LENS 11 — CROSS-LENS CONNECTIONS Red Team Brief --- LENS 04 (Constraint Mapping / WiFi cliff) The CTO's WiFi single-point-of-failure attack in Lens 11 directly confirms what Lens 04 identified as the 100ms WiFi cliff. Lens 04 showed the cliff as a physical constraint; Lens 11 shows it as an adversarial attack surface. UPDATE (2026-04-16): Activating the idle Hailo-8 AI HAT+ (26 TOPS, YOLOv8n @ 430 FPS, on-robot, zero WiFi dependency) moves obstacle detection off the WiFi cliff. The cliff still exists for semantic/goal reasoning on Panda. Flag for Lens 04 revision: add a note that naive 2.4 GHz jamming no longer disables the safety layer once Hailo-8 is activated. A sophisticated dual-band jam (2.4 GHz + 5 GHz) remains the full attack surface; both lenses should reference the Orin-NX-native successor as the architectural collapse path. LENS 06 (Moat Analysis) Lens 11 depends critically on the Lens 06 finding that the accumulated household semantic map is the durable competitive asset, not the VLM architecture. Lens 11's "open-source race to zero" card only has a counter because Lens 06 established the map-as-moat thesis. If Lens 06's analysis is wrong — if the map is not actually irreplaceable — then Lens 11's Card 5 counter collapses. The two lenses are load-bearing for each other. Key verification: Lens 06 should quantify how many navigation sessions are required before the semantic map becomes meaningfully differentiated. If the answer is "2 sessions," the moat thesis is weak. If the answer is "100 sessions," the 18-month window claim in Lens 11 is defensible. LENS 07 (Landscape Position) Lens 07 identified Annie as targeting the "edge + rich semantics" quadrant that is currently empty in the competitive landscape — the space that GR00T N1 is also targeting from above (cloud-trained, enterprise-priced). Lens 11's competitor card confirms this collision course. The 6-month competitive window in Lens 11 maps to Lens 07's observation that the quadrant will fill by mid-2026. The strategic implication: Annie's differentiation must shift from "architecture" to "data accumulation" before the quadrant fills. LENS 10 (Cognitive Biases / Decision-Making / Post-Mortem) Lens 10's post-mortem finding that "we built the fast path, forgot the slow path" maps to the evaluation vacuum in Lens 11's CTO card: measured navigation success rate and trajectory error are both planned but not running, so the slow-path failure modes (WiFi jitter, IMU watchdog gap, SLAM prerequisite chain) are invisible to the dashboard. Flag for Lens 10: the same cognitive-bias that hides the slow path hides the absence of evaluation metrics — "if the fast path works, the system works" is the same shortcut that "if a vendor ships an SDK, it fits our problem" would be, at a different layer. LENS 12 (Spectrum / RF Dependence / Comms Architecture) Lens 12 treats home WiFi as Annie's critical comms substrate. Lens 11's adversarial framing sharpens this: a sophisticated adversary targets both 2.4 GHz (semantic link to Panda) and 5 GHz (any backhaul/telemetry) to defeat the Hailo-8 mitigation. Flag for Lens 12: add a scenario for dual-band jam. Consider wired Ethernet fallback (Pi 5 has GbE) to Panda as an anti-jam mitigation for the semantic path — cheaper than an Orin-NX-native robot and closes the attack vector that Hailo-8 alone does not. Also: Hailo-8 activation creates new monitoring metrics (L1 detection recall, L1↔L2 handoff latency, safety-layer-alone survival rate under jam) that Lens 12 should include in its comms-dependence matrix. LENS 13 (Constraint Fragility) The CTO's WiFi attack and the malicious-user WiFi jamming attack both point to the fragility Lens 13 analyzed structurally. Lens 11 provides the adversarial instantiation: the fragility is not just accidental (router reboot) but exploitable (deliberate degradation). Hailo-8 activation drops the fragility score for the safety layer but leaves the goal-reasoning layer at the original score. Recommendation: cite Lens 13's fragility score for the WiFi constraint in any implementation plan for a local fallback planner, and have Lens 13 model the post-Hailo-8 fragility explicitly. LENS 25 (Ecosystem Dependencies / Vendor Lock-in) Large vendors ship multiple parallel, non-integrated stacks — e.g., NVIDIA's catalog partitions across robotics (Isaac ROS / Perceptor), video analytics, and separate VLM-reasoning products, with no automatic bridge between them. A naive consumer of the ecosystem cannot distinguish them without domain expertise. Flag for Lens 25: expand treatment of intra-vendor ecosystem fragmentation — the risk is not only between NVIDIA and non-NVIDIA, but between two products from the same vendor that a builder cannot tell apart without a pre-commitment fit check. LENS 02 (Abstraction Leak) The Pico RP2040 REPL crash identified in Lens 02 as an "invisible abstraction leak" is also an adversarial attack surface: a power glitch or deliberate USB disconnect causes the Pico to drop to REPL, the IMU reports unhealthy, the heading correction layer fails, and the robot executes turns with no drift correction. Lens 11 didn't model this attack explicitly (it focused on WiFi, voice injection, and vendor seduction), but the Pico failure mode is a physical attack vector. --- SYNTHESIS NOTE FOR ASSEMBLY The five cards in Lens 11 are not independent threats — they form a dependency chain: 1. The open-source race (Card 5) commoditizes the architecture. 2. Commoditized architecture means only data matters (Card 5 counter). 3. Data accumulation requires the semantic map to be built (Phase 2c). 4. Phase 2c requires Phase 1 SLAM to be stable and evaluated (CTO Card 3 gap). 5. SLAM stability requires the WiFi dependency to be hardened (CTO Card 2 gap, Malicious User Card 2). 6. WiFi hardening is partially achieved by Hailo-8 activation (safety layer) but requires local fallback planning or wired Ethernet for the semantic layer. The chain terminates at the highest-leverage near-term fix: Hailo-8 activation + local fallback planning + wired Ethernet fallback.