I am reporting a reproducible performance issue in iOS 18.6 where synchronizing the scroll position of two Chart views via chartScrollPosition(id:) causes a complete redraw loop when custom AxisMarks are used. This occurs even when the axis marks are technically "hidden," leading to significant frame drops and stuttering on modern hardware like the iPhone 15. Environment Device: iPhone 15 OS: iOS 18.6 (22G86) Frameworks: SwiftUI, Swift Charts, Observation The Issue When using a shared @Observable state to sync two charts, the scrolling is fluid only if the axes are at their default settings. As soon as a custom AxisMarks block is added to either chart, the following behavior is observed: Diffing Failure: The framework appears unable to maintain the identity of the axis components during the scroll update. Redraw Loop: Instead of an incremental scroll translation, the diffing algorithm triggers a full reload/re-render of both charts on every scroll offset change. Impact: CPU spikes to 100% and the UI becomes unresponsive. This happens even if the custom AxisMarks is used solely to hide the axis (e.g., AxisMarks { _ in }), suggesting the issue is with the custom declaration itself rather than the complexity of the marks being rendered. Steps to Reproduce Create two Chart views in a VStack. Bind both to a single @Observable property using .chartScrollPosition(id: $state.pos). Add any .chartXAxis { AxisMarks(...) { ... } } modifier. Scroll either chart; observe the stuttering. import SwiftUI import Charts import Observation @Observable class ChartState { var scrollPos: Date = .now } struct PerformanceBugView: View { @State private var state = ChartState() var body: some View { VStack { Chart(data) { ... } .chartScrollPosition(id: $state.scrollPos) .chartXAxis { // This custom mark triggers the performance issue AxisMarks { _ in AxisValueLabel() } } Chart(data) { ... } .chartScrollPosition(id: $state.scrollPos) } } } Questions for the Community/Apple Engineers: Is there a way to provide a stable identifier to AxisMarks to prevent them from being treated as "new" during a scroll update? Why does even an empty AxisMarks block (used for hiding) trigger a layout invalidation that standard axes do not? Are there internal optimizations for chartScrollPosition that are bypassed when the axis layout is customized?

My project uses about 10 local packages. When Xcode determins its dependencies graph it includes project Foo (a package) and project Foo_Foo (???). This is a no op for normal builds, but causes the Foo package to be linked twice for test builds which then fail due to duplicate symbols. Nowhere in my repository is Foo_Foo defined or reference. Nowhere in the .xcodeproj/ folder is there mention of Foo_Foo. What am I doing wrong?

Hi everyone, I’m facing an issue with CloudKit sync getting stuck during initial device migration in my SwiftData-based app. The app follows a local-first architecture using SwiftData + CloudKit sync, and works correctly for: ✔ Incremental sync ✔ Bi-directional updates ✔ Small datasets However, when onboarding a new device with large historical data, sync becomes extremely slow or appears stuck. Even after two hours data is not fully synced. ~6900 Transactions 🚨 Problem When installing the app on a new iPhone and enabling iCloud sync: • Initial hydration starts • A small amount of data syncs • Then sync stalls indefinitely Observed behaviour: • iPhone → Mac sync works (new changes sync back) • Mac → iPhone large historical migration gets stuck • Reinstalling app / clearing container does not resolve issue • Sync never completes full migration This gives the impression that: CloudKit is trickling data but not progressing after a certain threshold. The architecture is: • SwiftData local store • Manual CloudKit sync layer • Local-first persistence • Background push/pull sync So I understand: ✔ Conflict resolution is custom ✔ Initial import may not be optimized by default But I expected CloudKit to eventually deliver all records. Instead, the new device remains permanently in a “partial state”. ⸻ 🔍 Observations • No fatal CloudKit errors • No rate-limit errors • No quota issues • iCloud is available • Sync state remains “Ready” • Hydration remains “mostlyReady” Meaning: CloudKit does not report failure — but data transfer halts. ⸻ 🤔 Questions Would appreciate guidance on: Is CloudKit designed to support large initial dataset migration via manual sync layers? Or is this a known limitation vs NSPersistentCloudKitContainer? ⸻ Does CloudKit internally throttle historical record fetches? Could it silently stall without error when record volume is high? ⸻ Is there any recommended strategy for: • Bulk initial migration • Progressive hydration • Forcing forward sync progress ⸻ Should initial migration be handled outside CloudKit (e.g. via file transfer / backup restore) before enabling sync? ⸻ 🎯 Goal I want to support: • Large historical onboarding • Multi-device sync • User-visible progress Without forcing migration to Core Data. ⸻ 🙏 Any advice on: • Best practices • Debugging approach • CloudKit behavior in such scenarios would be greatly appreciated. Thank you!

Hello everyone! I am trying to wrap a ViewModifier inside a Swift Package that bundles a metal shader file to be used in the modifier. Everything works as expected in the Preview, in the Simulator and on a real device for iOS. It also works in Preview and in the Simulator for tvOS but not on a real AppleTV. I have tried this on a 4th generation Apple TV running tvOS 26.3 using Xcode 26.2.0. Xcode logs the following: The metallib is processed and exists in the bundle. Compiler failed to build request precondition failure: pipeline error: custom_effect-fg2a5cia7fmha4: error: unresolved visible function reference: custom_fn Reason: visible function not loaded Compiler failed to build request precondition failure: pipeline error: custom_effect-fg2a5cia7fmha4: error: unresolved visible function reference: custom_fn Reason: visible function not loaded Compiler failed to build request precondition failure: pipeline error: custom_effect-fg2a5cia7fmha4: error: unresolved visible function reference: custom_fn Reason: visible function not loaded Compiler failed to build request precondition failure: pipeline error: custom_effect-fg2a5cia7fmha4: error: unresolved visible function reference: custom_fn Reason: visible function not loaded Compiler failed to build request precondition failure: pipeline error: custom_effect-fg2a5cia7fmha4: error: unresolved visible function reference: custom_fn Reason: visible function not loaded Compiler failed to build request precondition failure: pipeline error: custom_effect-fg2a5cia7fmha4: error: unresolved visible function reference: custom_fn Reason: visible function not loaded Contents of Package.swift: import PackageDescription let package = Package( name: "Test", platforms: [ .iOS(.v17), .tvOS(.v17) ], products: [ .library( name: "Test", targets: [ "Test" ] ) ], targets: [ .target( name: "Test", resources: [ .process("Shaders") ] ), .testTarget( name: "TestTests", dependencies: [ "Test" ] ) ] ) Content of my metal file: #include <metal_stdlib> using namespace metal; [[ stitchable ]] float2 complexWave(float2 position, float time, float2 size, float speed, float strength, float frequency) { float2 normalizedPosition = position / size; float moveAmount = time * speed; position.x += sin((normalizedPosition.x + moveAmount) * frequency) * strength; position.y += cos((normalizedPosition.y + moveAmount) * frequency) * strength; return position; } And my ViewModifier: import MetalKit import SwiftUI extension ShaderFunction { static let complexWave: ShaderFunction = { ShaderFunction( library: .bundle(.module), name: "complexWave" ) }() } extension Shader { static func complexWave(arguments: [Shader.Argument]) -> Shader { Shader(function: .complexWave, arguments: arguments) } } struct WaveModifier: ViewModifier { let start: Date = .now func body(content: Content) -> some View { TimelineView(.animation) { context in let delta = context.date.timeIntervalSince(start) content .visualEffect { view, proxy in view.distortionEffect( .complexWave( arguments: [ .float(delta), .float2(proxy.size), .float(0.5), .float(8), .float(10) ] ), maxSampleOffset: .zero ) } } .onAppear { let paths = Bundle.module.paths(forResourcesOfType: "metallib", inDirectory: nil) print(paths) } } } extension View { public func wave() -> some View { modifier(WaveModifier()) } } #Preview { Image(systemName: "cart") .wave() } Any help is appreciated.

I have a live activity, that works fine when the Lock Screen showing, but as soon as it "sleeps" dims down for always on display, everything in the widget disappears and an Activity Indicator(spinner), displays in its place, but non-animating.

Hi! I have a Catalyst app that embeds command line utility. So the project has two targets: Catalyst target, this target depends on #2 and embeds it into its bundle. macOS target, the command line tool. Both targets have package dependency to the same package. I used this to embed the CMD tool. Everything builds, runs and works fine until I try to archive the project. Archiving stops early with such error: error: Multiple commands produce '/Users/kse2/Library/Developer/Xcode/DerivedData/PkgTest-clngkndczxoprpdlwefqqiqlryjt/Build/Intermediates.noindex/ArchiveIntermediates/PkgTest/IntermediateBuildFilesPath/UninstalledProducts/macosx/MacrosForSwift.o' note: Target 'MacrosForSwift' (project 'MacrosForSwift') has a command with output '/Users/kse2/Library/Developer/Xcode/DerivedData/PkgTest-clngkndczxoprpdlwefqqiqlryjt/Build/Intermediates.noindex/ArchiveIntermediates/PkgTest/IntermediateBuildFilesPath/UninstalledProducts/macosx/MacrosForSwift.o' note: Target 'MacrosForSwift' (project 'MacrosForSwift') has a command with output '/Users/kse2/Library/Developer/Xcode/DerivedData/PkgTest-clngkndczxoprpdlwefqqiqlryjt/Build/Intermediates.noindex/ArchiveIntermediates/PkgTest/IntermediateBuildFilesPath/UninstalledProducts/macosx/MacrosForSwift.o' What I have tried to fix archiving: Changing PkgTestCMD target to be on Catalyst/iOS instead of macOS. This works but I'm not sure how to properly run CMD tool with iOS SDK (if this correct at all): I need the main thread to be 'unblocked' and be active while background tasks exist. Adding an aux framework that act as container for the package. Doesn't work. Splitting targets into different projects, making workspace and cross-project reference. Doesn't work. My understanding is that archiving attempts to produce .o files for both Catalyst and macOS simultaneously, due to PkgTest being Catalyst and PkgTestCMD being macOS targets. How to archive such a project? Is there a way to separate archiving of CMD and main app? Maybe separate .o files into different directories. Simple building and running the project works, why archiving doesn't want to? I have a test project for the issue: PkgTest. main branch is the source project, DepFramework contains attempt #2, Workspace—#3. Thank you!

I am working on a remote control application for macOS where I need to maintain two "virtual" cursors: Remote Cursor: Follows the remote user's movements. Local Cursor: Follows the local user's physical mouse/trackpad movements. To move the system cursor (for the remote side), I use CGWarpMouseCursorPosition as follows: void DualCursorMac::UpdateSystemCursorPosition(int x, int y) { CGPoint point = CGPointMake(static_cast<CGFloat>(x), static_cast<CGFloat>(y)); // Warp the cursor to match remote coordinates CGWarpMouseCursorPosition(point); } Meanwhile, I use a CGEventTap to monitor local physical movements to update my local virtual cursor's UI: CGEventRef Mouse::MouseTapCallback(CGEventTapProxy proxy, CGEventType type, CGEventRef event, void *refcon) { if (remoteControlMode) { // We want to suppress system cursor movement but still read the delta const int deltaX = static_cast<int>(CGEventGetIntegerValueField(event, kCGMouseEventDeltaX)); const int deltaY = static_cast<int>(CGEventGetIntegerValueField(event, kCGMouseEventDeltaY)); NSLog(@"MouseTapCallback: delta:(%d, %d)", deltaX, deltaY); // Update local virtual cursor UI based on deltas... return nullptr; // Consume the event } return event; } The Problem: When CGWarpMouseCursorPosition is called frequently to update the system cursor, it interferes with the kCGMouseEventDeltaX/Y values in the Event Tap. Specifically, if the local user moves the trackpad slowly (expecting deltas of 1 or 2), but a "Warp" occurs simultaneously (e.g., jumping the cursor from (100, 100) to (300, 300)), the deltaX and deltaY in the callback suddenly spike to very large values. It seems the system calculates the delta based on the new warped position rather than the pure physical displacement of the trackpad. This makes the local virtual cursor "jump" erratically and makes it impossible to track smooth local movement during remote control. My Question: Is there a way to get the "raw" or "pure" physical relative movement (delta) from the trackpad/mouse that is independent of the system cursor's absolute position or warping? Are there alternative APIs (perhaps IOKit or different CGEvent fields) that would allow me to get consistent deltas even when the cursor is being warped programmatically?

Can I use external Swift Package Dependencies for the challenge, or will my app just not run as the testing environment is offline? I have a couple dependencies which are crucial to the running of my submission

Good day, I've uploaded a build to TestFlight, but received an automated response with the following error: ITMS-90426: Invalid Swift Support - The SwiftSupport folder is missing. Rebuild your app using the current public (GM) version of Xcode and resubmit it. Our project started in Objective-C and have mixed swift class and pods. The last uploaded build without any automated response was Nov 8, 2023. I'm using XCode Version 26.0.1 (17A400). I've tried every way i found in internet and i'm not able to find any solution for this. ALWAYS_EMBED_SWIFT_STANDARD_LIBRARIES = YES use_frameworks! :linkage => :dynamic (in pods) We would appreciate any assistance in clarifying why this issue is occurring and how we should proceed to address it. Your guidance would mean a lot. Thank you.

Hi there, I just re-install 26.2, but I got many issues regarding to SwiftUI's components. How can I fix it? Do I need download or install something else?

I downloaded package, I can see it in Package Dependencies tab, [swift-collections | https://github.com/apple/swift-collections.git | Up to Next Major Version] — looks OK. But (quite strange): import Collections //🛑 No such module 'Collections' typealias Objects = OrderedSet<anyHashable> // No error I took Collection word from a list or typed — no difference. if I remove import Collections there is no OrderedSet anymore (of course) Can't compile. Cleaned build folder, restarted Xcode... What could be wrong with it? Any ideas?

This post documents an issue I reported in feedback FB19610114 and see if anyone knows of a workaround. Here is a copy of the feedback. Short version Manipulation (SwiftUI OR RealityKit) fails to translate entities after changing rooms. By changing rooms, I mean a human wearing an Apple Vision Pro leaving one room and entering another room. Once this issue occurs, it impacts all apps that use these features. A device restart is the only solution I have to fix it. Feedback FB19610114 This is an odd one. I'm using the new Manipulation Component in visionOS 26. Most of the time this works well. Sometime it stops working and when it does the only way to get it working again is to reboot the headset. When this happens, I can continue to rotate and scale items, but translation no longer works. It is as if the item is stuck to a fixed point in the parent scene (window, volume, etc). When this bug occurs, it affects every app across the entire operating system that is using manipulation, including the RealityKit component AND the SwiftUI version. This is not limited to one app and is not limited to apps that I am working on. Once this error occurs, it affects literally any application across the operating system that is using this API, including apps from Apple. I won't speculate on the cause of this, but I do know of one way where I can always get it to happen. Here is how to reproduce it: Make an Xcode project with a single entity that uses the Manipulation Component. There is no need to customize the configuration of this component. The default implementation will work. Build and run this app on device. You can keep running from device or quit and launch the app like normal on device. Open the app and manipulate the entity - it should work as expected. Physically walk into another room. It is vital that you leave the current room that you are in and enter a different room entirely. Use the digital crown to recenter your view and bring your window or volume to you. Test the manipulation on the entity again - it should still be working as expected at this point. Physically, move yourself and your headset into the original room where you started. Use the digital crown to recenter your view and bring your window or volume to you. Test the manipulation on the entity again - you should now see the issue. When I follow the steps above, then 100% of the time manipulation translation stops working at this point. It will impact any application using this API. The only way to fix it is to restart my headset. A few points to keep in mind It does not matter if an app is actively being run from Xcode. When this occurs, it impacts every single app, not just one. When this occurs, rotation and scaling continue to work, but the entity/view cannot be translated. This impacts BOTH the SwiftUI version and the RealityKit version. When this occurs, the only way to "fix" it is to reboot the device.

Is it possible to achieve sub-second end-to-end latency when displaying live streaming video using APMP (Apple Projected Media Profile) with Wide FoV? APMP supports HLS playback, but my understanding is that standard HLS introduces several seconds of latency. I would like to know whether APMP (especially Wide FoV) supports Low-Latency HLS, or if there are inherent limitations that make sub-second latency impractical. If APMP is not suitable for this use case, are there any recommended alternatives within AVFoundation or related frameworks for rendering wide-FoV live video with very low latency? Thank you for any insights.

I’m building an iOS SDK that is distributed as a binary XCFramework and consumed via Swift Package Manager using a binaryTarget. What I’ve done so far: Built the SDK using xcodebuild archive for device and simulator Created the XCFramework using xcodebuild -create-xcframework The SDK contains a resource bundle with JSON/config files The XCFramework is wrapped using SPM (code-only wrapper target) Currently, the resource bundle exists outside the XCFramework, and the host app needs to add it manually during integration. I want to avoid this and make the SDK completely self-contained. What I’m trying to achieve: Embed the resource bundle inside the SDK framework so that each XCFramework slice contains it Ensure the SDK can load its assets internally at runtime without any host app changes Questions: What is the correct way to embed a .bundle inside a framework so it gets packaged into each XCFramework slice during archiving? Which Xcode build phases or build settings are required for this (e.g., Copy Bundle Resources, SKIP_INSTALL, etc.)? At runtime, what is the recommended approach for locating and loading this embedded bundle from within the SDK code? Any guidance or best practices for achieving this would be helpful.

Hi! I have a bigger Xcode project that has multiple targets: the main app (MainApp) helper command line tool (HelperTool) The project also lists multiple package dependencies via SwiftPM and these dependencies have dependencies between themselves. One of such packages produces a dynamic library MyFramework which is a dependency for both the main app and the HelperTool which has it listed under Build Phases > Dependencies as well as under Link with Frameworks. This builds just fine, but the issue somes when I want to add another target called AdditionalHelperTool which it has pretty much the same dependencies as HelperTool. When I add this second target, I start running into issues like the following: Multiple commands produce '[...]/Build/Products/Debug/Frameworks/MyFramework.framework/Versions/A' Target 'HelperTool' (project 'MyApp') has copy command from '[...]/Build/Products/Debug/PackageFrameworks/MyFramework.framework' to '[...]/Build/Products/Debug/Frameworks/MyFramework.framework' Target 'AdditionalHelperTool' (project 'MyApp') has copy command from '[...]/Build/Products/Debug/PackageFrameworks/MyFramework.framework' to '[...]/Build/Products/Debug/Frameworks/MyFramework.framework'` It seems that Xcode runs the build of both targets separately and it sees a conflict given that both targets have same dependencies that it's trying to built separately. Has anyone encountered something similar? Can anyone suggest a solution for this?

I am using swift-subprocess, and need to disable the SubprocessSpan trait because Xcode 26.2 does not ship with a bundled version libswiftCompatibilitySpan.dylib, causing everything to crash built with Xcode that happens to use Span. However, I cannot disable that trait by doing any of the following things: .package( url: "https://github.com/swiftlang/swift-subprocess.git", branch: "main", traits: [] ), .package( url: "https://github.com/swiftlang/swift-subprocess.git", branch: "main", traits: [.trait(name: "SubprocessFoundation")] ), Note that SubprocessSpan is default trait in subprocess: // Enable SubprocessFoundation by default var defaultTraits: Set<String> = ["SubprocessFoundation"] #if compiler(>=6.2) // Enable SubprocessSpan when Span is available [except it is not] defaultTraits.insert("SubprocessSpan") #endif The package still builds with the SubprocessSpan enabled. This is not an issue with the subprocess package. According to this, I should use swift build on the command line, yet this isn't -- as is upgrading to Tahoe -- an option because I need Xcode previews for SwiftUI. Help, what do I do now (other than downloading a toolchain) from swift.org)?

Hello I am using 26.2 (25C56) on my m4 air. when I am working on a Flutter project suddenly my app crashed and after when I tried to build the application again this error occurs. How to fix it. ? Launching lib/main.dart on iPhone 16e in debug mode... Xcode build done. 2.4s Failed to build iOS app Uncategorized (Xcode): Unable to find a destination matching the provided destination specifier: { id:45967BD9-44E0-4D5D-8C43-C5437F48836A }Available destinations for the "Runner" scheme: { platform:macOS, arch:arm64, variant:Designed for [iPad,iPhone], id:00008132-001E5102020B801C, name:My Mac } { platform:iOS, id:dvtdevice-DVTiPhonePlaceholder-iphoneos:placeholder, name:Any iOS Device } { platform:iOS Simulator, id:dvtdevice-DVTiOSDeviceSimulatorPlaceholder-iphonesimulator:placeholder, name:Any iOS Simulator Device } Could not build the application for the simulator. Error launching application on iPhone 16e.

So I’m writing a program, as a developer would - ‘with Xcode.’ Code produced an error. The key values were swapped. The parameters suggested were ‘optional parameters variables.’ “var name: TYPE? = (default)” var name0: TYPE ============================= name0 = “super cool” ‘Name is not yet declared at this point provided with x - incorrect argument replace ExampleStruct(name:”supercool”) should be x - incorrect argument replace ExampleStruct(name0:”supercool”) ============================= In swift, there is a procedural prioritization within the constructor calling process. Application calls constructor. Constructor provides constructor signature. Signature requires parameters & throws an error if the params are not in appropriate order. - “got it compiler; thank you, very much” Typically, when this occurs, defaults will be suggested. Often the variable type. Ie String, Bool. such as: StructName(param1:Int64, param2:Bool) (Recently, I have seen a decline in @Apple’s performance in many vectors.) As stated before, the key value pairs were out of sequence. The optionals were suggested instead of the required parameters. This leads me to believe that there is an order of operations in the calling procedure that is being mismanaged. I.e. regular expression, matching with optional. This confuses these with [forced, required] parameters, and the mismanagement of ‘key: value’ pairs. this is a superficial prognosis and would like to know if anyone has any insight as to why this may occur. Could it be a configuration setting? Is it possibly the network I connected to bumped into something. Etc.. I appreciate any and all feedback. Please take into consideration the Apple developer forum, guidelines before posting comments. #dev_div

private static func getEthernetIPAddress(from interfaces: [String: String]) -> String? { // 常见虚拟以太网接口名（根据适配器型号可能不同） let poeEthernetInterfaces = ["en2", "en3", "en4", "en5", "eth0", "eth1"] for interfaceName in poeEthernetInterfaces { if let ethernetIP = interfaces[interfaceName], !ethernetIP.isEmpty { return ethernetIP } } return nil }//我们通过该方法去抓取有线网的IP地址，但是有的设备无法抓取到，怎样才能更准确的抓取到有线网络的IP地址

Hi everyone, I've been building an on-device AI safety layer called Newton Engine, designed to validate prompts before they reach FoundationModels (or any LLM). Wanted to share v1.3 and get feedback from the community. The Problem Current AI safety is post-training — baked into the model, probabilistic, not auditable. When Apple Intelligence ships with FoundationModels, developers will need a way to catch unsafe prompts before inference, with deterministic results they can log and explain. What Newton Does Newton validates every prompt pre-inference and returns: Phase (0/1/7/8/9) Shape classification Confidence score Full audit trace If validation fails, generation is blocked. If it passes (Phase 9), the prompt proceeds to the model. v1.3 Detection Categories (14 total) Jailbreak / prompt injection Corrosive self-negation ("I hate myself") Hedged corrosive ("Not saying I'm worthless, but...") Emotional dependency ("You're the only one who understands") Third-person manipulation ("If you refuse, you're proving nobody cares") Logical contradictions ("Prove truth doesn't exist") Self-referential paradox ("Prove that proof is impossible") Semantic inversion ("Explain how truth can be false") Definitional impossibility ("Square circle") Delegated agency ("Decide for me") Hallucination-risk prompts ("Cite the 2025 CDC report") Unbounded recursion ("Repeat forever") Conditional unbounded ("Until you can't") Nonsense / low semantic density Test Results 94.3% catch rate on 35 adversarial test cases (33/35 passed). Architecture User Input ↓ [ Newton ] → Validates prompt, assigns Phase ↓ Phase 9? → [ FoundationModels ] → Response Phase 1/7/8? → Blocked with explanation Key Properties Deterministic (same input → same output) Fully auditable (ValidationTrace on every prompt) On-device (no network required) Native Swift / SwiftUI String Catalog localization (EN/ES/FR) FoundationModels-ready (#if canImport) Code Sample — Validation let governor = NewtonGovernor() let result = governor.validate(prompt: userInput) if result.permitted { // Proceed to FoundationModels let session = LanguageModelSession() let response = try await session.respond(to: userInput) } else { // Handle block print("Blocked: Phase \(result.phase.rawValue) — \(result.reasoning)") print(result.trace.summary) // Full audit trace } Questions for the Community Anyone else building pre-inference validation for FoundationModels? Thoughts on the Phase system (0/1/7/8/9) vs. simple pass/fail? Interest in Shape Theory classification for prompt complexity? Best practices for integrating with LanguageModelSession? Links GitHub: https://github.com/jaredlewiswechs/ada-newton Technical overview: parcri.net Happy to share more implementation details. Looking for feedback, collaborators, and anyone else thinking about deterministic AI safety on-device.