feat(rumble): host-authoritative self-terminating envelopes (0xCA v2)

Rumble was level-triggered, unbounded state on a lossy channel: a non-zero
level meant "buzz until further notice", healed only by the host re-sending
state every 500 ms, and every client guessed when the host had died with its own
magic timeout (SDL 1.5 s, Apple 1.6 s, Android up to 60 s). A lost stop, a
reordered start, or a dead host could drone the motor for seconds.

Make "stuck rumble" inexpressible on the wire. The 0xCA datagram grows a
length-tolerant tail — [u8 seq][u16 ttl_ms] — so it self-terminates: the host
authorizes a level for at most ttl_ms and renews it (~120 ms) while it holds,
letting an abandoned one lapse client-side. seq is a per-pad wrapping reorder
gate (reusing GamepadSnapshot::seq_newer) so a reordered stale start can't
re-light a stopped motor. Decoders read the first 7 bytes as a plain level and
ignore the tail, so no wire-version bump: an old client renders a new host's
levels, and a new client falls back to its prior staleness heuristic against an
old host (ttl = None). All four generation pairings render correctly.

- core: encode_rumble_datagram_v2 / decode_rumble_envelope (datagram.rs); the
  client demux applies the seq gate then forwards (pad, low, high, Option<ttl>);
  next_rumble is unchanged (drops ttl), next_rumble_ttl keeps it; ABI adds
  punktfunk_connection_next_rumble2 + PUNKTFUNK_RUMBLE_NO_TTL, ABI_VERSION 4->5
  (WIRE_VERSION unchanged — the tail is backward-compatible).
- host (punktfunk1.rs): the flat 500 ms refresh becomes a renewal loop that bumps
  seq + stamps a fresh TTL on active pads and drains a short post-stop zero burst,
  then goes quiet. Hatches: PUNKTFUNK_RUMBLE_ENVELOPE=0 (legacy v1 + flat refresh,
  a bisect switch), PUNKTFUNK_RUMBLE_TTL_MS (clamped [150, 5000]).
- renderers honor the TTL as their playback duration/deadline and keep their old
  heuristic only for a legacy (ttl=None) update: pf-client-core (the Deck haptic
  keep-alive is now deadline-bounded so it can't sustain a host-stopped rumble),
  clients/windows (SDL duration), android (JNI packs the lease out-of-band in bit
  48 so any u16 ttl is unambiguous; Kotlin createOneShot(ttl)), apple
  (RumbleRenderer.envelopeDeadline + nextRumble2; sessionStaleSeconds demoted to
  the legacy fallback).
- tests: codec round-trip + tail tolerance + seq-gate reorder (Rust); the probe
  asserts the v2 tail arrived under PUNKTFUNK_TEST_FEEDBACK; the Apple loopback
  asserts ttlMs round-trips end to end; RumbleTuning lease-decision cases.

The host-side idle-timeout from the previous commit is defense in depth on the
game side; this is the guarantee on the client side. Design:
punktfunk-planning/design/rumble-envelope-plan.md.

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

clients/apple/Sources/PunktfunkKit/Connection/PunktfunkConnection.swift

@@ -570,7 +570,8 @@ public final class PunktfunkConnection {
 
     /// Pull the next force-feedback update for the GCController haptics engine:
     /// `(pad, lowFrequency, highFrequency)` with 0...0xFFFF amplitudes, (0, 0) = stop.
-    /// Drain from the (single) feedback thread, alongside `nextHidOutput`.
+    /// Drain from the (single) feedback thread, alongside `nextHidOutput`. Drops the v2
+    /// self-termination TTL — use `nextRumble2` to honor the host lease.
     public func nextRumble(timeoutMs: UInt32 = 0) throws -> (pad: UInt16, low: UInt16, high: UInt16)? {
         feedbackLock.lock()
         defer { feedbackLock.unlock() }
@@ -590,6 +591,33 @@ public final class PunktfunkConnection {
         }
     }
 
+    /// Pull the next force-feedback update *including its self-termination TTL* (v2 envelopes):
+    /// `(pad, low, high, ttlMs)`. `ttlMs` is how long to render this level before silencing unless
+    /// the host renews it; `RumbleTuning.noTTL` (`UInt32.max`) means "no lease" — a legacy host, so
+    /// fall back to a client-side staleness timeout. The reorder gate (seq) already ran in the
+    /// core, so a stale/reordered envelope never surfaces here. Drain from the (single) feedback
+    /// thread, alongside `nextHidOutput`.
+    public func nextRumble2(timeoutMs: UInt32 = 0) throws
+        -> (pad: UInt16, low: UInt16, high: UInt16, ttlMs: UInt32)?
+    {
+        feedbackLock.lock()
+        defer { feedbackLock.unlock() }
+        guard let h = liveHandle() else { throw PunktfunkClientError.closed }
+
+        var pad: UInt16 = 0, low: UInt16 = 0, high: UInt16 = 0, ttl: UInt32 = .max
+        let rc = punktfunk_connection_next_rumble2(h, &pad, &low, &high, &ttl, timeoutMs)
+        switch rc {
+        case statusOK:
+            return (pad, low, high, ttl)
+        case statusNoFrame:
+            return nil
+        case statusClosed:
+            throw PunktfunkClientError.closed
+        default:
+            throw PunktfunkClientError.status(rc)
+        }
+    }
+
     /// One DualSense feedback event a game wrote to the host's virtual pad — replay it on
     /// the real controller (GCDeviceLight, GCControllerPlayerIndex,
     /// GCDualSenseAdaptiveTrigger). Only a `.dualSense` session emits these.

clients/apple/Sources/PunktfunkKit/Gamepad/GamepadFeedback.swift

@@ -92,15 +92,15 @@ public final class GamepadFeedback {
                     // drain: levels rendered up to ~130 ms late through the core's 16-deep queue,
                     // and its drop-newest overflow could shed a stop while stale nonzero states
                     // queued ahead of it — buzzing until the host's next 500 ms refresh.
-                    var newest: (low: UInt16, high: UInt16)?
+                    var newest: (low: UInt16, high: UInt16, ttl: UInt32)?
                     var rumbleBurst = 0
                     while rumbleBurst < 64, !flag.isStopped,
-                          let r = try connection.nextRumble(timeoutMs: 0) {
-                        if r.pad == 0 { newest = (r.low, r.high) }
+                          let r = try connection.nextRumble2(timeoutMs: 0) {
+                        if r.pad == 0 { newest = (r.low, r.high, r.ttlMs) }
                         rumbleBurst += 1
                     }
                     if let n = newest {
-                        self?.rumble.apply(low: n.low, high: n.high)
+                        self?.rumble.apply(low: n.low, high: n.high, ttlMs: n.ttl)
                     }
                     // Drain a BOUNDED burst of hidout events so sustained 0xCD traffic (a game writing
                     // per-frame LED/trigger reports) can't spin here or block stop() past one cycle.

clients/apple/Sources/PunktfunkKit/Gamepad/RumbleRenderer.swift

@@ -23,10 +23,23 @@ enum RumbleTuning {
     /// the churn that lost stops inside CoreHaptics. Newest level wins when the window opens;
     /// zero is never throttled.
     static let minRebakeSeconds: TimeInterval = 0.025
-    /// Session watchdog: silence the motors when no wire command arrived for this long. The
-    /// host re-sends the current rumble state every 500 ms as its loss heal, so this trips only
-    /// after 3 consecutive refreshes vanished — i.e. the channel or host died while audible.
+    /// Session watchdog: silence the motors when no wire command arrived for this long. This is
+    /// the **legacy-host fallback only** — an old host sends no self-termination lease, so its
+    /// periodic re-send (every 500 ms) is the sole liveness signal and 3 vanished refreshes means
+    /// the channel or host died while audible. A v2 host instead supplies a per-command TTL (see
+    /// [`leaseSeconds`]); that deadline supersedes this watchdog.
     static let sessionStaleSeconds: TimeInterval = 1.6
+
+    /// The legacy no-lease sentinel a v2 `ttl_ms` carries for an old host (mirrors the C ABI's
+    /// `PUNKTFUNK_RUMBLE_NO_TTL`). `UInt32.max` by construction.
+    static let noTTL: UInt32 = .max
+
+    /// Interpret a wire TTL (ms) from a rumble update: `nil` for the legacy no-lease sentinel
+    /// ([`noTTL`]) — the renderer falls back to [`sessionStaleSeconds`] — else the self-termination
+    /// lease in seconds (render the level for at most this long unless the host renews it).
+    static func leaseSeconds(ttlMs: UInt32) -> TimeInterval? {
+        ttlMs == noTTL ? nil : TimeInterval(ttlMs) / 1000
+    }
     /// Levels closer than this (≈0.4 % of full scale) are the same level — an identical host
     /// refresh must never rebuild a player.
     static let levelEpsilon: Float = 1.0 / 256.0
@@ -139,6 +152,10 @@ final class RumbleRenderer: @unchecked Sendable {
     /// Wire-truth target (raw wire units) and when it was last confirmed by any command.
     private var target: (low: UInt16, high: UInt16) = (0, 0)
     private var lastCommand = DispatchTime(uptimeNanoseconds: 0)
+    /// The v2 envelope lease: the active level is authorized until here unless the host renews it
+    /// (`tick` silences at the deadline). `nil` against a legacy host (no lease — the
+    /// `sessionStaleSeconds` watchdog is the backstop) and while silent.
+    private var envelopeDeadline: DispatchTime?
     /// Runs while anything is (or should be) audible: staleness watchdog, segment re-arm,
     /// throttled-level catch-up, engine rebuild after a reset, HID keepalive. Nil while silent,
     /// so an idle controller costs no timer wakeups and no radio traffic.
@@ -212,13 +229,23 @@ final class RumbleRenderer: @unchecked Sendable {
         }
     }
 
-    /// Set the wire-truth target. Called with every 0xCA state the host sends — level changes
-    /// AND the 500 ms refreshes; refreshes stamp liveness for the watchdog and are otherwise
-    /// free (invariant 2).
-    func apply(low lowAmp: UInt16, high highAmp: UInt16) {
+    /// Set the wire-truth target. Called with every 0xCA state the host sends — level changes AND
+    /// renewals (v2) / 500 ms refreshes (legacy); both stamp liveness and, for v2, refresh the
+    /// self-termination deadline. `ttlMs` is the envelope lease in ms, or [`RumbleTuning.noTTL`]
+    /// against a legacy host (no lease → the staleness watchdog is the backstop). Renewals at an
+    /// unchanged level extend the deadline before the idempotence guard, so a held rumble never
+    /// lapses mid-effect.
+    func apply(low lowAmp: UInt16, high highAmp: UInt16, ttlMs: UInt32 = RumbleTuning.noTTL) {
         queue.async {
             self.lastCommand = .now()
             let active = lowAmp != 0 || highAmp != 0
+            // v2 lease: a nonzero level gets an explicit deadline; a stop or a legacy update clears
+            // it. Set BEFORE the idempotence guard so an identical renewal still extends the lease.
+            if let lease = RumbleTuning.leaseSeconds(ttlMs: ttlMs), active {
+                self.envelopeDeadline = .now() + lease
+            } else {
+                self.envelopeDeadline = nil
+            }
             if active != self.wasActive {
                 self.wasActive = active
                 log.debug(
@@ -236,6 +263,7 @@ final class RumbleRenderer: @unchecked Sendable {
             self.ticker?.cancel()
             self.ticker = nil
             self.target = (0, 0)
+            self.envelopeDeadline = nil
             self.wasActive = false
             self.teardown()
             self.closeHID()
@@ -293,9 +321,18 @@ final class RumbleRenderer: @unchecked Sendable {
 
     /// Watchdog + housekeeping heartbeat while audible.
     private func tick() {
-        if let after = policy.staleAfter, target != (0, 0), seconds(since: lastCommand) > after {
-            // The host refreshes rumble state every 500 ms; this much silence means the channel
-            // (or host) died while a motor was on. A direct-connected pad would have been
+        if let deadline = envelopeDeadline {
+            // v2 host lease: silence the moment it lapses unrenewed. This firing in the wild is the
+            // observable signature of a host that stopped renewing (a dropped stop, or a dead host)
+            // — the whole point of the envelope model: the motor can't outlive the host's intent.
+            if target != (0, 0), DispatchTime.now() >= deadline {
+                log.warning("rumble: envelope expired unrenewed — silencing")
+                target = (0, 0)
+                envelopeDeadline = nil
+            }
+        } else if let after = policy.staleAfter, target != (0, 0), seconds(since: lastCommand) > after {
+            // Legacy host (no lease): it re-sends state every 500 ms, so this much silence means the
+            // channel (or host) died while a motor was on. A direct-connected pad would have been
             // stopped by its game long ago — force the same outcome.
             log.warning(
                 "rumble: no wire refresh for \(after, format: .fixed(precision: 1), privacy: .public)s — auto-silencing")

clients/apple/Tests/PunktfunkKitTests/LoopbackIntegrationTests.swift

@@ -88,16 +88,19 @@ final class LoopbackIntegrationTests: XCTestCase {
         // one feedback burst on the host→client planes — drain both and verify, end to
         // end through the xcframework: rumble (0xCA) + the three hidout kinds (0xCD).
         if ProcessInfo.processInfo.environment["PUNKTFUNK_TEST_FEEDBACK"] == "1" {
-            var rumble: (pad: UInt16, low: UInt16, high: UInt16)?
+            var rumble: (pad: UInt16, low: UInt16, high: UInt16, ttlMs: UInt32)?
             var hidout: [PunktfunkConnection.HidOutputEvent] = []
             let feedbackDeadline = Date().addingTimeInterval(10)
             while (rumble == nil || hidout.count < 3), Date() < feedbackDeadline {
-                if rumble == nil, let r = try conn.nextRumble(timeoutMs: 100) { rumble = r }
+                if rumble == nil, let r = try conn.nextRumble2(timeoutMs: 100) { rumble = r }
                 if let ev = try conn.nextHidOutput(timeoutMs: 100) { hidout.append(ev) }
             }
             XCTAssertEqual(rumble?.pad, 0)
             XCTAssertEqual(rumble?.low, 0x4000)
             XCTAssertEqual(rumble?.high, 0x8000)
+            // The synthetic host emits a v2 envelope (400 ms TTL) — assert the self-terminating tail
+            // survived the full wire → C ABI → Swift path, not just the level.
+            XCTAssertEqual(rumble?.ttlMs, 400)
             XCTAssertTrue(
                 hidout.contains(.led(pad: 0, r: 10, g: 20, b: 30)),
                 "missing the scripted lightbar event: \(hidout)")

clients/apple/Tests/PunktfunkKitTests/RumbleTuningTests.swift

@@ -75,6 +75,40 @@ final class RumbleTuningTests: XCTestCase {
         renderer.stop()
     }
 
+    func testLeaseSecondsInterpretsWireTTL() {
+        // The legacy no-lease sentinel → nil (fall back to the staleness watchdog).
+        XCTAssertNil(RumbleTuning.leaseSeconds(ttlMs: RumbleTuning.noTTL))
+        XCTAssertEqual(RumbleTuning.noTTL, UInt32.max)
+        // A real lease → its duration in seconds.
+        XCTAssertEqual(RumbleTuning.leaseSeconds(ttlMs: 400), 0.4, accuracy: 1e-9)
+        XCTAssertEqual(RumbleTuning.leaseSeconds(ttlMs: 0), 0, accuracy: 1e-9)
+        XCTAssertEqual(RumbleTuning.leaseSeconds(ttlMs: 150), 0.15, accuracy: 1e-9)
+    }
+
+    func testEnvelopeLeaseBoundsMotorLifeTighterThanTheLegacyWatchdog() {
+        // The whole point of v2: a host-supplied lease silences the motor faster than the
+        // legacy staleness watchdog ever could (which needs sessionStaleSeconds of silence). The
+        // default 400 ms TTL is well under that, on every platform.
+        let defaultTTL = RumbleTuning.leaseSeconds(ttlMs: 400)
+        XCTAssertNotNil(defaultTTL)
+        XCTAssertLessThan(defaultTTL!, RumbleTuning.sessionStaleSeconds)
+        // The ticker must be able to observe an expired lease promptly (well within one TTL).
+        XCTAssertLessThan(RumbleTuning.tickSeconds, defaultTTL!)
+    }
+
+    /// A v2 envelope with a short TTL, left unrenewed, must self-silence — the renderer's core
+    /// promise. Drive the real queue/ticker (no physical pad) and confirm it doesn't wedge.
+    func testEnvelopeExpiresWhenUnrenewed() {
+        let renderer = RumbleRenderer(policy: .session)
+        renderer.retarget(nil)
+        // A 100 ms lease, then no renewal — the ticker (50 ms) must silence it on its own.
+        renderer.apply(low: 0x8000, high: 0x8000, ttlMs: 100)
+        Thread.sleep(forTimeInterval: 0.3)
+        // No assertion on private state; this exercises the expiry path + serial-queue teardown
+        // without deadlock (the ticker fires on the same queue stop() sync-hops onto).
+        renderer.stop()
+    }
+
     func testTuningRelationsTheDesignDependsOn() {
         // The watchdog must tolerate a couple of lost 500 ms host refreshes (heals, not gaps)
         // but trip well before a stuck rumble reads as "still going".