@camstack/addon-pipeline 1.1.1 → 1.1.2

package/dist/detection-pipeline/index.js

@@ -121,7 +121,6 @@ function npuInfoFrom(hw) {
 	return { type };
 }
 function envToHardwareInfo(env) {
-	if (!env.hardware) return null;
 	return {
 		platform: toKnownPlatform(env.platform),
 		arch: toKnownArch(env.arch),
@@ -129,8 +128,8 @@ function envToHardwareInfo(env) {
 		cpuCores: 0,
 		totalRAM_MB: 0,
 		availableRAM_MB: 0,
-		gpu: gpuInfoFrom(env.hardware),
-		npu: npuInfoFrom(env.hardware)
+		gpu: env.hardware ? gpuInfoFrom(env.hardware) : null,
+		npu: env.hardware ? npuInfoFrom(env.hardware) : null
 	};
 }
 function probedToHardwareInfo(hw) {
@@ -4096,6 +4095,17 @@ var ONNX_FLOOR = {
 * instead of duplicated inline.
 */
 function onnxFloorPick() {
+	return ONNX_FLOOR;
+}
+/**
+* The platform-deterministic engine pick computed SYNCHRONOUSLY from this node's
+* own `process.platform`/`arch` alone (no probe): darwin → coreml, else → onnx
+* (gpu-dependent openvino/cuda need the probe and converge via the auto-pick).
+* Used as the ENGINE fallback when a persisted selection is unsupported on this
+* node — so a stale GLOBAL engine config (e.g. the cluster's OpenVINO choice)
+* can never force an impossible engine onto a node whose platform rejects it.
+*/
+function platformDefaultPick() {
 	const pick = pickBestRuntime(runtimeEnvFromProcess(null), null);
 	return {
 		runtime: "python",
@@ -4104,6 +4114,19 @@ function onnxFloorPick() {
 		device: pick.device
 	};
 }
+/**
+* Is `backend` even POSSIBLE on this node's OS/arch (ignoring gpu detail)?
+* coreml ⇒ darwin only; openvino ⇒ x64 non-darwin only; onnx ⇒ anywhere. Used to
+* reject a persisted/global engine choice that the node's PLATFORM fundamentally
+* cannot run (e.g. the cluster's OpenVINO default landing on a Mac) — distinct
+* from the gpu-dependent support (linux without a probed Intel iGPU still keeps
+* openvino as a valid platform choice; the device falls back to cpu).
+*/
+function backendPossibleOnPlatform(backend) {
+	if (backend === "coreml") return process.platform === "darwin";
+	if (backend === "openvino") return process.arch === "x64" && process.platform !== "darwin";
+	return true;
+}
 var DetectionPipelineProvider = class DetectionPipelineProvider {
 	modelsDir;
 	eventBus;
@@ -4348,17 +4371,11 @@ var DetectionPipelineProvider = class DetectionPipelineProvider {
 	*/
 	async autoPickAndPersist() {
 		let hardware = null;
-		let bestBackendHint = null;
 		try {
 			const api = this.addonCtx?.api;
-			if (api) {
-				const caps = await api.platformProbe.getCapabilities.query({ nodeId: this.localProbeNodeId() });
-				hardware = caps?.hardware ?? null;
-				const bs = caps?.bestScore;
-				if (bs && bs.runtime === "python") bestBackendHint = bs.backend;
-			}
+			if (api) hardware = (await api.platformProbe.getCapabilities.query({ nodeId: this.localProbeNodeId() }))?.hardware ?? null;
 		} catch {}
-		const pick = pickBestRuntime(runtimeEnvFromProcess(toProbedHardware(hardware)), bestBackendHint);
+		const pick = pickBestRuntime(runtimeEnvFromProcess(toProbedHardware(hardware)), null);
 		const engine = {
 			runtime: "python",
 			backend: pick.runtimeId,
@@ -4366,8 +4383,8 @@ var DetectionPipelineProvider = class DetectionPipelineProvider {
 			device: pick.device
 		};
 		this.currentEngine = engine;
-		if (!(hardware !== null || bestBackendHint !== null)) {
-			this.log.warn("Auto-pick: probe returned no hardware/hint — using onnx floor WITHOUT persisting", { meta: {
+		if (!(pick.runtimeId !== "onnx" || hardware !== null)) {
+			this.log.info("Auto-pick: onnx floor pending gpu probe — NOT persisting (re-pick on done)", { meta: {
 				backend: pick.runtimeId,
 				device: pick.device
 			} });
@@ -4378,10 +4395,10 @@ var DetectionPipelineProvider = class DetectionPipelineProvider {
 			[nodeEngineKey("engineBackend", apNode)]: pick.runtimeId,
 			[nodeEngineKey("engineDevice", apNode)]: pick.device
 		});
-		this.log.info("Auto-picked engine at first boot", { meta: {
+		this.log.info("Auto-picked engine (platform-deterministic)", { meta: {
 			backend: pick.runtimeId,
 			device: pick.device,
-			hint: bestBackendHint ?? "none"
+			hadProbeHardware: hardware !== null
 		} });
 	}
 	/** Map a backend string to a known RuntimeId, flooring to onnx. */
@@ -4441,7 +4458,9 @@ var DetectionPipelineProvider = class DetectionPipelineProvider {
 		const runtimeId = this.toRuntimeId(engine.backend);
 		const device = engine.device ?? "cpu";
 		const snapshot = this.provisioner.state;
-		if (snapshot.runtimeId === runtimeId && snapshot.device === device && snapshot.state !== "idle") return;
+		const sameSelection = snapshot.runtimeId === runtimeId && snapshot.device === device;
+		if (sameSelection && snapshot.state !== "idle") return;
+		if (!sameSelection) this.currentSteps = null;
 		this.provisioner.select(runtimeId, device);
 	}
 	/**
@@ -4677,13 +4696,52 @@ var DetectionPipelineProvider = class DetectionPipelineProvider {
 	async getDefaultSteps(engine) {
 		return buildDefaultStepTree(engine.format);
 	}
+	/**
+	* Substitute, AT RUNTIME ONLY (never persisted), any step whose configured
+	* model lacks a build for that step's active engine format. A GLOBAL pipeline
+	* config can pin a model that's valid on the cluster's Intel nodes (e.g. the
+	* OpenVINO-only `yolov9t-int8`) but impossible on a CoreML/ONNX node — without
+	* this the node crash-loops `Model "X" has no <format> format`. The operator's
+	* choice stays in the persisted config; this node just runs the smallest
+	* catalog model that DOES have its format. Mirrors the per-node engine
+	* fallback in `loadEngine`. Catalog models only — a custom model that lacks
+	* the format is left as-is (operator's responsibility).
+	*/
+	substituteIncompatibleModels(steps) {
+		const fix = (step) => {
+			const format = step.engine?.format ?? this.currentEngine.format;
+			let modelId = step.modelId;
+			try {
+				const entry = getStepDefinition(step.addonId).models.find((m) => m.id === step.modelId);
+				if (entry && !entry.formats[format]) {
+					const fallback = getDefaultModelForFormat(step.addonId, format);
+					if (fallback !== step.modelId) {
+						this.log.info("Step model lacks engine format — substituting format default (runtime)", { meta: {
+							step: step.addonId,
+							configured: step.modelId,
+							substitute: fallback,
+							format
+						} });
+						modelId = fallback;
+					}
+				}
+			} catch {}
+			const children = step.children?.length ? step.children.map(fix) : step.children;
+			return modelId === step.modelId && children === step.children ? step : {
+				...step,
+				modelId,
+				...children ? { children } : {}
+			};
+		};
+		return steps.map(fix);
+	}
 	async getGlobalSteps() {
 		if (this.currentSteps) return this.currentSteps;
 		const raw = (await this.readStore())[KEY_STEPS];
 		if (!raw) {
 			const defaults = buildDefaultStepTree(this.currentEngine.format);
 			if (defaults.length === 0) return null;
-			this.currentSteps = defaults;
+			this.currentSteps = this.substituteIncompatibleModels(defaults);
 			this.writeStore({ [KEY_STEPS]: JSON.stringify(defaults) });
 			this.log.info("Bootstrapped default pipeline — object-detection + face + plate recognition enabled by default", { meta: { rootSteps: defaults.length } });
 			return this.currentSteps;
@@ -4721,7 +4779,7 @@ var DetectionPipelineProvider = class DetectionPipelineProvider {
 					this.log.info("Migration: added audio-classifier step to persisted pipeline config");
 				}
 			}
-			this.currentSteps = steps;
+			this.currentSteps = this.substituteIncompatibleModels(steps);
 			return this.currentSteps;
 		} catch {
 			throw new Error(`Failed to parse persisted pipeline steps: corrupt data in key "${KEY_STEPS}"`);
@@ -5872,12 +5930,13 @@ var DetectionPipelineProvider = class DetectionPipelineProvider {
 			const storedDevice = typeof storedDeviceRaw === "string" ? storedDeviceRaw : "";
 			const floor = onnxFloorPick();
 			const migratedBackend = typeof storedRuntime === "string" && storedRuntime === "node" && backend === "cpu" ? "onnx" : backend;
-			if (!DetectionPipelineProvider.isPythonBackendAvailable(migratedBackend, this.executorOptions.pythonPath ?? "")) {
-				this.log.warn("Stored engine backend unavailable on this node — falling back to onnx floor", { meta: {
+			if (!backendPossibleOnPlatform(migratedBackend)) {
+				const platformDefault = platformDefaultPick();
+				this.log.warn("Stored engine backend impossible on this platform — using platform default", { meta: {
 					stored: migratedBackend,
-					fallback: `${floor.backend}`
+					fallback: platformDefault.backend
 				} });
-				return floor;
+				return platformDefault;
 			}
 			const device = storedDevice || floor.device;
 			return {
@@ -6049,44 +6108,61 @@ var DetectionPipelineProvider = class DetectionPipelineProvider {
 	}
 	async reprobeEngine() {
 		const api = this.addonCtx?.api;
-		let best;
+		let hardware = null;
 		if (api) try {
-			const caps = await api.platformProbe.getCapabilities.query({ nodeId: this.localProbeNodeId() });
-			const bs = caps?.bestScore;
-			if (bs && bs.runtime === "python") {
-				const probeBackend = bs.backend;
-				const probeDevice = (() => {
-					const hw = caps.hardware;
-					if (probeBackend === "openvino") return defaultDeviceFor("openvino");
-					if (probeBackend === "coreml") return defaultDeviceFor("coreml");
-					if (probeBackend === "onnx") return hw?.gpu?.type === "nvidia" ? "cuda" : "cpu";
-					return "cpu";
-				})();
-				best = {
-					runtime: "python",
-					backend: probeBackend,
-					format: backendToFormat(probeBackend),
-					device: probeDevice
-				};
-			} else best = onnxFloorPick();
-		} catch {
-			best = onnxFloorPick();
-		}
-		else best = onnxFloorPick();
+			hardware = (await api.platformProbe.getCapabilities.query({ nodeId: this.localProbeNodeId() }))?.hardware ?? null;
+		} catch {}
+		const pick = pickBestRuntime(runtimeEnvFromProcess(toProbedHardware(hardware)), null);
+		const best = {
+			runtime: "python",
+			backend: pick.runtimeId,
+			format: modelFormatFor(pick.runtimeId),
+			device: pick.device
+		};
 		const probedLabel = `${best.backend}/${best.device ?? "default"}`;
 		const rpNode = this.localProbeNodeId();
-		await this.writeStore({
-			[nodeEngineKey("probedBestEngine", rpNode)]: probedLabel,
-			[nodeEngineKey("engineBackend", rpNode)]: best.backend,
-			[nodeEngineKey("engineDevice", rpNode)]: best.device ?? "cpu"
-		});
-		this.log.info("Re-probed engine — wrote back engineBackend + engineDevice", { meta: {
+		if (pick.runtimeId !== "onnx" || hardware !== null) {
+			await this.writeStore({
+				[nodeEngineKey("probedBestEngine", rpNode)]: probedLabel,
+				[nodeEngineKey("engineBackend", rpNode)]: best.backend,
+				[nodeEngineKey("engineDevice", rpNode)]: best.device ?? "cpu"
+			});
+			this.log.info("Re-probed engine (platform-deterministic) — wrote back", { meta: {
+				backend: best.backend,
+				device: best.device ?? null,
+				probedBestEngine: probedLabel
+			} });
+		} else this.log.info("Re-probe: onnx floor pending gpu probe — NOT persisting (re-pick on done)", { meta: {
 			backend: best.backend,
-			device: best.device ?? null,
-			probedBestEngine: probedLabel
+			device: best.device ?? null
 		} });
 		return best;
 	}
+	/**
+	* Re-pick the engine when the platform-probe finishes its async hardware +
+	* Python detection (the `platform-probe.phase` `done` event). At boot the
+	* probe's accelerator result may not be ready yet, so the engine floored to
+	* onnx; once the probe answers (e.g. a Mac's CoreML/ANE surfaces after the
+	* embedded Python is installed) this re-runs the probe-driven pick and
+	* re-provisions. Idempotent: `startProvisioningForCurrentEngine` skips a
+	* no-op when the selection is unchanged.
+	*/
+	async repickEngineOnProbeReady() {
+		const before = `${this.currentEngine.backend}/${this.currentEngine.device ?? "default"}`;
+		await this.reprobeEngine();
+		const stored = await this.loadEngine();
+		if (stored) {
+			this.currentEngine = stored;
+			this.needsAutoPick = false;
+			this.cancelDeferredAutoPick();
+		}
+		const after = `${this.currentEngine.backend}/${this.currentEngine.device ?? "default"}`;
+		if (before !== after) this.log.info("Engine re-picked after platform-probe completed", { meta: {
+			before,
+			after
+		} });
+		this.startProvisioningForCurrentEngine();
+	}
 	async getReferenceAudioFiles() {
 		const dir = resolveReferenceAudioDir();
 		if (!dir) return [];
@@ -6554,6 +6630,7 @@ var DetectionPipelineAddon = class extends require_dist.BaseAddon {
 	nodeEngineBackend = DEFAULT_CONFIG.engineBackend;
 	nodeProbedBestEngine = "";
 	engineMetricsTimer = null;
+	probePhaseUnsub = null;
 	/** Snapshot-equality cache for engine-metrics emit. Most ticks
 	*  the engine inventory is unchanged (no model load/unload), so
 	*  we skip the bus emit and let the heartbeat re-emit at
@@ -6960,11 +7037,35 @@ var DetectionPipelineAddon = class extends require_dist.BaseAddon {
 			numWorkers: num(t["numWorkers"], 1)
 		};
 	}
-	async onInitialize() {
-		const modelsDir = await this.ctx.api.storage.resolve.query({
+	/**
+	* Resolve the directory models are downloaded into, resilient across nodes.
+	* The `models` storage location is GLOBAL (hub-seeded) and can resolve to a
+	* path that exists only on the hub — e.g. Docker's `/data/models`, which an
+	* agent on a different filesystem (a Mac) cannot create (`ENOENT mkdir
+	* /data/models`). Verify the resolved dir is creatable; otherwise fall back to
+	* this node's LOCAL addon data-dir so models always land on writable disk.
+	*/
+	async resolveModelsDir() {
+		const fallback = node_path.join(this.ctx.dataDir, "models");
+		const candidate = await this.ctx.api.storage.resolve.query({
 			location: "models",
 			relativePath: ""
-		}).catch(() => "camstack-data/models");
+		}).catch(() => null) ?? fallback;
+		try {
+			await node_fs.promises.mkdir(candidate, { recursive: true });
+			return candidate;
+		} catch (err) {
+			this.ctx.logger.warn("models dir not creatable on this node — using local data-dir", { meta: {
+				resolved: candidate,
+				fallback,
+				error: err instanceof Error ? err.message : String(err)
+			} });
+			await node_fs.promises.mkdir(fallback, { recursive: true });
+			return fallback;
+		}
+	}
+	async onInitialize() {
+		const modelsDir = await this.resolveModelsDir();
 		if (!this.ctx.settings) throw new Error("DetectionPipelineAddon: ctx.settings not available");
 		await this.refreshNodeEngineFromStore();
 		this.pythonAddonDir = resolveAddonPythonDir();
@@ -6994,6 +7095,12 @@ var DetectionPipelineAddon = class extends require_dist.BaseAddon {
 		await this.provider.ensureBootEngineProvisioned().catch((err) => {
 			this.ctx.logger.warn("ensureBootEngineProvisioned failed", { meta: { error: err instanceof Error ? err.message : String(err) } });
 		});
+		this.probePhaseUnsub = this.ctx.eventBus?.subscribe({ category: require_dist.EventCategory.PlatformProbePhase }, (event) => {
+			if (event.data?.phase !== "done") return;
+			this.provider.repickEngineOnProbeReady().catch((err) => {
+				this.ctx.logger.warn("repick on platform-probe done failed", { meta: { error: err instanceof Error ? err.message : String(err) } });
+			});
+		}) ?? null;
 		await this.provider.warmPool();
 		this.engineMetricsTimer = setInterval(() => this.emitEngineMetricsSnapshot(), ENGINE_METRICS_SNAPSHOT_INTERVAL_MS);
 		this.lastAppliedPoolConfig = this.snapshotPoolConfig();
@@ -7067,6 +7174,10 @@ var DetectionPipelineAddon = class extends require_dist.BaseAddon {
 		}
 	}
 	async onShutdown() {
+		if (this.probePhaseUnsub) {
+			this.probePhaseUnsub();
+			this.probePhaseUnsub = null;
+		}
 		if (this.engineMetricsTimer) {
 			clearInterval(this.engineMetricsTimer);
 			this.engineMetricsTimer = null;
@@ -7132,10 +7243,7 @@ var DetectionPipelineAddon = class extends require_dist.BaseAddon {
 		} catch (err) {
 			this.ctx.logger.warn("provider shutdown failed during tuning respawn", { meta: { error: err instanceof Error ? err.message : String(err) } });
 		}
-		const modelsDir = await this.ctx.api.storage.resolve.query({
-			location: "models",
-			relativePath: ""
-		}).catch(() => "camstack-data/models");
+		const modelsDir = await this.resolveModelsDir();
 		if (!this.ctx.settings) throw new Error("DetectionPipelineAddon: ctx.settings not available during respawn");
 		const effectiveTuning = this.resolveBackendTuning();
 		this.provider = new DetectionPipelineProvider(this.ctx.settings, modelsDir, this.ctx.logger, this.ctx.eventBus ?? null, () => ({ sections: [] }), {

package/dist/detection-pipeline/index.mjs

@@ -114,7 +114,6 @@ function npuInfoFrom(hw) {
 	return { type };
 }
 function envToHardwareInfo(env) {
-	if (!env.hardware) return null;
 	return {
 		platform: toKnownPlatform(env.platform),
 		arch: toKnownArch(env.arch),
@@ -122,8 +121,8 @@ function envToHardwareInfo(env) {
 		cpuCores: 0,
 		totalRAM_MB: 0,
 		availableRAM_MB: 0,
-		gpu: gpuInfoFrom(env.hardware),
-		npu: npuInfoFrom(env.hardware)
+		gpu: env.hardware ? gpuInfoFrom(env.hardware) : null,
+		npu: env.hardware ? npuInfoFrom(env.hardware) : null
 	};
 }
 function probedToHardwareInfo(hw) {
@@ -4089,6 +4088,17 @@ var ONNX_FLOOR = {
 * instead of duplicated inline.
 */
 function onnxFloorPick() {
+	return ONNX_FLOOR;
+}
+/**
+* The platform-deterministic engine pick computed SYNCHRONOUSLY from this node's
+* own `process.platform`/`arch` alone (no probe): darwin → coreml, else → onnx
+* (gpu-dependent openvino/cuda need the probe and converge via the auto-pick).
+* Used as the ENGINE fallback when a persisted selection is unsupported on this
+* node — so a stale GLOBAL engine config (e.g. the cluster's OpenVINO choice)
+* can never force an impossible engine onto a node whose platform rejects it.
+*/
+function platformDefaultPick() {
 	const pick = pickBestRuntime(runtimeEnvFromProcess(null), null);
 	return {
 		runtime: "python",
@@ -4097,6 +4107,19 @@ function onnxFloorPick() {
 		device: pick.device
 	};
 }
+/**
+* Is `backend` even POSSIBLE on this node's OS/arch (ignoring gpu detail)?
+* coreml ⇒ darwin only; openvino ⇒ x64 non-darwin only; onnx ⇒ anywhere. Used to
+* reject a persisted/global engine choice that the node's PLATFORM fundamentally
+* cannot run (e.g. the cluster's OpenVINO default landing on a Mac) — distinct
+* from the gpu-dependent support (linux without a probed Intel iGPU still keeps
+* openvino as a valid platform choice; the device falls back to cpu).
+*/
+function backendPossibleOnPlatform(backend) {
+	if (backend === "coreml") return process.platform === "darwin";
+	if (backend === "openvino") return process.arch === "x64" && process.platform !== "darwin";
+	return true;
+}
 var DetectionPipelineProvider = class DetectionPipelineProvider {
 	modelsDir;
 	eventBus;
@@ -4341,17 +4364,11 @@ var DetectionPipelineProvider = class DetectionPipelineProvider {
 	*/
 	async autoPickAndPersist() {
 		let hardware = null;
-		let bestBackendHint = null;
 		try {
 			const api = this.addonCtx?.api;
-			if (api) {
-				const caps = await api.platformProbe.getCapabilities.query({ nodeId: this.localProbeNodeId() });
-				hardware = caps?.hardware ?? null;
-				const bs = caps?.bestScore;
-				if (bs && bs.runtime === "python") bestBackendHint = bs.backend;
-			}
+			if (api) hardware = (await api.platformProbe.getCapabilities.query({ nodeId: this.localProbeNodeId() }))?.hardware ?? null;
 		} catch {}
-		const pick = pickBestRuntime(runtimeEnvFromProcess(toProbedHardware(hardware)), bestBackendHint);
+		const pick = pickBestRuntime(runtimeEnvFromProcess(toProbedHardware(hardware)), null);
 		const engine = {
 			runtime: "python",
 			backend: pick.runtimeId,
@@ -4359,8 +4376,8 @@ var DetectionPipelineProvider = class DetectionPipelineProvider {
 			device: pick.device
 		};
 		this.currentEngine = engine;
-		if (!(hardware !== null || bestBackendHint !== null)) {
-			this.log.warn("Auto-pick: probe returned no hardware/hint — using onnx floor WITHOUT persisting", { meta: {
+		if (!(pick.runtimeId !== "onnx" || hardware !== null)) {
+			this.log.info("Auto-pick: onnx floor pending gpu probe — NOT persisting (re-pick on done)", { meta: {
 				backend: pick.runtimeId,
 				device: pick.device
 			} });
@@ -4371,10 +4388,10 @@ var DetectionPipelineProvider = class DetectionPipelineProvider {
 			[nodeEngineKey("engineBackend", apNode)]: pick.runtimeId,
 			[nodeEngineKey("engineDevice", apNode)]: pick.device
 		});
-		this.log.info("Auto-picked engine at first boot", { meta: {
+		this.log.info("Auto-picked engine (platform-deterministic)", { meta: {
 			backend: pick.runtimeId,
 			device: pick.device,
-			hint: bestBackendHint ?? "none"
+			hadProbeHardware: hardware !== null
 		} });
 	}
 	/** Map a backend string to a known RuntimeId, flooring to onnx. */
@@ -4434,7 +4451,9 @@ var DetectionPipelineProvider = class DetectionPipelineProvider {
 		const runtimeId = this.toRuntimeId(engine.backend);
 		const device = engine.device ?? "cpu";
 		const snapshot = this.provisioner.state;
-		if (snapshot.runtimeId === runtimeId && snapshot.device === device && snapshot.state !== "idle") return;
+		const sameSelection = snapshot.runtimeId === runtimeId && snapshot.device === device;
+		if (sameSelection && snapshot.state !== "idle") return;
+		if (!sameSelection) this.currentSteps = null;
 		this.provisioner.select(runtimeId, device);
 	}
 	/**
@@ -4670,13 +4689,52 @@ var DetectionPipelineProvider = class DetectionPipelineProvider {
 	async getDefaultSteps(engine) {
 		return buildDefaultStepTree(engine.format);
 	}
+	/**
+	* Substitute, AT RUNTIME ONLY (never persisted), any step whose configured
+	* model lacks a build for that step's active engine format. A GLOBAL pipeline
+	* config can pin a model that's valid on the cluster's Intel nodes (e.g. the
+	* OpenVINO-only `yolov9t-int8`) but impossible on a CoreML/ONNX node — without
+	* this the node crash-loops `Model "X" has no <format> format`. The operator's
+	* choice stays in the persisted config; this node just runs the smallest
+	* catalog model that DOES have its format. Mirrors the per-node engine
+	* fallback in `loadEngine`. Catalog models only — a custom model that lacks
+	* the format is left as-is (operator's responsibility).
+	*/
+	substituteIncompatibleModels(steps) {
+		const fix = (step) => {
+			const format = step.engine?.format ?? this.currentEngine.format;
+			let modelId = step.modelId;
+			try {
+				const entry = getStepDefinition(step.addonId).models.find((m) => m.id === step.modelId);
+				if (entry && !entry.formats[format]) {
+					const fallback = getDefaultModelForFormat(step.addonId, format);
+					if (fallback !== step.modelId) {
+						this.log.info("Step model lacks engine format — substituting format default (runtime)", { meta: {
+							step: step.addonId,
+							configured: step.modelId,
+							substitute: fallback,
+							format
+						} });
+						modelId = fallback;
+					}
+				}
+			} catch {}
+			const children = step.children?.length ? step.children.map(fix) : step.children;
+			return modelId === step.modelId && children === step.children ? step : {
+				...step,
+				modelId,
+				...children ? { children } : {}
+			};
+		};
+		return steps.map(fix);
+	}
 	async getGlobalSteps() {
 		if (this.currentSteps) return this.currentSteps;
 		const raw = (await this.readStore())[KEY_STEPS];
 		if (!raw) {
 			const defaults = buildDefaultStepTree(this.currentEngine.format);
 			if (defaults.length === 0) return null;
-			this.currentSteps = defaults;
+			this.currentSteps = this.substituteIncompatibleModels(defaults);
 			this.writeStore({ [KEY_STEPS]: JSON.stringify(defaults) });
 			this.log.info("Bootstrapped default pipeline — object-detection + face + plate recognition enabled by default", { meta: { rootSteps: defaults.length } });
 			return this.currentSteps;
@@ -4714,7 +4772,7 @@ var DetectionPipelineProvider = class DetectionPipelineProvider {
 					this.log.info("Migration: added audio-classifier step to persisted pipeline config");
 				}
 			}
-			this.currentSteps = steps;
+			this.currentSteps = this.substituteIncompatibleModels(steps);
 			return this.currentSteps;
 		} catch {
 			throw new Error(`Failed to parse persisted pipeline steps: corrupt data in key "${KEY_STEPS}"`);
@@ -5865,12 +5923,13 @@ var DetectionPipelineProvider = class DetectionPipelineProvider {
 			const storedDevice = typeof storedDeviceRaw === "string" ? storedDeviceRaw : "";
 			const floor = onnxFloorPick();
 			const migratedBackend = typeof storedRuntime === "string" && storedRuntime === "node" && backend === "cpu" ? "onnx" : backend;
-			if (!DetectionPipelineProvider.isPythonBackendAvailable(migratedBackend, this.executorOptions.pythonPath ?? "")) {
-				this.log.warn("Stored engine backend unavailable on this node — falling back to onnx floor", { meta: {
+			if (!backendPossibleOnPlatform(migratedBackend)) {
+				const platformDefault = platformDefaultPick();
+				this.log.warn("Stored engine backend impossible on this platform — using platform default", { meta: {
 					stored: migratedBackend,
-					fallback: `${floor.backend}`
+					fallback: platformDefault.backend
 				} });
-				return floor;
+				return platformDefault;
 			}
 			const device = storedDevice || floor.device;
 			return {
@@ -6042,44 +6101,61 @@ var DetectionPipelineProvider = class DetectionPipelineProvider {
 	}
 	async reprobeEngine() {
 		const api = this.addonCtx?.api;
-		let best;
+		let hardware = null;
 		if (api) try {
-			const caps = await api.platformProbe.getCapabilities.query({ nodeId: this.localProbeNodeId() });
-			const bs = caps?.bestScore;
-			if (bs && bs.runtime === "python") {
-				const probeBackend = bs.backend;
-				const probeDevice = (() => {
-					const hw = caps.hardware;
-					if (probeBackend === "openvino") return defaultDeviceFor("openvino");
-					if (probeBackend === "coreml") return defaultDeviceFor("coreml");
-					if (probeBackend === "onnx") return hw?.gpu?.type === "nvidia" ? "cuda" : "cpu";
-					return "cpu";
-				})();
-				best = {
-					runtime: "python",
-					backend: probeBackend,
-					format: backendToFormat(probeBackend),
-					device: probeDevice
-				};
-			} else best = onnxFloorPick();
-		} catch {
-			best = onnxFloorPick();
-		}
-		else best = onnxFloorPick();
+			hardware = (await api.platformProbe.getCapabilities.query({ nodeId: this.localProbeNodeId() }))?.hardware ?? null;
+		} catch {}
+		const pick = pickBestRuntime(runtimeEnvFromProcess(toProbedHardware(hardware)), null);
+		const best = {
+			runtime: "python",
+			backend: pick.runtimeId,
+			format: modelFormatFor(pick.runtimeId),
+			device: pick.device
+		};
 		const probedLabel = `${best.backend}/${best.device ?? "default"}`;
 		const rpNode = this.localProbeNodeId();
-		await this.writeStore({
-			[nodeEngineKey("probedBestEngine", rpNode)]: probedLabel,
-			[nodeEngineKey("engineBackend", rpNode)]: best.backend,
-			[nodeEngineKey("engineDevice", rpNode)]: best.device ?? "cpu"
-		});
-		this.log.info("Re-probed engine — wrote back engineBackend + engineDevice", { meta: {
+		if (pick.runtimeId !== "onnx" || hardware !== null) {
+			await this.writeStore({
+				[nodeEngineKey("probedBestEngine", rpNode)]: probedLabel,
+				[nodeEngineKey("engineBackend", rpNode)]: best.backend,
+				[nodeEngineKey("engineDevice", rpNode)]: best.device ?? "cpu"
+			});
+			this.log.info("Re-probed engine (platform-deterministic) — wrote back", { meta: {
+				backend: best.backend,
+				device: best.device ?? null,
+				probedBestEngine: probedLabel
+			} });
+		} else this.log.info("Re-probe: onnx floor pending gpu probe — NOT persisting (re-pick on done)", { meta: {
 			backend: best.backend,
-			device: best.device ?? null,
-			probedBestEngine: probedLabel
+			device: best.device ?? null
 		} });
 		return best;
 	}
+	/**
+	* Re-pick the engine when the platform-probe finishes its async hardware +
+	* Python detection (the `platform-probe.phase` `done` event). At boot the
+	* probe's accelerator result may not be ready yet, so the engine floored to
+	* onnx; once the probe answers (e.g. a Mac's CoreML/ANE surfaces after the
+	* embedded Python is installed) this re-runs the probe-driven pick and
+	* re-provisions. Idempotent: `startProvisioningForCurrentEngine` skips a
+	* no-op when the selection is unchanged.
+	*/
+	async repickEngineOnProbeReady() {
+		const before = `${this.currentEngine.backend}/${this.currentEngine.device ?? "default"}`;
+		await this.reprobeEngine();
+		const stored = await this.loadEngine();
+		if (stored) {
+			this.currentEngine = stored;
+			this.needsAutoPick = false;
+			this.cancelDeferredAutoPick();
+		}
+		const after = `${this.currentEngine.backend}/${this.currentEngine.device ?? "default"}`;
+		if (before !== after) this.log.info("Engine re-picked after platform-probe completed", { meta: {
+			before,
+			after
+		} });
+		this.startProvisioningForCurrentEngine();
+	}
 	async getReferenceAudioFiles() {
 		const dir = resolveReferenceAudioDir();
 		if (!dir) return [];
@@ -6547,6 +6623,7 @@ var DetectionPipelineAddon = class extends BaseAddon {
 	nodeEngineBackend = DEFAULT_CONFIG.engineBackend;
 	nodeProbedBestEngine = "";
 	engineMetricsTimer = null;
+	probePhaseUnsub = null;
 	/** Snapshot-equality cache for engine-metrics emit. Most ticks
 	*  the engine inventory is unchanged (no model load/unload), so
 	*  we skip the bus emit and let the heartbeat re-emit at
@@ -6953,11 +7030,35 @@ var DetectionPipelineAddon = class extends BaseAddon {
 			numWorkers: num(t["numWorkers"], 1)
 		};
 	}
-	async onInitialize() {
-		const modelsDir = await this.ctx.api.storage.resolve.query({
+	/**
+	* Resolve the directory models are downloaded into, resilient across nodes.
+	* The `models` storage location is GLOBAL (hub-seeded) and can resolve to a
+	* path that exists only on the hub — e.g. Docker's `/data/models`, which an
+	* agent on a different filesystem (a Mac) cannot create (`ENOENT mkdir
+	* /data/models`). Verify the resolved dir is creatable; otherwise fall back to
+	* this node's LOCAL addon data-dir so models always land on writable disk.
+	*/
+	async resolveModelsDir() {
+		const fallback = path$1.join(this.ctx.dataDir, "models");
+		const candidate = await this.ctx.api.storage.resolve.query({
 			location: "models",
 			relativePath: ""
-		}).catch(() => "camstack-data/models");
+		}).catch(() => null) ?? fallback;
+		try {
+			await fs.promises.mkdir(candidate, { recursive: true });
+			return candidate;
+		} catch (err) {
+			this.ctx.logger.warn("models dir not creatable on this node — using local data-dir", { meta: {
+				resolved: candidate,
+				fallback,
+				error: err instanceof Error ? err.message : String(err)
+			} });
+			await fs.promises.mkdir(fallback, { recursive: true });
+			return fallback;
+		}
+	}
+	async onInitialize() {
+		const modelsDir = await this.resolveModelsDir();
 		if (!this.ctx.settings) throw new Error("DetectionPipelineAddon: ctx.settings not available");
 		await this.refreshNodeEngineFromStore();
 		this.pythonAddonDir = resolveAddonPythonDir();
@@ -6987,6 +7088,12 @@ var DetectionPipelineAddon = class extends BaseAddon {
 		await this.provider.ensureBootEngineProvisioned().catch((err) => {
 			this.ctx.logger.warn("ensureBootEngineProvisioned failed", { meta: { error: err instanceof Error ? err.message : String(err) } });
 		});
+		this.probePhaseUnsub = this.ctx.eventBus?.subscribe({ category: EventCategory.PlatformProbePhase }, (event) => {
+			if (event.data?.phase !== "done") return;
+			this.provider.repickEngineOnProbeReady().catch((err) => {
+				this.ctx.logger.warn("repick on platform-probe done failed", { meta: { error: err instanceof Error ? err.message : String(err) } });
+			});
+		}) ?? null;
 		await this.provider.warmPool();
 		this.engineMetricsTimer = setInterval(() => this.emitEngineMetricsSnapshot(), ENGINE_METRICS_SNAPSHOT_INTERVAL_MS);
 		this.lastAppliedPoolConfig = this.snapshotPoolConfig();
@@ -7060,6 +7167,10 @@ var DetectionPipelineAddon = class extends BaseAddon {
 		}
 	}
 	async onShutdown() {
+		if (this.probePhaseUnsub) {
+			this.probePhaseUnsub();
+			this.probePhaseUnsub = null;
+		}
 		if (this.engineMetricsTimer) {
 			clearInterval(this.engineMetricsTimer);
 			this.engineMetricsTimer = null;
@@ -7125,10 +7236,7 @@ var DetectionPipelineAddon = class extends BaseAddon {
 		} catch (err) {
 			this.ctx.logger.warn("provider shutdown failed during tuning respawn", { meta: { error: err instanceof Error ? err.message : String(err) } });
 		}
-		const modelsDir = await this.ctx.api.storage.resolve.query({
-			location: "models",
-			relativePath: ""
-		}).catch(() => "camstack-data/models");
+		const modelsDir = await this.resolveModelsDir();
 		if (!this.ctx.settings) throw new Error("DetectionPipelineAddon: ctx.settings not available during respawn");
 		const effectiveTuning = this.resolveBackendTuning();
 		this.provider = new DetectionPipelineProvider(this.ctx.settings, modelsDir, this.ctx.logger, this.ctx.eventBus ?? null, () => ({ sections: [] }), {

package/dist/stream-broker/{_virtual_mf-localSharedImportMap___mfe_internal__addon_stream_broker_widgets-Do7lgO8N.mjs → _virtual_mf-localSharedImportMap___mfe_internal__addon_stream_broker_widgets-DLgk22-S.mjs}

@@ -3,7 +3,7 @@ import "./dist-CYZr2fwk.mjs";
 var e = {
 	"@camstack/sdk": {
 		name: "@camstack/sdk",
-		version: "1.1.1",
+		version: "1.1.2",
 		scope: ["default"],
 		loaded: !1,
 		from: "addon_stream_broker_widgets",
@@ -18,7 +18,7 @@ var e = {
 	},
 	"@camstack/types": {
 		name: "@camstack/types",
-		version: "1.1.1",
+		version: "1.1.2",
 		scope: ["default"],
 		loaded: !1,
 		from: "addon_stream_broker_widgets",
@@ -33,7 +33,7 @@ var e = {
 	},
 	"@camstack/ui-library": {
 		name: "@camstack/ui-library",
-		version: "1.1.1",
+		version: "1.1.2",
 		scope: ["default"],
 		loaded: !1,
 		from: "addon_stream_broker_widgets",

package/dist/stream-broker/{hostInit-D5y5VMK8.mjs → hostInit-pRSjUAJj.mjs}

@@ -36,7 +36,7 @@ async function r() {
 				}
 			},
 			"@camstack/types": {
-				version: "1.1.1",
+				version: "1.1.2",
 				scope: "default",
 				shareConfig: {
 					singleton: !0,
@@ -45,7 +45,7 @@ async function r() {
 				}
 			},
 			"@camstack/sdk": {
-				version: "1.1.1",
+				version: "1.1.2",
 				scope: "default",
 				shareConfig: {
 					singleton: !0,
@@ -81,7 +81,7 @@ async function r() {
 				}
 			},
 			"@camstack/ui-library": {
-				version: "1.1.1",
+				version: "1.1.2",
 				scope: "default",
 				shareConfig: {
 					singleton: !0,

package/dist/stream-broker/remoteEntry.js

@@ -30,7 +30,7 @@ async function d(e) {
 	}
 }
 async function f() {
-	return l ||= d(() => import("./_virtual_mf-localSharedImportMap___mfe_internal__addon_stream_broker_widgets-Do7lgO8N.mjs")).catch((e) => {
+	return l ||= d(() => import("./_virtual_mf-localSharedImportMap___mfe_internal__addon_stream_broker_widgets-DLgk22-S.mjs")).catch((e) => {
 		throw l = void 0, e;
 	}), l;
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@camstack/addon-pipeline",
-  "version": "1.1.1",
+  "version": "1.1.2",
   "description": "CamStack Pipeline bundle — runner, detection, motion, decoders, audio + stream broker. Multi-entry npm package shipping 7 addons under a single bundle.",
   "keywords": [
     "camstack",