnodeconf)…'
+    case 'done': return 'Flash complete'
+    case 'failed': return 'Flash failed'
+    default: return ''
+  }
+})
+
+function openFlashStep() {
+  flashFamily.value = (probe.value?.kind as FlashFirmwareFamily) ?? ''
+  flashBoard.value = resolvedFlashBoard.value
+  flashConfirmed.value = false
+  flashJob.value = null
+  error.value = ''
+  step.value = 'flash'
+}
+
+function stopFlashPoll() {
+  if (flashPollTimer) {
+    clearInterval(flashPollTimer)
+    flashPollTimer = null
+  }
+}
+
+async function pollFlashStatus() {
+  try {
+    const status = await mesh.flashStatus()
+    flashJob.value = status
+    if (!status.active) {
+      stopFlashPoll()
+      if (status.done && !status.error) {
+        // Mirrors the unplug/replug hot-swap flow: re-probe so the details
+        // card reflects whatever firmware is actually on the board now.
+        const path = devicePath.value
+        probing.value = true
+        try {
+          probe.value = await mesh.probeDevice(path)
+        } catch {
+          probe.value = null
+        } finally {
+          probing.value = false
+        }
+      }
+    }
+  } catch {
+    stopFlashPoll()
+  }
+}
+
+async function startFlash() {
+  if (!flashFamily.value || !flashBoard.value || !flashConfirmed.value) return
+  starting.value = true
+  error.value = ''
+  try {
+    await mesh.flashDevice(devicePath.value, flashFamily.value, flashBoard.value)
+    flashJob.value = { active: true, stage: 'downloading', log_tail: [] }
+    stopFlashPoll()
+    flashPollTimer = setInterval(pollFlashStatus, 1500)
+  } catch (e) {
+    error.value = e instanceof Error ? e.message : 'Failed to start flashing'
+  } finally {
+    starting.value = false
+  }
+}
+
+async function cancelFlash() {
+  try {
+    await mesh.flashCancel()
+  } catch (e) {
+    error.value = e instanceof Error ? e.message : 'Failed to cancel'
+  }
+}
+
+function closeFlashStep() {
+  stopFlashPoll()
+  step.value = 1
+}
 </script>
 
 <style scoped>

neode-ui/src/stores/mesh.ts

@@ -57,6 +57,23 @@ export interface MeshDeviceProbe {
   max_contacts: number | null
 }
 
+export type FlashFirmwareFamily = 'meshcore' | 'meshtastic' | 'reticulum'
+export type FlashBoard = 'heltec-v3' | 'heltec-v4'
+export type FlashStage = 'downloading' | 'erasing' | 'writing' | 'autoinstalling' | 'done' | 'failed'
+
+/** Live progress for the one flash job that can run at a time. */
+export interface FlashJobStatus {
+  active: boolean
+  board?: FlashBoard
+  family?: FlashFirmwareFamily
+  path?: string
+  stage?: FlashStage
+  percent?: number | null
+  log_tail?: string[]
+  done?: boolean
+  error?: string | null
+}
+
 /** Params accepted by mesh.configure (superset of the status fields). */
 export interface MeshConfigureParams {
   enabled?: boolean
@@ -274,12 +291,16 @@ export const useMeshStore = defineStore('mesh', () => {
   async function fetchStatus() {
     try {
       loading.value = true
-      error.value = null
       const res = await rpcClient.call<MeshStatus>({ method: 'mesh.status' })
       status.value = res
       trackDetectedDevices(res)
     } catch (err: unknown) {
-      error.value = err instanceof Error ? err.message : 'Failed to fetch mesh status'
+      // Don't clobber a user-action error (broadcast/configure/send) — this
+      // runs on a 5s poll, and the old `error.value = null` on entry meant
+      // any real error banner survived at most one poll tick.
+      if (!error.value) {
+        error.value = err instanceof Error ? err.message : 'Failed to fetch mesh status'
+      }
     } finally {
       loading.value = false
     }
@@ -358,6 +379,32 @@ export const useMeshStore = defineStore('mesh', () => {
       timeout: 45000, // serial probes are slow (multi-firmware handshakes)
     })
   }
+  /** Available firmware version(s) for a family — v1 only ever returns
+   *  ["latest"], since firmware is always fetched from upstream at flash
+   *  time rather than pinned/bundled. */
+  async function flashListFirmware(family: FlashFirmwareFamily): Promise<string[]> {
+    const res = await rpcClient.call<{ versions: string[] }>({
+      method: 'mesh.flash-list-firmware',
+      params: { family },
+    })
+    return res.versions
+  }
+  /** Erase + reflash a detected radio. `board` is optional — omit it to let
+   *  the backend auto-resolve from the port's USB vid:pid; if that fails
+   *  (e.g. Heltec V4 not yet in the vid:pid table), it errors and the UI
+   *  must ask the user to pick the board explicitly. Always erases first. */
+  async function flashDevice(path: string, family: FlashFirmwareFamily, board?: FlashBoard): Promise<void> {
+    await rpcClient.call({
+      method: 'mesh.flash-device',
+      params: board ? { path, family, board } : { path, family },
+    })
+  }
+  async function flashStatus(): Promise<FlashJobStatus> {
+    return rpcClient.call<FlashJobStatus>({ method: 'mesh.flash-status' })
+  }
+  async function flashCancel(): Promise<void> {
+    await rpcClient.call({ method: 'mesh.flash-cancel' })
+  }
   let globalDetectTimer: ReturnType<typeof setInterval> | null = null
   /** App-wide light poll so the detected-device modal works on every page
    *  (the Mesh view's own 5s poll takes over while it is mounted). */
@@ -975,6 +1022,18 @@ export const useMeshStore = defineStore('mesh', () => {
     await Promise.all([fetchStatus(), fetchPeers(), fetchMessages(), fetchDeadmanStatus(), fetchBlockHeaders()])
   }
 
+  /** Ask the backend to actively re-query the radio's contact table (and by
+   *  extension re-drain daemon events for Reticulum) — the server-side half
+   *  of the Refresh button; refreshAll() alone only re-reads caches. */
+  async function refreshRadio(): Promise<boolean> {
+    try {
+      const res = await rpcClient.call<{ refreshed: boolean }>({ method: 'mesh.refresh' })
+      return !!res.refreshed
+    } catch {
+      return false
+    }
+  }
+
   return {
     status,
     peers,
@@ -994,6 +1053,10 @@ export const useMeshStore = defineStore('mesh', () => {
     undismissedDetectedDevices,
     dismissDetectedDevice,
     probeDevice,
+    flashListFirmware,
+    flashDevice,
+    flashStatus,
+    flashCancel,
     startGlobalDetection,
     fetchPeers,
     fetchMessages,
@@ -1002,6 +1065,7 @@ export const useMeshStore = defineStore('mesh', () => {
     broadcastIdentity,
     configure,
     refreshAll,
+    refreshRadio,
     markChatRead,
     clearViewingChat,
     sendInvoice,

neode-ui/src/views/Mesh.vue

@@ -38,6 +38,8 @@ const activeChatChannel = ref<{ index: number; name: string } | null>(null)
 const messageText = ref('')
 const sendError = ref('')
 const broadcasting = ref(false)
+const broadcastResult = ref<string | null>(null) // 'ok' | error message
+const refreshing = ref(false)
 const configuring = ref(false)
 const connectingDevice = ref<string | null>(null)
 // Device-detected onboarding now lives in the global MeshDeviceSetupModal (App.vue).
@@ -383,12 +385,22 @@ onMounted(async () => {
     archPollInterval = setInterval(loadArchMessages, 15000)
   }
   if (!pollInterval) {
+    let tick = 0
     pollInterval = setInterval(() => {
       mesh.fetchStatus()
       mesh.fetchPeers()
       mesh.fetchMessages()
       mesh.fetchDeadmanStatus()
       mesh.fetchBlockHeaders()
+      // Contacts/aliases, federation nodes and the outbox badge previously
+      // loaded ONCE at mount and went permanently stale — new federation
+      // peers or renames never appeared without a full page reload. Every
+      // 6th tick (~30s) keeps them fresh without adding per-5s load.
+      if (++tick % 6 === 0) {
+        void refreshContacts()
+        void refreshFederationNodes()
+        void refreshOutboxCount()
+      }
     }, 5000)
   }
 
@@ -1021,7 +1033,37 @@ function onChatWheel(e: WheelEvent) {
 
 async function handleBroadcast() {
   broadcasting.value = true
-  try { await mesh.broadcastIdentity() } finally { broadcasting.value = false }
+  broadcastResult.value = null
+  try {
+    await mesh.broadcastIdentity()
+    broadcastResult.value = 'ok'
+  } catch (e) {
+    broadcastResult.value = e instanceof Error ? e.message : 'Broadcast failed'
+  } finally {
+    broadcasting.value = false
+    setTimeout(() => { broadcastResult.value = null }, 4000)
+  }
+}
+
+async function handleRefresh() {
+  if (refreshing.value) return
+  refreshing.value = true
+  try {
+    // Backend first: re-query the radio's contact table (mesh.refresh), then
+    // re-read EVERYTHING the list is built from — peers, contacts/aliases,
+    // federation nodes, outbox — not just the mesh caches.
+    await Promise.allSettled([
+      mesh.refreshRadio(),
+      mesh.refreshAll(),
+      refreshContacts(),
+      refreshFederationNodes(),
+      refreshOutboxCount(),
+    ])
+    // Radio contact refresh is async on the backend — pick up its result.
+    await mesh.fetchPeers()
+  } finally {
+    refreshing.value = false
+  }
 }
 
 async function handleToggleEnabled() {
@@ -1830,8 +1872,14 @@ async function downloadAttachment(payload: MeshAttachmentPayload) {
           <button class="glass-button mesh-action-btn" :disabled="configuring" @click="handleToggleEnabled">
             {{ mesh.status?.enabled ? 'Disable' : 'Enable' }}
           </button>
-          <button class="glass-button mesh-action-btn" :disabled="!mesh.status?.device_connected || broadcasting" @click="handleBroadcast">
-            {{ broadcasting ? 'Sending...' : 'Broadcast' }}
+          <button
+            class="glass-button mesh-action-btn"
+            :class="broadcastResult === 'ok' ? 'mesh-action-ok' : ''"
+            :disabled="!mesh.status?.device_connected || broadcasting"
+            :title="broadcastResult && broadcastResult !== 'ok' ? broadcastResult : 'Announce this node so nearby radios learn about it'"
+            @click="handleBroadcast"
+          >
+            {{ broadcasting ? 'Sending…' : broadcastResult === 'ok' ? 'Sent ✓' : broadcastResult ? 'Failed ✕' : 'Broadcast' }}
           </button>
           <button
             class="glass-button mesh-action-btn"
@@ -1841,7 +1889,10 @@ async function downloadAttachment(payload: MeshAttachmentPayload) {
           >
             {{ transport.meshOnly ? 'Go Online' : 'Off-Grid' }}
           </button>
-          <button class="glass-button mesh-action-btn" @click="mesh.refreshAll()">Refresh</button>
+          <button class="glass-button mesh-action-btn" :disabled="refreshing" @click="handleRefresh">
+            <span v-if="refreshing" class="mesh-refresh-spinner" aria-hidden="true"></span>
+            {{ refreshing ? 'Refreshing…' : 'Refresh' }}
+          </button>
         </div>
 
         <!-- Peers list -->
@@ -1871,7 +1922,10 @@ async function downloadAttachment(payload: MeshAttachmentPayload) {
             >&times;</button>
           </div>
 
-          <div v-if="mesh.peers.length === 0 && !mesh.status?.device_connected" class="mesh-empty">
+          <!-- Only claim "no peers" when the MERGED list (radio + federation)
+               is truly empty — with no radio attached the federation rows and
+               the two channel rows must still render. -->
+          <div v-if="displayedPeers.length === 0 && !mesh.status?.device_connected" class="mesh-empty">
             No peers discovered yet.
           </div>
 
@@ -2140,8 +2194,14 @@ async function downloadAttachment(payload: MeshAttachmentPayload) {
                         <button
                           class="mesh-typed-content-download-btn"
                           title="Download"
+                          aria-label="Download image"
                           @click="downloadAttachment(msg.typed_payload as any)"
-                        >&#x2B07;</button>
+                        >
+                          <svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true">
+                            <path d="M12 4v11m0 0l-4.5-4.5M12 15l4.5-4.5" />
+                            <path d="M5 19h14" />
+                          </svg>
+                        </button>
                       </div>
                       <audio
                         v-else-if="(msg.typed_payload.mime || '').startsWith('audio/')"
@@ -2165,10 +2225,19 @@ async function downloadAttachment(payload: MeshAttachmentPayload) {
                         @click="openMeshLightbox(msg.typed_payload as any)"
                       />
                       <button
-                        class="btn"
+                        class="mesh-typed-content-fetch-btn"
                         :disabled="fetchingCids.has(msg.typed_payload.cid)"
                         @click="handleFetchContent(msg.typed_payload as any)"
                       >
+                        <span
+                          v-if="fetchingCids.has(msg.typed_payload.cid)"
+                          class="mesh-refresh-spinner"
+                          aria-hidden="true"
+                        ></span>
+                        <svg v-else viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true">
+                          <path d="M12 4v11m0 0l-4.5-4.5M12 15l4.5-4.5" />
+                          <path d="M5 19h14" />
+                        </svg>
                         {{ fetchingCids.has(msg.typed_payload.cid) ? 'Fetching…' : 'Download' }}
                       </button>
                     </template>

neode-ui/src/views/mesh/MeshDevicePanel.vue

@@ -142,12 +142,15 @@ async function saveSettings() {
       lora_region: form.value.region,
       device_kind: form.value.deviceKind,
       channel_name: form.value.channel.trim() || 'archipelago',
-      ...(form.value.name.trim() ? { advert_name: form.value.name.trim() } : {}),
+      // Always sent: an empty string CLEARS the custom mesh name (backend
+      // maps "" -> None -> fall back to the server name). The old omit-when-
+      // empty made clearing impossible once a name was ever set.
+      advert_name: form.value.name.trim(),
       broadcast_identity: form.value.broadcastIdentity,
       ...(rfParams ? { lora_radio_params: rfParams } : {}),
     })
     saveDone.value = true
-    setTimeout(() => { saveDone.value = false }, 3000)
+    setTimeout(() => { saveDone.value = false }, 5000)
   } catch (e) {
     saveError.value = e instanceof Error ? e.message : 'Failed to save mesh settings'
   } finally {
@@ -286,7 +289,7 @@ async function saveSettings() {
         >
           {{ saving ? 'Saving…' : 'Save Settings' }}
         </button>
-        <span v-if="saveDone" class="text-xs text-green-400">Saved — applies on next radio session</span>
+        <span v-if="saveDone" class="text-xs text-green-400">Saved — applying to the radio now…</span>
         <span v-if="saveError" class="text-xs text-red-400">{{ saveError }}</span>
       </div>
     </div>

neode-ui/src/views/mesh/mesh-styles.css

@@ -83,6 +83,19 @@
 .mesh-offgrid-active { border-color: rgba(251, 146, 60, 0.4) !important; color: #fb923c !important; }
 .mesh-actions { display: flex; gap: 8px; flex-shrink: 0; }
 .mesh-action-btn { flex: 1; padding: 8px 0; font-size: 0.8rem; }
+.mesh-action-ok { color: #34d399; border-color: rgba(52, 211, 153, 0.4); }
+.mesh-refresh-spinner {
+  display: inline-block;
+  width: 10px;
+  height: 10px;
+  margin-right: 4px;
+  border-radius: 9999px;
+  border: 2px solid rgba(251, 146, 60, 0.7);
+  border-top-color: transparent;
+  animation: mesh-refresh-spin 0.8s linear infinite;
+  vertical-align: -1px;
+}
+@keyframes mesh-refresh-spin { to { transform: rotate(360deg); } }
 .mesh-peers-card { padding: 14px; flex: 1; min-height: 0; display: flex; flex-direction: column; }
 .mesh-peers-card .mesh-section-title { margin-bottom: 10px; flex-shrink: 0; }
 .mesh-peer-list { display: flex; flex-direction: column; gap: 4px; overflow-y: auto; flex: 1; min-height: 0; }
@@ -361,13 +374,35 @@
 .mesh-typed-content-audio { width: 220px; max-width: 100%; display: block; }
 .mesh-typed-content-image-wrap { position: relative; display: inline-block; }
 .mesh-typed-content-download-btn {
-  position: absolute; bottom: 6px; right: 6px; width: 1.75rem; height: 1.75rem;
-  border-radius: 50%; border: 1px solid rgba(255,255,255,0.15);
-  background: rgba(0,0,0,0.55); color: rgba(255,255,255,0.85); font-size: 0.85rem;
+  position: absolute; bottom: 8px; right: 8px;
+  width: 2.25rem; height: 2.25rem; min-width: 2.25rem; flex-shrink: 0;
+  border-radius: 50%; border: 1px solid rgba(255,255,255,0.18);
+  background: rgba(10,10,14,0.55); color: rgba(255,255,255,0.9);
   display: flex; align-items: center; justify-content: center; cursor: pointer;
-  backdrop-filter: blur(6px);
+  backdrop-filter: blur(10px); -webkit-backdrop-filter: blur(10px);
+  box-shadow: 0 2px 8px rgba(0,0,0,0.35);
+  transition: background 0.15s ease, transform 0.15s ease;
 }
-.mesh-typed-content-download-btn:hover { background: rgba(0,0,0,0.75); color: #fff; }
+.mesh-typed-content-download-btn svg { width: 1.05rem; height: 1.05rem; }
+.mesh-typed-content-download-btn:hover { background: rgba(251,146,60,0.35); color: #fff; transform: scale(1.06); }
+.mesh-typed-content-download-btn:active { transform: scale(0.96); }
+/* Pre-fetch "Download" pill under an incoming attachment. The generic .btn it
+   replaced collapsed to its text width inside the narrow mobile bubble and
+   looked squashed — this is a full-width glass pill in the house style. */
+.mesh-typed-content-fetch-btn {
+  display: flex; align-items: center; justify-content: center; gap: 7px;
+  width: 100%; min-height: 2.4rem; padding: 8px 14px; margin-top: 2px;
+  border-radius: 12px; border: 1px solid rgba(255,255,255,0.14);
+  background: rgba(255,255,255,0.07); color: rgba(255,255,255,0.9);
+  font-size: 0.82rem; font-weight: 500; cursor: pointer; white-space: nowrap;
+  backdrop-filter: blur(10px); -webkit-backdrop-filter: blur(10px);
+  transition: background 0.15s ease, border-color 0.15s ease;
+}
+.mesh-typed-content-fetch-btn svg { width: 1rem; height: 1rem; flex-shrink: 0; }
+.mesh-typed-content-fetch-btn:hover:not(:disabled) {
+  background: rgba(251,146,60,0.18); border-color: rgba(251,146,60,0.4); color: #fff;
+}
+.mesh-typed-content-fetch-btn:disabled { opacity: 0.6; cursor: default; }
 .mesh-tab-bar { display: flex; gap: 2px; background: rgba(0,0,0,0.3); border-radius: 10px; padding: 3px; flex-shrink: 0; }
 .mesh-tab { flex: 1; padding: 8px 12px; border: none; background: transparent; color: rgba(255,255,255,0.5); font-size: 0.82rem; font-weight: 500; border-radius: 8px; cursor: pointer; transition: all 0.2s ease; display: flex; align-items: center; justify-content: center; gap: 6px; }
 .mesh-tab:hover { color: rgba(255,255,255,0.8); background: rgba(255,255,255,0.05); }

neode-ui/src/views/server/FipsNetworkCard.vue

@@ -121,6 +121,7 @@
 <script setup lang="ts">
 import { computed, onMounted, onUnmounted, ref } from 'vue'
 import { rpcClient } from '@/api/rpc-client'
+import { useCachedResource } from '@/composables/useCachedResource'
 import { safeClipboardWrite } from '@/views/web5/utils'
 import FipsSeedAnchorsCard from './FipsSeedAnchorsCard.vue'
 
@@ -136,7 +137,14 @@ interface FipsStatus {
   anchor_connected?: boolean
 }
 
-const status = ref<FipsStatus>({
+// Shares `server.fips-summary` with the Local Network card's FIPS row, so
+// both paint from the same cache instantly on revisit and never disagree.
+const statusRes = useCachedResource<FipsStatus>({
+  key: 'server.fips-summary',
+  ttlMs: 15_000,
+  fetcher: (signal) => rpcClient.call<FipsStatus>({ method: 'fips.status', signal, dedup: true, maxRetries: 1 }),
+})
+const status = computed<FipsStatus>(() => statusRes.data.value ?? {
   installed: false,
   version: null,
   service_state: 'unknown',
@@ -199,18 +207,11 @@ function flash(msg: string, isError = false) {
   setTimeout(() => { statusMessage.value = '' }, 6000)
 }
 
-async function loadStatus() {
-  try {
-    status.value = await rpcClient.call<FipsStatus>({ method: 'fips.status' })
-  } catch (e) {
-    if (import.meta.env.DEV) console.warn('fips.status failed', e)
-  }
-}
-
 async function installAndActivate() {
   installing.value = true
   try {
-    status.value = await rpcClient.call<FipsStatus>({ method: 'fips.install' })
+    const next = await rpcClient.call<FipsStatus>({ method: 'fips.install' })
+    statusRes.optimistic(() => next) // confirmed server state, not a guess
     flash('FIPS started')
   } catch (e: unknown) {
     const msg = e instanceof Error ? e.message : String(e)
@@ -235,7 +236,7 @@ async function reconnectAnchor() {
     }>({ method: 'fips.reconnect', timeout: 60_000 })
     // Update the card with the post-reconnect status returned by the
     // backend — avoids an extra status fetch race.
-    status.value = { ...status.value, ...res.after }
+    statusRes.optimistic((cur) => ({ ...(cur ?? status.value), ...res.after }))
     if (res.recovered) {
       flash('Anchor reconnected.')
     } else if (res.likely_cause === 'connected') {
@@ -258,8 +259,10 @@ async function reconnectAnchor() {
 // stuck showing whatever anchor state existed at mount time forever.
 let statusInterval: ReturnType<typeof setInterval> | null = null
 onMounted(() => {
-  loadStatus()
-  statusInterval = setInterval(loadStatus, 15000)
+  statusInterval = setInterval(() => {
+    if (document.hidden) return
+    void statusRes.refresh()
+  }, 15000)
 })
 onUnmounted(() => {
   if (statusInterval) clearInterval(statusInterval)

neode-ui/src/views/server/FipsSeedAnchorsCard.vue

@@ -88,8 +88,9 @@
 </template>
 
 <script setup lang="ts">
-import { onMounted, reactive, ref } from 'vue'
+import { computed, reactive, ref } from 'vue'
 import { rpcClient } from '@/api/rpc-client'
+import { useCachedResource } from '@/composables/useCachedResource'
 
 defineProps<{ closable?: boolean }>()
 defineEmits<{ (e: 'close'): void }>()
@@ -107,7 +108,16 @@ interface ApplyResult {
   message: string
 }
 
-const anchors = ref<SeedAnchor[]>([])
+const anchorsRes = useCachedResource<SeedAnchor[]>({
+  key: 'server.fips-seed-anchors',
+  fetcher: async (signal) => {
+    const res = await rpcClient.call<{ seed_anchors: SeedAnchor[] }>({
+      method: 'fips.list-seed-anchors', signal, dedup: true, maxRetries: 1,
+    })
+    return res.seed_anchors
+  },
+})
+const anchors = computed(() => anchorsRes.data.value ?? [])
 const adding = ref(false)
 const applying = ref(false)
 const statusMessage = ref('')
@@ -126,15 +136,6 @@ function flash(msg: string, isError = false) {
   setTimeout(() => { statusMessage.value = '' }, 6000)
 }
 
-async function load() {
-  try {
-    const res = await rpcClient.call<{ seed_anchors: SeedAnchor[] }>({ method: 'fips.list-seed-anchors' })
-    anchors.value = res.seed_anchors
-  } catch (e: unknown) {
-    if (import.meta.env.DEV) console.warn('fips.list-seed-anchors failed', e)
-  }
-}
-
 async function addAnchor() {
   if (!draft.npub.trim() || !draft.address.trim()) return
   adding.value = true
@@ -148,7 +149,7 @@ async function addAnchor() {
         label: draft.label.trim(),
       },
     })
-    anchors.value = res.seed_anchors
+    anchorsRes.optimistic(() => res.seed_anchors) // authoritative post-add list
     draft.npub = ''
     draft.address = ''
     draft.label = ''
@@ -168,7 +169,7 @@ async function removeAnchor(npub: string) {
       method: 'fips.remove-seed-anchor',
       params: { npub },
     })
-    anchors.value = res.seed_anchors
+    anchorsRes.optimistic(() => res.seed_anchors)
     flash('Anchor removed.')
   } catch (e: unknown) {
     const msg = e instanceof Error ? e.message : String(e)
@@ -189,6 +190,4 @@ async function applyAll() {
     applying.value = false
   }
 }
-
-onMounted(load)
 </script>

reticulum-daemon/build.sh

@@ -47,11 +47,16 @@ if [ -n "$RNODECONF_SRC" ] && [ -f "$RNODECONF_SRC" ]; then
     # exit()/quit() builtins, which only exist in interactive Python (site.py
     # injects them) — a frozen app hits NameError right as it tries to quit
     # cleanly, after all the real work already succeeded. See
-    # pyi_rthook_exit_builtins.py.
+    # pyi_rthook_exit_builtins.py. A second hook fixes rnodeconf's board-flash
+    # step, which shells out to a bundled esptool.py via `sys.executable` —
+    # under a frozen binary that's the binary itself, not a real interpreter,
+    # so the flash subprocess call breaks. See
+    # pyi_rthook_fix_flasher_executable.py.
     .venv/bin/pyinstaller --onefile --name archy-rnodeconf --clean --noconfirm \
         --collect-submodules RNS \
         --collect-data RNS \
         --runtime-hook pyi_rthook_exit_builtins.py \
+        --runtime-hook pyi_rthook_fix_flasher_executable.py \
         -d noarchive \
         "$RNODECONF_SRC"
     echo "Built dist/archy-rnodeconf ($(du -h dist/archy-rnodeconf | cut -f1))"

reticulum-daemon/pyi_rthook_fix_flasher_executable.py

@@ -0,0 +1,34 @@
+# PyInstaller runtime hook — see build.sh.
+#
+# rnodeconf's own board-flashing code shells out to a bundled esptool.py as
+# `[sys.executable, flasher_path, "--chip", ..., "write_flash", ...]` (RNS's
+# rnodeconf.py, ~line 2794 as of RNS 1.3.5). That's correct for a normal
+# `python rnodeconf.py` invocation, but under a frozen PyInstaller binary
+# `sys.executable` is the frozen binary itself, not a real interpreter — so
+# the "subprocess" just re-invokes archy-rnodeconf's OWN argparse CLI with
+# esptool-shaped flags, which it doesn't recognize, and the flash step fails
+# immediately with "unrecognized arguments: --chip ...". Confirmed live
+# against a real Heltec V4 (2026-07-23): device selection, band selection,
+# and firmware download all worked; only the final `write_flash` subprocess
+# call broke this way.
+#
+# Fix: point sys.executable at a real Python interpreter that has rnodeconf's
+# own runtime deps available (esptool.py only needs pyserial, which RNS
+# already depends on) before any of rnodeconf's code runs. Prefer the build
+# venv this exact binary was frozen from — see build.sh — falling back to a
+# bare `python3` on PATH if that venv isn't present on this node.
+import os
+import sys
+
+if getattr(sys, "frozen", False):
+    _candidates = [
+        os.environ.get("ARCHY_RNODECONF_PYTHON", ""),
+        os.path.join(os.path.dirname(os.path.abspath(sys.argv[0])), "..", "reticulum-daemon", ".venv", "bin", "python3"),
+        os.path.expanduser("~/archy/reticulum-daemon/.venv/bin/python3"),
+    ]
+    for _candidate in _candidates:
+        if _candidate and os.path.isfile(_candidate):
+            sys.executable = _candidate
+            break
+    else:
+        sys.executable = "python3"

reticulum-daemon/reticulum_daemon.py

@@ -14,12 +14,13 @@ Security posture (see the plan's "most secure way" section):
 RPC (one JSON object per line, both directions):
   in : {"cmd":"send","dest_hash":"<hex16>","content":"…","title":"…","method":"direct|opportunistic"}
        {"cmd":"announce"}
+       {"cmd":"set_name","name":"…"}
        {"cmd":"status"}
        {"cmd":"send_resource","id":"<correlation>","dest_hash":"<hex16>","data_b64":"…"}
        {"cmd":"shutdown"}
   out: {"event":"ready","dest_hash":"<hex16>","display_name":"…"}
        {"event":"recv","source_hash":"<hex16>","content":"…","title":"…","fields":{…},"app_data":"<hex>","rssi":n,"snr":n,"stamp":t}
-       {"event":"announce","dest_hash":"<hex16>","app_data":"<hex>"}
+       {"event":"announce","dest_hash":"<hex16>","app_data":"<hex>","display_name":"…"|null,"archy_blob":"ARCHY:2:…"|null}
        {"event":"delivered","dest_hash":"<hex16>","state":"delivered|failed","id":"<hex>"}
        {"event":"status","connected":bool,"dest_hash":"<hex16>","interfaces":[…]}
        {"event":"resource_progress","id":"<correlation>","transferred":n,"total":n}
@@ -227,31 +228,47 @@ class ReticulumDaemon:
         if self.delivery_destination is not None:
             self.delivery_destination.announce(app_data=self._announce_app_data())
 
-    def _announce_app_data(self) -> bytes:
-        """Carry the Archy identity so peers bind this RNS destination onto the
-        existing contact, the same way a meshcore/Meshtastic identity advert does.
-
-        Reuses the exact ``ARCHY:2:{ed25519_hex}:{x25519_hex}`` wire format the
-        Rust side already parses (``protocol::parse_identity_broadcast``) and
-        binds via ``handle_identity_received``/``bind_federation_twins`` — so a
-        Reticulum-carried identity merges into the SAME conversation as the
-        meshcore/Meshtastic/federation twins of the same Archy node, satisfying
-        cross-protocol DM convergence. The keys are the node's real Archipelago
-        ed25519/x25519 pubkeys (passed in by the Rust side, which already has
-        them) — NOT this daemon's internally-HKDF-derived RNS keys, which exist
-        only to make the RNS destination hash deterministic and are never
-        themselves treated as an Archy identity.
-
-        Falls back to a plain display-name string (undetected as an identity
-        blob — no `ARCHY:2:` prefix) if the Archy pubkeys weren't supplied, e.g.
-        a dev/selftest run with no `--archy-ed-pubkey-hex`.
-        """
+    def _archy_identity_blob(self):
+        """The ``ARCHY:2:{ed25519_hex}:{x25519_hex}`` identity string the Rust
+        side parses (``protocol::parse_identity_broadcast``) and binds via
+        ``handle_identity_received`` — so a Reticulum-carried identity merges
+        into the SAME conversation as the meshcore/Meshtastic/federation twins
+        of the same Archy node. The keys are the node's real Archipelago
+        pubkeys (passed in by the Rust side) — NOT this daemon's
+        internally-HKDF-derived RNS keys, which exist only to make the RNS
+        destination hash deterministic. ``None`` when the pubkeys weren't
+        supplied (dev/selftest run)."""
         if self.args.archy_ed_pubkey_hex and self.args.archy_x25519_pubkey_hex:
             return (
                 f"ARCHY:2:{self.args.archy_ed_pubkey_hex}:"
                 f"{self.args.archy_x25519_pubkey_hex}"
             ).encode("ascii")
-        return (self.args.display_name or "").encode("utf-8")
+        return None
+
+    def _announce_app_data(self) -> bytes:
+        """LXMF-standard announce app_data — msgpack ``[display_name,
+        stamp_cost, supported_functionality]`` via the router, so Sideband/
+        NomadNet/MeshChat (and upgraded archy nodes) all see our real display
+        name — with the Archy identity blob appended as an EXTRA list element.
+        Stock clients only read the elements they know ([0]/[1]), so the blob
+        rides along invisibly instead of replacing the name the way the old
+        blob-only app_data did (which left every archy node nameless on RNS).
+        """
+        import RNS.vendor.umsgpack as msgpack
+        app_data = self.router.get_announce_app_data(self.delivery_destination.hash)
+        blob = self._archy_identity_blob()
+        if blob is None:
+            return app_data
+        try:
+            peer_data = msgpack.unpackb(app_data)
+            if not isinstance(peer_data, list):
+                raise ValueError("unexpected announce app_data shape")
+            peer_data.append(blob)
+            return msgpack.packb(peer_data)
+        except Exception:
+            # Never let announce formatting kill announcing entirely — fall
+            # back to the legacy blob-only format (identity binding > name).
+            return blob
 
     # ---- RNS-thread callbacks → asyncio ----
     def _on_lxmf_delivery(self, message):
@@ -328,6 +345,15 @@ class ReticulumDaemon:
             self._send(req)
         elif cmd == "announce":
             self.announce()
+        elif cmd == "set_name":
+            # Live rename: update the LXMF delivery destination's display name
+            # (what get_announce_app_data reads) and re-announce immediately so
+            # peers learn the new name without waiting for the next advert tick.
+            name = (req.get("name") or "").strip()
+            if name and self.delivery_destination is not None:
+                self.args.display_name = name
+                self.delivery_destination.display_name = name
+                self.announce()
         elif cmd == "status":
             self._broadcast(self._status())
         elif cmd == "send_resource":
@@ -521,10 +547,41 @@ class _AnnounceHandler:
         self.receive_path_responses = True
 
     def received_announce(self, destination_hash, announced_identity, app_data):
+        # Decode what we can here (both the LXMF-standard display name and our
+        # appended ARCHY identity blob — see _announce_app_data) so the Rust
+        # side gets clean typed fields instead of re-implementing msgpack.
+        display_name = None
+        archy_blob = None
+        raw = app_data or b""
+        try:
+            import LXMF
+            display_name = LXMF.display_name_from_app_data(raw)
+            # A legacy blob-only announce is plain ascii, so LXMF's decoder
+            # returns the whole ARCHY identity blob as a "name" — drop it.
+            if display_name and display_name.startswith("ARCHY:"):
+                display_name = None
+        except Exception:
+            display_name = None
+        try:
+            if raw[:1] and ((0x90 <= raw[0] <= 0x9F) or raw[0] == 0xDC):
+                import RNS.vendor.umsgpack as msgpack
+                peer_data = msgpack.unpackb(raw)
+                if isinstance(peer_data, list):
+                    for el in peer_data[3:]:
+                        if isinstance(el, bytes) and el.startswith(b"ARCHY:"):
+                            archy_blob = el.decode("ascii", "ignore")
+                            break
+            elif raw.startswith(b"ARCHY:"):
+                # Legacy (pre-upgrade archy node): app_data IS the blob.
+                archy_blob = raw.decode("ascii", "ignore")
+        except Exception:
+            archy_blob = None
         self.daemon._emit_threadsafe({
             "event": "announce",
             "dest_hash": destination_hash.hex(),
-            "app_data": (app_data or b"").hex(),
+            "app_data": raw.hex(),
+            "display_name": display_name,
+            "archy_blob": archy_blob,
         })
 
 
@@ -604,8 +661,30 @@ def main(argv=None) -> int:
     if args.selftest:
         args.no_radio = True
         daemon.bring_up()
+        # Announce app_data round-trip: the LXMF-standard msgpack name must be
+        # decodable by stock clients AND (with archy keys set) the appended
+        # identity blob must survive as an extra list element — this is the
+        # exact wire contract the Rust announce handler and Sideband both
+        # depend on, so verify it here where there's a real router to build it.
+        import LXMF as _LXMF
+        import RNS.vendor.umsgpack as _msgpack
+        args.archy_ed_pubkey_hex = args.archy_ed_pubkey_hex or "ab" * 32
+        args.archy_x25519_pubkey_hex = args.archy_x25519_pubkey_hex or "cd" * 32
+        app_data = daemon._announce_app_data()
+        decoded_name = _LXMF.display_name_from_app_data(app_data)
+        assert decoded_name == args.display_name, (
+            f"announce name round-trip failed: {decoded_name!r} != {args.display_name!r}"
+        )
+        blob_elems = [e for e in _msgpack.unpackb(app_data)[3:]
+                      if isinstance(e, bytes) and e.startswith(b"ARCHY:")]
+        assert blob_elems, "identity blob missing from announce app_data"
+        # Live rename: set_name must change what the next announce carries.
+        daemon.delivery_destination.display_name = "selftest-renamed"
+        renamed = _LXMF.display_name_from_app_data(daemon._announce_app_data())
+        assert renamed == "selftest-renamed", f"rename round-trip failed: {renamed!r}"
         print(f"selftest ok — dest_hash={daemon.dest_hash_hex} "
-              f"display_name={args.display_name!r} lxmf_router=up")
+              f"display_name={args.display_name!r} lxmf_router=up "
+              f"announce_app_data=verified set_name=verified")
         return 0
 
     for sig in (signal.SIGINT, signal.SIGTERM):

scripts/self-update.sh

@@ -84,6 +84,64 @@ if ! command -v nano >/dev/null 2>&1; then
     fi
 fi
 
+if ! command -v ping >/dev/null 2>&1; then
+    log "Installing iputils-ping..."
+    if sudo apt-get update -qq 2>>"$LOG_FILE" && sudo apt-get install -y -qq iputils-ping 2>>"$LOG_FILE"; then
+        ok "ping installed"
+    else
+        warn "Unable to install ping automatically; continuing update"
+    fi
+fi
+
+if ! command -v esptool >/dev/null 2>&1; then
+    log "Installing esptool for LoRa radio firmware flashing..."
+    if sudo apt-get update -qq 2>>"$LOG_FILE" && sudo apt-get install -y -qq esptool 2>>"$LOG_FILE"; then
+        ok "esptool installed"
+    else
+        warn "Unable to install esptool automatically; radio firmware flashing will be unavailable"
+    fi
+fi
+
+# Debian's esptool package (4.7.0+dfsg-0.1) ships without the precompiled
+# esp32s3 "stub flasher" blob (stripped for DFSG compliance — no
+# buildable-from-source path Debian could verify). Without it, esptool's
+# normal stub-loader mode fails outright (FileNotFoundError), and the ROM
+# bootloader fallback (--no-stub) doesn't implement a full-chip erase at
+# all — confirmed live 2026-07-23 flashing a real Heltec V4, both ways.
+# Fetching the exact same file from the matching upstream esptool release
+# tag restores full (and correct) flashing behavior — it's the same
+# open-source codebase, just the one blob Debian's packaging couldn't
+# include.
+if command -v esptool >/dev/null 2>&1; then
+    STUB_DIR="/usr/lib/python3/dist-packages/esptool/targets/stub_flasher"
+    STUB_FILE="$STUB_DIR/stub_flasher_32s3.json"
+    if [ ! -f "$STUB_FILE" ]; then
+        log "Fetching esptool's esp32s3 stub flasher (missing from the Debian package)..."
+        ESPTOOL_VERSION=$(esptool version 2>/dev/null | tail -1 | tr -d ' \t')
+        if [ -n "$ESPTOOL_VERSION" ] && sudo curl -fsSL -o "$STUB_FILE" \
+            "https://raw.githubusercontent.com/espressif/esptool/v${ESPTOOL_VERSION}/esptool/targets/stub_flasher/stub_flasher_32s3.json" \
+            2>>"$LOG_FILE"; then
+            sudo chmod 644 "$STUB_FILE"
+            ok "esp32s3 stub flasher installed"
+        else
+            sudo rm -f "$STUB_FILE" 2>/dev/null
+            warn "Unable to fetch esp32s3 stub flasher; LoRa firmware flashing will be unavailable"
+        fi
+    fi
+fi
+
+# Build-time prerequisites for reticulum-daemon/build.sh's PyInstaller step
+# below (discovered the hard way: ensurepip needs python3-venv, and
+# PyInstaller itself needs objdump + libpython3.13.so at build time — none
+# of these are pulled in by a bare `python3` package on Debian trixie).
+for pkg in python3-venv binutils libpython3.13; do
+    if ! dpkg -s "$pkg" >/dev/null 2>&1; then
+        log "Installing $pkg (reticulum-daemon build prerequisite)..."
+        sudo apt-get update -qq 2>>"$LOG_FILE" && sudo apt-get install -y -qq "$pkg" 2>>"$LOG_FILE" \
+            || warn "Unable to install $pkg automatically; reticulum-daemon tools build may fail"
+    fi
+done
+
 # Fetch latest
 log "Fetching from origin..."
 git fetch origin main --quiet 2>>"$LOG_FILE"
@@ -155,6 +213,30 @@ sudo cp "$BUILT_BIN" "$INSTALL_BIN"
 sudo chmod +x "$INSTALL_BIN"
 ok "Backend installed"
 
+# Build + install reticulum-daemon tools (archy-reticulum-daemon, archy-rnodeconf).
+# Non-fatal: archipelago falls back to its dev venv path if the packaged
+# binaries aren't present, so a missing/failed build here degrades mesh
+# Reticulum support rather than breaking the update. This mirrors
+# deploy-to-target.sh's existing manual-deploy step, which until now was the
+# only path that ever installed these — a node that only ever received OTA
+# self-updates had neither binary.
+if [ -f "$REPO_DIR/reticulum-daemon/build.sh" ]; then
+    log "Building reticulum-daemon tools (archy-reticulum-daemon, archy-rnodeconf)..."
+    if (cd "$REPO_DIR/reticulum-daemon" && ./build.sh) 2>>"$LOG_FILE"; then
+        for tool in archy-reticulum-daemon archy-rnodeconf; do
+            if [ -f "$REPO_DIR/reticulum-daemon/dist/$tool" ]; then
+                sudo cp "$REPO_DIR/reticulum-daemon/dist/$tool" /usr/local/bin/
+                sudo chmod +x "/usr/local/bin/$tool"
+                ok "$tool installed"
+            else
+                warn "$tool not built — leaving existing /usr/local/bin/$tool (if any) in place"
+            fi
+        done
+    else
+        warn "reticulum-daemon tools build failed — continuing without updating them"
+    fi
+fi
+
 # Build frontend
 log "Building Vue frontend (production)..."
 cd "$FRONTEND_DIR"

tests/mesh/run-mesh-tests.sh

@@ -0,0 +1,85 @@
+#!/bin/bash
+# Mesh / Reticulum test suite — the "is the mesh stack healthy" gate.
+#
+# Three layers, cheapest first:
+#   1. Rust unit tests        (no hardware, ~2s once built)
+#   2. Daemon selftest        (full RNS+LXMF bring-up, no radio; also verifies
+#                              the announce app_data wire contract + set_name)
+#   3. Live-node assertions   (optional; needs a running archipelago with a
+#                              radio — set MESH_TEST_LIVE=1 MESH_TEST_PW=...)
+#
+# Usage:
+#   tests/mesh/run-mesh-tests.sh                # layers 1+2
+#   MESH_TEST_LIVE=1 MESH_TEST_PW='...' tests/mesh/run-mesh-tests.sh
+#   MESH_TEST_HOST=100.113.100.55 ...           # live-test a remote node
+set -u
+cd "$(dirname "$0")/../.."
+FAIL=0
+ok()   { echo "ok  - $1"; }
+bad()  { echo "not ok - $1"; FAIL=1; }
+
+# ── 1. Rust unit tests ────────────────────────────────────────────────
+RUST_RESULTS=$(cd core && cargo test -p archipelago --bin archipelago mesh 2>&1 | grep "^test result:")
+if [ -n "$RUST_RESULTS" ] && ! echo "$RUST_RESULTS" | grep -vq " 0 failed"; then
+    ok "rust mesh unit tests ($(echo "$RUST_RESULTS" | grep -o '[0-9]* passed' | head -1))"
+else
+    bad "rust mesh unit tests"
+fi
+
+# ── 2. Reticulum daemon selftest (no radio) ───────────────────────────
+TMP=$(mktemp -d)
+trap 'rm -rf "$TMP"' EXIT
+head -c 32 /dev/urandom > "$TMP/key"
+DAEMON=reticulum-daemon/.venv/bin/python
+if [ -x "$DAEMON" ]; then
+    if "$DAEMON" reticulum-daemon/reticulum_daemon.py \
+        --identity-key "$TMP/key" --rns-config "$TMP/rns" \
+        --socket "$TMP/sock" --display-name "SelftestNode" --selftest 2>/dev/null \
+        | grep -q "announce_app_data=verified set_name=verified"; then
+        ok "daemon selftest (announce wire contract + set_name)"
+    else
+        bad "daemon selftest"
+    fi
+else
+    echo "skip - daemon selftest (no venv at $DAEMON)"
+fi
+
+# ── 3. Live node assertions (opt-in) ──────────────────────────────────
+if [ "${MESH_TEST_LIVE:-0}" = "1" ]; then
+    HOST="${MESH_TEST_HOST:-127.0.0.1}"
+    PW="${MESH_TEST_PW:?set MESH_TEST_PW}"
+    # Nodes differ: dev boxes serve plain http on :80, ISO installs https.
+    RPC=""
+    for base in "http://$HOST" "https://$HOST" "http://$HOST:5678"; do
+        code=$(curl -ksS -o /dev/null -w '%{http_code}' -m 5 -X POST "$base/rpc/v1" 2>/dev/null || true)
+        case "$code" in 000|"") continue ;; *) RPC="$base/rpc/v1"; break ;; esac
+    done
+    [ -n "$RPC" ] || { bad "live: no RPC endpoint reachable on $HOST"; echo FAIL; exit 1; }
+    JAR="$TMP/jar"
+    curl -ksS -c "$JAR" -H "Content-Type: application/json" \
+        -d "{\"jsonrpc\":\"2.0\",\"method\":\"auth.login\",\"params\":{\"password\":\"$PW\"},\"id\":1}" \
+        "$RPC" > "$TMP/login"
+    if grep -q '"error":null' "$TMP/login"; then ok "live: rpc login"; else bad "live: rpc login"; fi
+    call() {
+        local csrf; csrf=$(awk '/^[^#]/ && /csrf_token/ {print $7; exit}' "$JAR")
+        curl -ksS -b "$JAR" -c "$JAR" -H "Content-Type: application/json" \
+            -H "X-CSRF-Token: $csrf" \
+            -d "{\"jsonrpc\":\"2.0\",\"method\":\"$1\",\"params\":${2:-{\}},\"id\":2}" \
+            --max-time 60 "$RPC"
+    }
+    ST=$(call mesh.status)
+    echo "$ST" | grep -q '"device_connected":true' \
+        && ok "live: radio connected ($(echo "$ST" | grep -o '"device_type":"[a-z]*"'))" \
+        || bad "live: radio connected"
+    echo "$ST" | grep -q '"self_advert_name":"[^"]' \
+        && ok "live: node has a mesh name" || bad "live: node has a mesh name"
+    call mesh.refresh | grep -q '"refreshed":true' \
+        && ok "live: mesh.refresh" || bad "live: mesh.refresh"
+    call mesh.broadcast | grep -q '"broadcast":true' \
+        && ok "live: mesh.broadcast" || bad "live: mesh.broadcast"
+    # No peer may ever display a raw identity blob as its name.
+    call mesh.peers | grep -q '"advert_name":"ARCHY:' \
+        && bad "live: no ARCHY-blob peer names" || ok "live: no ARCHY-blob peer names"
+fi
+
+[ "$FAIL" = 0 ] && echo "PASS" || { echo "FAIL"; exit 1; }