Flow reaches the vessel openings.

What changed

• The four pulmonary veins, IVC and both exported upper branches feeding the SVC have separate inflow boundaries.
• The full aortic arch leads to its three upper branches and descending outlet. Both pulmonary arteries have terminal outflow.
• Tracers can leave only through outward-flowing arterial terminal faces. Aortic tracers return via the venae cavae; pulmonary-artery tracers return via the pulmonary veins. An unrendered staggered wait spreads their return across the cycle.
• Wall correction uses the animated tissue as well as reference clearance. Slow flow during a closed-valve interval no longer triggers a stalled-particle reset.
• Colour Doppler uses the whole scan sector. The rectangular box and its region selector are removed.

The route follows the NHLBI description of circulation. The reference-grid field, motion correction and branch flow shares remain model assumptions.

Five angles, two phases, before and after

These are actual captures from the shipped WebGPU renderer using native Dawn, taken at the same cardiac phases. Click an image for its full resolution. They are offline captures, not browser screenshots or phone performance measurements.

Front angle · ejection and filling

Back angle · ejection and filling

Right angle · ejection and filling

Left angle · ejection and filling

Above angle · ejection and filling

All 13 external openings checked

The final transport ran for 15 complete beats on each geometry profile: 840 updates with the desktop mask and 420 with the mobile mask. Both used 5,000 tracer slots. All four valves recorded forward transfers, with no forbidden label transitions. External tracer returns entered the downstream venous circuit.

Venous events count accepted seeds, including tracers awaiting return; arterial events count actual terminal exits. Counts are visual tracer statistics, not mL of blood. Branch shares come from projected opening area, not measured vascular resistance. The two SVC entries are the separate upper cut ends in this export; they merge into the SVC.

The quantized field was checked on every designated opening and every remaining wall face. Prescribed opening fluxes agree to better than 0.01%; non-port wall-normal flux is zero. The four shared valves remain conservative, filling and ejection windows remain separate, and left/right stroke volume remains 54.52 mL per beat.

In the first three beats, the sampled terminal corridors had no displacements below 1 mm/s during their active flow windows on either profile. Small local changes of direction still occur near curved walls; this is not a guarantee that every tracer follows a physiological streamline.

Desktop required 18 wall-stall reseeds over the full 15 beats; mobile required 46. These remaining corrections cluster near the moving valve/chamber junctions. The original persistent blockage at the aortic arch and missing venous supply are resolved, but local moving-wall transport is still approximate.

Doppler without the box

Native rendering checks confirm matching loop endpoints and that old saved region values also render the whole sector. The healthy and geometry ECG modes still select their matching activation patterns.

What this does not establish

This remains a reduced kinematic circulation, not patient-specific CFD. The aorta and pulmonary arteries are zero-storage conduits with prescribed systolic outflow; arterial elastic recoil and diastolic runoff are not simulated locally. Venous waveforms, branch resistance, evolving vortices, actual external transit time and absolute Doppler velocity remain unvalidated. The surrounding vessels and coronary network are still outside the tracer domain. No FBX geometry was changed for this correction.

Mobile was checked with its 192³ tissue mask at a simulated 30-update-per-second cadence. This validates transport behavior at that timestep, not a physical phone's achievable frame rate.