Six kitchen-scale experiments systematically falsify classical thermodynamic predictions for gradient-dominated regimesand establish thermal gradients as attractive vector force fields capable of performing sustained mechanical work againstconservative fields.Experiment 1 (Iceberg): Water frozen on a stainless steel tray inclined at θ ≈ 10° receives heat from a gas flame (Tflame≈ 1200°C) at its upper end. Ice (Tc ≤ 0°C) is positioned at the lower end. Within 90 seconds, meltwater reverses fromdownward to upward flow against gravity. The system sustains simultaneous coexistence of ice (0°C, solid), liquid bridge,and boiling water (100°C, Leidenfrost gas phase) across L ≈ 15 cm for > 720 s — a factor of 4.56× beyond the classicalequilibration prediction (τeq = L2/α ≈ 160 s). Individual Leidenfrost droplets undergo orbital trapping and are trackedcontinuously for > 4.67 min.Experiment 2 (Butter): Lipid samples (butter, ρ ≈ 920 kg/m3, Tmelt ≈ 32°C) on a horizontal frying pan execute curvedtrajectories toward the heat source — direct evidence of vector field geometry incompatible with any scalar mechanism.On an inclined tray (θ ≈ 10°, Tsurface,max = 65°C, |∇T| ≈ 267 K/m), butter climbs against gravity for 25+ min. A spatialshearing boundary forms at rthreshold ≈ 10 cm, yielding kT ≈ 6.4×10−3 m·s−2·K−1.Experiment 3 (Boiling Water): Water in a stainless steel pot monitored with digital and infrared thermometers. Bulkmotion correlates with flame state (ON/OFF), not with temperature. Water at 100.0°C with flame OFF exhibits zero visiblemotion, while water at 72.7°C with flame ON shows active convective movement — a ∆T = 27.3°C reversal of theclassical prediction.Experiment 4 (Boiling Milk): Milk monitored across 4 heating/cooling cycles. Milk executes a completestillness→near-overflow→stillness cycle within a 0.3°C range (100.5–100.8°C), with behavior determined entirely byflame state. IR surface temperature reaches 102.6°C when bulk temperature is 75.4°C (∆ = 27.2°C).Experiment 5 (Beans & Lentils — Short): Water with black beans and lentils heated at maximum flame. Temperaturedrops from 17.3°C to 15.8°C in 41 seconds of continuous heating — a 1.5°C decrease with fire ON. Two simultaneousthermometers document an inverted gradient: the bottom of the pot (directly above the flame) reads 16.6°C while theliquid above reads 21.5°C — the bottom is 4.9°C colder than the top. Video E1: youtu.be/1dV24cUDyl4Experiment 6 (Beans & Lentils — Long): Extended 27-minute experiment with the same mixture. Baseline establishedat 22.3°C (stable, fogo OFF). Upon ignition, temperature plateaus at 22.0°C for 46 seconds (0.3°C below baseline) undermaximum flame. After flame OFF at 26.4°C (t = 2:34), temperature continues rising without any heat source for 17+minutes, reaching 38.1°C. During this entire period, the surface (IR) is consistently hotter than the bottom (probe), withinitial ∆ = 9.5°C converging to 0°C at t = 20:19. Second flame cycle at t = 26:04 produces IR 57.4°C vs probe 37.8°C (∆ =19.6°C) in under 48 seconds. Video E2: youtu.be/4P31ecyaeesThe combined evidence establishes Fnthermal = −kTm∇T as a fundamental force law, implying that gravity contains athermal component — consistent with Gravity = Magnetism + Heat.