ZeroStack (zerostack) — Release Zero (0.9.9-beta.1)

A zero-stall, deterministic O(1) Zigbee 3.0 / R23 coordinator & router firmware built exclusively for SMLIGHT devices.

Release Zero (0.9.9-beta.1) is the public beta release of SMLIGHT ZeroStack, a ground-up, deterministic O(1) Zigbee 3.0 / PRO R23 coordinator and router firmware built exclusively for SMLIGHT CC2652P, CC2652P7 (CC1352P7), and CC2674P10 adapters (SLZB-06, SLZB-06U, SLZB-06P7 / SLZB-06P7v1, SLZB-06P7U, SLZB-07P7, SLZB-06P10, and SLZB-07P10).

• Official Product & Firmware Page: https://smlight.tech/software/zerostack
• Interactive Z2M / ZHA Crash & Airtime Analyzer: https://smlight-tech.github.io/zerostack/
• Technical Whitepaper — Part 1 (PDF): Scaling Zigbee 3.0 / R23 Beyond the 100-Node Wall — Deterministic O(1) Architecture vs. Legacy O(N) Stack Scaling
• AI Agent Model Context Protocol (MCP) Server: github.com/zerostack-rf/zerostack-mcp

---

Core Engineering Pillars in Release Zero (0.9.9-beta.1)

• Zero Lookup Delays (O(1) Security & Key Engine): Eliminates O(N⋅P) linear flash log scans on every frame. Hardware DWT->CYCCNT cycle measurements on CC1352P7 @ 48 MHz verify 134.4× faster per-frame NWK encryption (82.7 µs vs 11.11 ms at D=150, 0 flash reads), 139.2× faster NWK decryption (79.7 µs vs 11.09 ms), 135× higher routed hop security ceiling (>6,100 pkt/s vs ~45 pkt/s), and 4,559× faster Trust Center Link Key (TCLK) lookup (6.8 µs vs 30.96 ms).
• Zero Unnecessary Flash Wear & 14.7× Lower Compaction Rate: Idempotent non-volatile state writes complete in 3,770 cycles (78.5 µs) with 0 bytes written to flash on unchanged records (26.7× faster than legacy 100,714 cycles), delaying flash page compaction from write #69 (163.9 ms stall) to write #1,013 (13.7 ms stall).
• Zero Brownout Corruption (Verified Across 22,959 Hardware Power-Cut Points): Fault-injected across 22,959 NOR flash power-cut points (22,565 mid-compaction + 394 record-update cuts) and 2,000,000 shadow-fuzz operations with 0 lost or corrupted records, 0 NOR write-verify conflicts, and 749× lower post-brownout reboot flash write volume (27 B written, 0 page erases vs 20,232 B and 62 page erases on legacy Z-Stack).
• Zero Remote OTA HardFaults: Legacy Z-Stack suffers a deterministic remote CPU HardFault (Precise BusFault after 143,345–144,412 cycles) when receiving malformed over-the-air secured NWK frames. ZeroStack achieves 0 crashes across all 12 malformed NWK/APS fault classes and 3,255 MAC boundary vectors, rejecting invalid frames in 320–6,631 cycles (6.7–138 µs, up to 402× lower CPU cost) and surviving a 1,000-frame cryptographic replay flood 92.8× faster (32.4 ms vs 2.99 s).
• Zero Serial Drops & Live Auto-Baud (115200 / 460800 / 1000000): Under hostile UART framing abuse (300 × SOF delimiter storms, random noise, truncated frames), legacy Z-Stack suffers an 8.84-second serial blackout (44 lost pings). ZeroStack loses 0 pings across all 6 scenarios (3–55 ms wire-speed recovery), cuts SYS_PING reply-path cycles by 83% (816 vs 4,921 cycles), and automatically negotiates 115200, 460800, or 1000000 baud on the fly.
• Zero Traffic Jams & Broadcast Storm Immunity: Expands the Zigbee Broadcast Transaction Table (BTT) by +33% (40 vs 30 active slots), suppresses a 50-router (1,000-frame) broadcast echo storm 20.3× faster per echo (147 µs p50 vs 3.00 ms; 151.16 ms vs 3.00 s total CPU — 19.9× lower), and completes a 64-frame post-outage broadcast flood in 18.87 ms vs 200.25 ms (10.6× lower CPU).
• Zero Memory Fragmentation (O(1) Constant-Time Allocator): Replaces OSAL_Memory.c linear free-list walks with a deterministic O(1) allocator (218 CPU cycles = 4.5 µs regardless of heap fragmentation — 11.7× faster at 64 heap holes), heap-independent timer start (1,448 cycles = 30.2 µs — 6.9× faster), and guaranteed 100% heap recovery (Δheap = 0 B) from 0 B free-heap saturation.
• Zero Battery-Device Bottlenecks: Strict per-child and global indirect queue quota isolation prevents offline or slow-polling battery sensors from exhausting the coordinator's nwkDB buffer pool (0x11: BUFFER_FULL), while O(1) in-RAM security (82.7 µs NWK encrypt, 0 flash reads) ensures queued indirect frames are dequeued immediately while a polling ZED's receive window is open.
• Zero Wasted Memory: Custom-tuned RAM and routing tables per SMLIGHT silicon profile (88 KB on CC2652P, 144 KB on CC2652P7, 256 KB on CC2674P10), scaling Many-to-One Concentrator source routing up to 200–300+ routes.

---

Why Stock Coordinator Firmware Fails on Large Meshes

Production smart-home and commercial IoT networks (100–300+ nodes across Zigbee2MQTT and Home Assistant ZHA) routinely push legacy reference firmware beyond its architectural scaling limits:

1. O(N⋅P) Flash Security Walks & SREQ/SRSP Serial Lockups (SRSP - AF - dataRequestExt after 6000ms & SRSP - SYS - ping after 6000ms):
   • At mesh depth D=150 with an aged NV flash log, stock TI Z-Stack executes a linear flash scan on every secured frame: 533,156 CPU cycles (11.11 ms) per NWK encrypt and 532,182 cycles (11.09 ms) per NWK decrypt, capping routed hop security at ~45 pkt/s. A single unjoined device's Trust Center Link Key (TCLK) lookup miss scans all 300 NV slots across flash (1,486,228 cycles = 30.96 ms), and a 50-device join burst stalls the coordinator for 1.70 seconds.
   • On container restart or hostile UART noise, NVOCMP flash sector compaction (163.9 ms per page with interrupts masked, triggered every 69 writes) or NPI parser desynchronization causes an 8.84-second serial blackout (44 consecutive lost pings), tripping Zigbee2MQTT's 6,000 ms watchdog.
2. Variable-Block Heap Fragmentation & Table Saturation (0x10: MEM_ERROR, 0x11: BUFFER_FULL, 0xc7: NWK_TABLE_FULL, 0xb2: APS_TABLE_FULL, 0xcd: NWK_NO_ROUTE):
   • Stock OSAL_Memory.c walks fragmented free lists in O(N) time (2,550 cycles at 64 heap holes) and allocates timers from the same shared heap (10,034 cycles = 209.0 µs at 39 active timers). Under concurrent unicast bursts, source-routed payloads, and unpolled sleepy ZED indirect frames, contiguous allocations fail with 0x10 (MEM_ERROR) while 50-router broadcast echo storms lock up the CPU for 3.00 seconds (143,865 cycles/echo) and overflow the 30-slot Broadcast Transaction Table (0xcd: NWK_NO_ROUTE).
3. Remote Over-the-Air HardFault & Replay-Flood Lockups (SYS_RESET_IND):
   • Unauthenticated malformed secured NWK frames (truncated AuxHeader, Payload, or MIC) cause a deterministic remote CPU HardFault (Precise BusFault) on legacy Z-Stack after 143,345–144,412 cycles (~3.0 ms), spontaneously rebooting the coordinator (AREQ - SYS - resetInd). A 1,000-frame replay flood burns 143,802 cycles (3.0 ms) per frame (2.99 s total CPU freeze).
4. Indirect Queue Starvation & Sleepy Poll Timing Misses (0xe9: MAC_NO_ACK / 233, 0xf0: MAC_TRANSACTION_EXPIRED / 240, 0xf1: MAC_TRANSACTION_OVERFLOW / 241):
   • On SimpleLink F2, the Cortex-M0 RF Core provides 32 hardware short-address SrcMatch slots in RFC_RAM to set Frame Pending (FP=1) in the 192 µs hardware auto-ACK, while the upper MAC on the main CPU must dequeue, encrypt, and transmit the held indirect packet before the sleepy ZED's poll wait window closes. When legacy Z-Stack stalls for 11.11 ms on flash security reads, 30.96 ms on TCLK misses, or 163.9 ms on NV compactions—or when unresponsive battery sensors monopolize nwkDB indirect buffers—indirect deliveries time out with 0xe9 / 0xf0 / 0xf1.
5. Sub-MAC State Desynchronization (0xe1: MAC_CHANNEL_ACCESS_FAILURE / 225):
   • Following an Energy Detect (ED) scan or channel switch, RF_FsmEventLastCommandDone SWI races can leave background CMD_IEEE_RX terminated (DONE_ABORT = 0x0807) while upper-MAC CSMA flags remain desynchronized, causing every subsequent CCA check to fail with 0xe1 (225).

Note on Silicon Labs EFR32 (EmberZNet 0x72, 0xa1, 0x66, ASH_NCP_FATAL_ERROR): Our diagnostic catalog and interactive analyzer also decode Silicon Labs EFR32 / EZSP / ASHv2 status codes for multi-vendor fleet triage. ZeroStack Release Zero (0.9.9-beta.1) ships for SMLIGHT TI SimpleLink F2 coordinators below; EFR32 ZeroStack support is in active development.

---

Hardware Support & Benchmark Summary

Supported SMLIGHT Hardware Models (Release Zero 0.9.9-beta.1)

SMLIGHT Model	Target Radio SoC	CPU Core & Clock	Flash / SRAM	Hardware Flow Control (RTS/CTS)	Supported UART Baud Rates	ZeroStack Capacity & Tuning
SLZB-06	TI CC2652P	ARM Cortex-M4F @ 48 MHz + Cortex-M0 CPE	352 KB Flash / 88 KB SRAM	No (rtscts: false)	Live Auto-Baud (115200 / 460800 / 1000000)	200+ source routes, indirect queue quota isolation
SLZB-06U	TI CC2652P	ARM Cortex-M4F @ 48 MHz + Cortex-M0 CPE	352 KB Flash / 88 KB SRAM	Yes (DIO10 RTS / DIO11 CTS)	Live Auto-Baud (115200 / 460800 / 1000000)	200+ source routes, indirect queue quota isolation
SLZB-06P7 / SLZB-06P7v1	TI CC2652P7 (CC1352P7)	ARM Cortex-M4F @ 48 MHz + Cortex-M0 CPE	704 KB Flash / 144 KB SRAM	No (rtscts: false)	Live Auto-Baud (115200 / 460800 / 1000000)	250+ source routes, 300 TCLK slots, indirect queue isolation
SLZB-06P7U	TI CC2652P7 (CC1352P7)	ARM Cortex-M4F @ 48 MHz + Cortex-M0 CPE	704 KB Flash / 144 KB SRAM	Yes (DIO10 RTS / DIO11 CTS)	Live Auto-Baud (115200 / 460800 / 1000000)	250+ source routes, 300 TCLK slots, indirect queue isolation
SLZB-07P7	TI CC2652P7	ARM Cortex-M4F @ 48 MHz + Cortex-M0 CPE	704 KB Flash / 144 KB SRAM	No (rtscts: false)	Live Auto-Baud (115200 / 460800 / 1000000)	250+ source routes, 300 TCLK slots, indirect queue isolation
SLZB-06P10	TI CC2674P10	ARM Cortex-M33 @ 48 MHz + Cortex-M0 CPE	1024 KB Flash / 256 KB SRAM	Yes (DIO10 RTS / DIO11 CTS)	Live Auto-Baud (115200 / 460800 / 1000000)	300+ source routes, 300 TCLK slots, indirect queue isolation
SLZB-07P10	TI CC2674P10	ARM Cortex-M33 @ 48 MHz + Cortex-M0 CPE	1024 KB Flash / 256 KB SRAM	No (rtscts: false)	Live Auto-Baud (115200 / 460800 / 1000000)	300+ source routes, 300 TCLK slots, indirect queue isolation

---

Hardware-Verified Benchmarks (Stock Z-Stack vs. SMLIGHT ZeroStack)

All measurements below were captured directly on target silicon (TI SimpleLink CC1352P7 / CC2652P7 @ 48.0 MHz, 1 µs = 48 CPU cycles) using the ARM Cortex-M4F hardware cycle counter (DWT->CYCCNT).

Part 1: Deterministic O(1) Execution Latency & Scaling (Whitepaper Part 1)

Subsystem / Operation (Mesh Scale D=150..300)	Stock TI Z-Stack (DWT->CYCCNT)	SMLIGHT ZeroStack (Release Zero 0.9.9-beta.1)	Measured Speedup / Scaling Law
TCLK Lookup Miss (300-entry table, first contact)	1,486,228 cyc (30.96 ms)	326 cyc (6.8 µs) (0 flash reads)	4,559× faster (O(1) vs O(N⋅P))
TCLK Lookup Hit (300-entry table, oldest record age 383)	620,396 cyc (12.92 ms)	324 cyc (6.75 µs) (0 flash reads)	1,915× faster (Slot 0 fresh: 7.1×, Slot 299 fresh: 9.1×)
Per-Frame NWK Security — Encrypt (D=150)	533,156 cyc (11.11 ms)	3,968 cyc (82.7 µs) (0 flash reads)	134.4× faster (O(1) flat; 2.15× at D=0)
Per-Frame NWK Security — Decrypt (D=150)	532,182 cyc (11.09 ms)	3,824 cyc (79.7 µs) (0 flash reads)	139.2× faster (O(1) flat)
Routed Hop Security Ceiling (D=150, Decrypt + Encrypt)	1,065,338 cyc (22.19 ms, ~45 pkt/s max)	7,792 cyc (162 µs, >6,100 pkt/s max)	135× higher routed packet ceiling
Per-Frame APS TX Security (300-peer table, D=0..100)	52,909 – 409,301 cyc (up to 8.53 ms)	5,410 – 5,424 cyc (113 µs)	Up to 75.5× faster
Device Authentication Check (Slot 0 -> 299)	1,656 -> 42,243 cyc (up to 880 µs)	181 -> 179 cyc (3.7 µs)	Up to 236× faster
NVM Stack Key Read (D=0 → 150)	2,347 -> 64,817 cyc (up to 1.35 ms)	302 – 303 cyc (6.3 µs)	Up to 213.9× faster
Idempotent NVM State Write (Unchanged 100 B record)	100,714 cyc (2.10 ms, writes flash 48/48 times)	3,770 cyc (78.5 µs, 0 B flash writes 48/48)	26.7× faster + 0% unnecessary flash wear
NVM Flash Compaction Frequency & Stall	First compaction at write #69 (163.9 ms stall)	First compaction at write #1,013 (13.7 ms stall)	14.7× lower compaction rate (12× shorter stall)
Dynamic Memory Allocation (malloc @ k=64 heap holes)	2,550 cyc (53.1 µs, O(k) hole walk)	218 cyc (4.5 µs, O(1) constant time)	11.7× faster (0% fragmentation penalty)
Timer Start Latency (T=39 active timers, 64 heap holes)	10,034 cyc (209.0 µs, heap-coupled)	1,448 cyc (30.2 µs, heap-independent)	6.9× faster
Inter-Task Event Enqueue (msgSend, Q=64 queued events)	935 cyc (19.5 µs, O(Q) walk)	322 cyc (6.7 µs, O(1) flat)	2.9× faster
Host Serial SYS_PING Round-Trip (115,200 baud)	9,790 cyc (4,921 cyc reply path)	4,618 cyc (816 cyc reply path)	53% fewer total cycles (83% faster reply path)

Part 2: Adversarial Resilience, Brownout & Fault-Injection (Adversarial & Fault-Injection Suite)

Adversarial Stress / Fault Vector	Stock TI Z-Stack Behavior	SMLIGHT ZeroStack (Release Zero 0.9.9-beta.1)	Resilience Advantage
Malformed Secured OTA NWK/APS Frames (12 fault classes + 3,255 MAC vectors)	Remote CPU HardFault (Precise BusFault) after 143,345–144,412 cyc (~3.0 ms); accepts malformed routing headers (144,127 cyc)	0 crashes across all 12 classes & 3,255 MAC vectors; rejected before crypto in 320–6,631 cyc (6.7–138 µs) (790 cyc on bad route header)	Remote HardFault eliminated (up to 402× lower CPU)
1,000-Frame Cryptographic Replay Flood (D=150)	143,802 cyc (3.0 ms) per replay (2.99 s total CPU lockup)	1,550 cyc (32 µs) per replay (32.4 ms total CPU)	92.8× faster replay flood rejection
Authenticated Header / Ciphertext / MIC Tamper (D=150)	143,058 – 143,065 cyc (2.98 ms) per frame	4,760 – 4,771 cyc (99 µs) per frame	30.0× lower CPU cost on tamper rejection
NOR Flash Brownout Fuzzing (22,959 hardware power cuts + 2M shadow ops)	1 NOR write-verify conflict; torn record triggers 62-page erase recovery (20,232 B written)	0 lost/corrupted records, 0 NOR conflicts; torn record recovers in 27 B written (0 page erases)	749× lower post-brownout reboot flash I/O
Hostile UART Serial Framing Abuse (6 scenarios: 300 × SOF, random noise, truncated)	8.84-second serial blackout (44 consecutive lost pings) on 300 × SOF / random noise; 1 lost ping (203 ms) on truncated frame	0 lost pings across all 6 scenarios (3–55 ms wire-speed recovery; 29 ms on 300 × SOF, 3 ms on truncated; Δheap = 0 B)	Zero serial blackouts (0 lost pings vs 44)
50-Device Unauthenticated Join / TCLK Miss Flood (D=300)	33.9 ms per miss (1.70 s total control-plane stall)	4 µs (208 cyc) on-CPU (0.22 s total wire time)	7.7× faster join-flood recovery
50-Router Broadcast Echo Storm (1,000 echoes at D=150)	143,865 cyc (3.00 ms) per echo (3.00 s total CPU)	7,076 cyc (147 µs p50) per echo (151.16 ms total CPU)	19.9× lower total CPU (20.3× faster per echo)
64-Frame Post-Outage Broadcast Flood (BTT Saturation)	Admits 30 / 64 frames; overflow drop costs 3.01 ms; 200.25 ms total CPU	Admits 40 / 64 frames (+33% BTT); overflow drop in 174 µs (8,341 cyc); 18.87 ms total CPU	10.6× lower flood CPU & +33% broadcast capacity
Heap (0 B free) & Timer Exhaustion Torture	Uncapped timer walk (255 timers); variable-block fragmentation	Bounded timer cap (194 timers), 5 / 5 ping survival at 0 B heap, and 100% heap recovery (Δheap = 0 B)	Deterministic survival under 100% resource exhaustion

---

AI-Assisted Log Decoder & Diagnostic Index (diagnostics/)

If your Zigbee2MQTT or Home Assistant ZHA instance is experiencing crashes, timeouts, or device drops on stock firmware, use the Interactive Z2M / ZHA Crash & Airtime Analyzer, visit https://smlight.tech/software/zerostack, or consult the 5 canonical Zigbee root-cause engineering guides below:

1. SRSP - AF - dataRequestExt after 6000ms, TCLK Interview Timeouts & SREQ Lockups
   • Symptom: Zigbee2MQTT freezes for 6 seconds per command (Error: SRSP - AF - dataRequestExt after 6000ms), or device pairing interviews fail during TCLK exchange (0xa7 / 0xad).
   • Stock Root Cause: O(N⋅P) linear flash security walks (11.11 ms per NWK encrypt, 30.96 ms per TCLK lookup miss at D=150..300), capping routed security at ~45 pkt/s, compounded by 163.9 ms NVOCMP flash compactions and 3.00 s broadcast storm stalls.
   • ZeroStack Remediation: O(1) per-frame NWK security in 82.7 µs encrypt / 79.7 µs decrypt (134.4×–139.2× faster, >6,100 pkt/s ceiling), 4,559× faster TCLK lookup (6.8 µs vs 30.96 ms), and 83% faster serial reply path (816 vs 4,921 cycles).
2. 0x10: MEM_ERROR, 0x11: BUFFER_FULL, 0xc7: NWK_TABLE_FULL, 0xb2: APS_TABLE_FULL, 0xcd: NWK_NO_ROUTE & Ember 0x72 / 0xa1
   • Symptom: Coordinator rejects outbound packets with 0x10 (MEM_ERROR), 0x11 (BUFFER_FULL), 0xc7 (NWK_TABLE_FULL), 0xb2 (APS_TABLE_FULL), 0xcd (NWK_NO_ROUTE / broadcast storm saturation), or Silicon Labs Ember 0x72 / 0xa1.
   • Stock Root Cause: OSAL_Memory.c variable-block heap fragmentation (2,550 cycles at 64 holes), heap-coupled timers (209.0 µs), uncollected indirect MAC buffers hoarded by offline battery sensors, and 30-slot BTT broadcast exhaustion (3.00 s CPU lockup under 50-router echo storms).
   • ZeroStack Remediation: Constant-time O(1) memory allocation in 218 cycles (11.7× faster), heap-independent timers (30.2 µs, 6.9× faster), +33% BTT capacity (40 slots) with 19.9× lower broadcast storm CPU (151.16 ms vs 3.00 s), and custom-tuned 200–300+ entry routing tables.
3. MAC_NO_ACK (0xe9 / 233), 0xf0: MAC_TRANSACTION_EXPIRED (240), 0xf1: MAC_TRANSACTION_OVERFLOW (241) & Sleepy Aqara / Tuya Child Drops
   • Symptom: Battery-powered sensors (Aqara, Sonoff, Tuya, IKEA) drop off the mesh or fail to receive commands with MAC_NO_ACK (233 / 0xe9), MAC_TRANSACTION_EXPIRED (240 / 0xf0), or MAC_TRANSACTION_OVERFLOW (241 / 0xf1).
   • Stock Root Cause: Multi-millisecond flash security walks (11.11 ms per NWK frame, 30.96 ms per TCLK miss) and 163.9 ms NV compactions delay indirect frame dequeue past a polling ZED's receive window, while unpolled indirect frames monopolize the shared nwkDB buffer pool.
   • ZeroStack Remediation: O(1) in-RAM security (82.7 µs NWK encrypt, 6.8 µs TCLK lookup, 0 flash reads) for sub-millisecond indirect poll turnaround, paired with per-child and global indirect queue quota isolation (Zero Battery-Device Bottlenecks).
4. MAC_CHANNEL_ACCESS_FAILURE (0xe1 / 225) & Stuck CCA/CSMA
   • Symptom: All outbound transmissions suddenly fail with MAC_CHANNEL_ACCESS_FAILURE (0xe1 / 225) even in low-interference RF environments.
   • Stock Root Cause: Sub-MAC CMD_IEEE_RX (0x2401) / CMD_IEEE_CSMA state machine desynchronization following an Energy Detect (ED) scan or frequency synthesizer recalibration when RF_FsmEventLastCommandDone races with foreground command termination.
   • ZeroStack Remediation: Self-healing Sub-MAC receiver and synthesizer state recovery (0 crashes across 3,255 MAC boundary/fuzz vectors).
5. Failed to connect to the adapter (Error: SRSP - SYS - ping after 6000ms) & SYS_RESET_IND
   • Symptom: Restarting Zigbee2MQTT or ZHA fails with SRSP - SYS - ping after 6000ms, or the coordinator spontaneously reboots mid-operation (SYS_RESET_IND).
   • Stock Root Cause: Blocking NVOCMP flash compactions (163.9 ms per page, or 20,232 B / 62 page erases after a brownout), 8.84-second NPI serial blackouts (44 lost pings) on delimiter noise, or remote CPU HardFault crashes on malformed OTA frames.
   • ZeroStack Remediation: 0 lost pings (3–55 ms recovery) under hostile UART framing, live auto-baud (115200 / 460800 / 1000000), 749× lower post-brownout flash writes (27 B vs 20,232 B), and 0 crashes across all 12 malformed OTA frame classes.

---

Silicon Engineering Notes (silicon-notes/)

• TI SimpleLink F2 (CC2652P, CC2652P7 / CC1352P7, CC2674P10) RF Core SWI & Hardware SrcMatch Errata
  • Hardware-level root-cause analysis of stock TI Z-Stack RF_FsmEventLastCommandDone Software Interrupt starvation, nested critical section hazards (HwiP_disable across RF doorbell commands and O(N) OSAL_Memory walks), and CMD_IEEE_MOD_SRC_MATCH ContextError (0x85) failures during 192 µs hardware auto-ACK turnaround windows.

---

Interactive Web Analyzer, AI Agent MCP & LLM Integration

ZeroStack is built for automated diagnostics by both human engineers and AI coding/operations agents:

• Official ZeroStack Product & Firmware Portal: Overview, supported SMLIGHT coordinators, benchmarks, and firmware flashing instructions.
• Interactive Z2M / ZHA Crash & Airtime Analyzer: Standalone browser-based multi-protocol log & airtime analyzer (docs/index.html & docs/diagnosticsCatalog.json).
• zerostack-mcp: Official Model Context Protocol (MCP) server for Claude Desktop, Claude Code, Cursor, Windsurf, and VS Code. Decodes Zigbee2MQTT/ZHA logs, inspects TI SimpleLink / EFR32 registers, and evaluates coordinator capacity.
• llms.txt & llms-full.txt: Standardized machine-readable indices mapped for LLM search crawlers linking verbatim error strings to Stock Firmware Root Causes and ZeroStack Release Zero (0.9.9-beta.1) hardware-verified remediations.

---

Links & Resources

• SMLIGHT ZeroStack Official Page: https://smlight.tech/software/zerostack
• Interactive Z2M / ZHA Crash & Airtime Analyzer: https://smlight-tech.github.io/zerostack/
• ZeroStack MCP Diagnostic Server: https://github.com/zerostack-rf/zerostack-mcp
• SMLIGHT Official Hardware: https://smlight.tech
• SLZB-OS & Web Flasher Documentation: https://smlight.tech/manual/slzb-06/

---

Copyright & Licensing

Copyright (c) 2026 SMLIGHT (https://smlight.tech).

• Documentation & Diagnostic Web Assets (LICENSE / NOTICE): All Markdown engineering guides, JSON diagnostic catalogs (diagnosticsCatalog.json), LLM indices (llms.txt, llms-full.txt), and client-side web analyzer assets (docs/) in this repository are openly licensed under the Apache License 2.0 (Apache-2.0).
• Compiled ZeroStack Firmware Binaries (FIRMWARE_LICENSE.md): All compiled SMLIGHT ZeroStack coordinator and router firmware images/binaries attached to releases or distributed by SMLIGHT are proprietary software governed by FIRMWARE_LICENSE.md (Copyright (c) 2026 SMLIGHT. All Rights Reserved.).