Boundary-layer Sliding Mode Control (SMC) has been applied to vision-based stabilization, but identifying its activation condition against simpler causal baselines (LPF, PID) has been elusive. The operational condition is plant binding: the disturbance velocity approaches or exceeds the actuator slew limit, so the bounded per-step actuation budget binds. We validate this in two independent vision domains: handheld translation stabilization (12 1080p clips; LPF, PID, SMC, L1) and angular re-framing against a virtual slew-rate-limited pan-tilt-zoom (PTZ) plant. Outside binding, leave-one-out cross-validation makes SMC and LPF on translation statistically indistinguishable, and PID beats SMC on the angular 99th-percentile pointing offset by 5.85 degrees (95% CI clear of zero); inside binding -- reached only by sweeping the simulated slew limit, as both captured datasets land outside it -- the translation gap stays a tie and SMC beats PID on the angular worst case (Cohen's d up to +0.83). The PTZ plant's 100 degrees/s default sits on the cross-over, which explains a small-margin PID result previously reported in the angular setting. A complementary ablation shows raising the SMC reaching gain k costs mean tracking outside a narrow window.
A 3D Gaussian Splatting (3DGS) reconstruction inherits arbitrary units from structure-from-motion, so it cannot be measured or used to place AR content until someone supplies a scale. Both standard recipes are fragile on real captures: on five indoor scenes with tape-measured ceilings, a quantile floor-to-ceiling heuristic errs by 27% and metric-depth alignment by 35-71% across three networks. We present METRIC-GS, a selective scale estimator for real indoor photographs: it reads the scale off the reconstruction's own semantic content and, just as importantly, refuses when the evidence is thin. A frozen 2D teacher labels Gaussians by visibility-aware voting; instance-level size priors for architectural and furniture classes (doors, tables, the floor-to-ceiling gap) are fused by a robust maximum-likelihood step; and an inlier-mass test turns thin evidence into an explicit abstention that doubles as a router to a heavyweight metric model. On 47 ScanNet++ scenes with laser ground truth, with a configuration frozen before public evaluation, the estimator alone cuts the heuristic's mean error from 21.1% to 16.6% (9.3% on the 28 scenes it acts on) at ~11 s per scene. MASt3R remains the more accurate standalone estimator (9.9%), but routing only our refusals to it matches running it everywhere with 60% fewer invocations; this selective hybrid is the configuration we advocate. Visibility-aware voting raises the scenes the estimator can act on from 16 to 28 at unchanged deployed accuracy. Applied unchanged to 13 phone and drone captures of our own, on which MASt3R fails its own registration gate 12 times, the system improves all five captures it acts on and degrades none. We also report that a current feed-forward metric model scores 2.3% on a benchmark inside its training mixture and 28-35% on captures it has never seen: a published metric-scale number is a best case, not a deployment estimate.