LENS 13 — CROSS-LENS CONVERGENCE POINTS LENS 13 (Constraint Analysis) is the structural backbone of the entire analysis. It identifies WHERE and WHY the system is fragile, which every other lens either discovers independently or builds upon. --- CONVERGENCE WITH LENS 01 (Temporal Geometry) Lens 01 identified temporal surplus as Annie's primary free resource: at 58Hz VLM and 1m/s travel, each frame captures <2cm of displacement, leaving substantial decision headroom. Lens 13 reveals that the WiFi-IMU-overshoot compounding cluster is the mechanism by which that surplus is annihilated. When all three constraints fail simultaneously, temporal surplus goes from ~400ms to effectively zero: stale commands arrive late to a robot spinning past its target with no correction signal available. Lens 01's optimism about temporal headroom must be qualified by Lens 13's compounding failure analysis. The surplus exists in the median case; it does not exist in the tail cases that matter most. Lens 01 identified constraint hierarchies — physics→convention→dissolved. Lens 13 maps where each of Annie's nine constraints sits in that hierarchy: glass surfaces are physics (unremovable), SLAM prerequisite is convention (temporary architectural decision), single-camera FOV is dissolved (artificial, removable in 30 minutes). The three categories from Lens 01 directly predict which of Lens 13's constraints yield to effort and which do not. --- CONVERGENCE WITH LENS 04 (Sensitivity Surface) Lens 04 identified WiFi latency as the parameter with a cliff edge at ~100ms P95, beyond which multiple latency spikes compound within a single nav cycle. Lens 13 elaborates on WHY this constraint is so dangerous: it is the only constraint that is both HIGH fragility AND uncontrollable. Every other HIGH-fragility constraint (glass surfaces, no encoders, Pico IMU crashes, motor overshoot) either has a hardware fix path or an architectural workaround. WiFi has neither — household RF is shared infrastructure outside the system's ownership boundary. Lens 04 found that VLM frame rate above 15Hz is surprisingly insensitive. Lens 13 confirms that the multi-query pipeline's value (diverse queries across alternating frames) does not depend on any of the nine constraints being resolved — it works within the existing constraint envelope. This makes Phase 2a/2b the highest-confidence roadmap items: they improve the system without touching any fragile constraint. --- CONVERGENCE WITH LENS 10 (Failure Pre-Mortem) Lens 10 named WiFi as the "boring" production failure mode — looks fine in testing on a clear channel, causes mysterious incidents when a microwave or neighboring network is active. Lens 13 provides the mechanism: WiFi is the only HIGH-fragility constraint that can spike from healthy to catastrophic in milliseconds without any local indicator. Lens 10's pre-mortem named glass doors as an unresolved hazard; Lens 13 maps this as the most fundamentally hard constraint in the entire matrix — dual sensor failure with zero fallback and no technology path to resolution without physical hardware change. The compounding failure scenario (WiFi spike + IMU crash + motor overshoot, all simultaneously) was partially articulated in Lens 10 but not fully named. Lens 13's matrix makes it explicit: the three constraints interact because IMU crash forces open-loop fallback (the overshoot regime), and WiFi spike means commands arrive after the robot has already overshot. The policy prescription — ESTOP if IMU absent AND WiFi P95 > 80ms — emerges from this analysis. --- CONVERGENCE WITH LENS 11 (Adversarial Analysis) Lens 11 identified glass door as the highest-probability unresolved safety issue in adversarial testing. Lens 13 provides the constraint-level explanation: this is the only scenario where the "VLM proposes, lidar disposes" fusion rule fails catastrophically because BOTH channels agree on the wrong answer. VLM may correctly identify "glass door" from visual context clues (frame edges, handle, partial reflection), but lidar reads "clear" and the safety daemon has no basis to ESTOP. The dual-sensor simultaneous failure means no amount of algorithmic improvement fixes this within the current hardware envelope. --- CONVERGENCE WITH LENS 03 (Dependency Map — llama-server embedding blocker) Lens 03 identified the llama-server embedding blocker as the highest-leverage addressable dependency: unblocks Phase 2d (visual place recognition), semantic map annotation, and loop closure assist — all with a single 2-day implementation task (SigLIP 2 as separate extractor). Lens 13 confirms this assessment: the llama-server constraint is MEDIUM fragility and has a WORKAROUND that is a clean architectural separation (not a hack), with a clear technology relaxation path (llama.cpp PR in review). It is one of only two constraints in the matrix rated both MEDIUM fragility AND HIGH removability — making it the best combination of urgent and fixable. The Hailo-8 activation discovery from the session-119 hardware audit creates a new Lens 03 ↔ Lens 13 coupling: if L1 safety detection (YOLOv8n) moves to the Hailo-8 NPU on the Pi 5, approximately 800 MB of Panda VRAM is freed — which is exactly the SigLIP 2 budget Lens 03 identified. In other words, activating Hailo-8 doesn't just relax the WiFi constraint, it also relaxes the VRAM constraint enough to unblock the Lens 03 dependency. Two constraint relaxations from a single hardware activation — but the cost is adopting the HailoRT runtime as a parallel stack to maintain. --- CONVERGENCE WITH LENS 15 (Constraint Relaxation) Lens 15 analyzed what changes if specific constraints are lifted. Lens 13 provides the prerequisite map for which relaxations are available now versus blocked. Motor overshoot (one-session firmware fix) and llama-server embeddings (two-day SigLIP 2 deploy) are available NOW. SLAM prerequisite is one deploy away (Zenoh fix pending). Glass surfaces require a hardware purchase (~$100 ToF sensor). No wheel encoders requires a motor retrofit (~$40). Lens 15's "3 constraints relaxable NOW for <$200" finding aligns with Lens 13's identification of motor overshoot, llama-server API, and glass detection as the highest-value, lowest-cost fixes in the matrix. --- CENTRAL FINDING FOR SYNTHESIS Lens 13 reveals a structural asymmetry that runs through the entire research: the constraints most amenable to technology relaxation (VRAM, model size, API limits) are the ones least urgently dangerous, while the constraints most urgently dangerous (WiFi jitter, Pico crash, glass) are the ones technology cannot fix — or requires physical hardware changes to address. The system's production readiness depends not on waiting for better models but on addressing the constraints that sit outside the software improvement curve. This is the constraint analysis version of OK-Robot's finding: "What really matters is not fancy models but clean integration" — except applied to failure modes rather than success cases. The session-119 hardware audit surfaced the Hailo-8 on the Pi 5 as the highest-leverage idle resource in the current inventory: 26 TOPS on-board, untouched for navigation, validated by IROS arXiv 2601.21506 as the System 1 half of a dual-process nav loop (66% latency reduction, 67.5% success vs 5.83% VLM-only). Activating Hailo-8 as the L1 reflex layer is the single hardware move that partially relaxes both the WiFi and VRAM constraints at once.