Shared VSB and output BUS16

DECIMA-8 has two distinct planes that must not be conflated.

Plane	Direction	Purpose
`VSB_INGRESS16[8]`	Conductor → swarm	One common input frame for every ACTIVE tile
`BUS16[8]`	swarm → Conductor	Saturated sum of eligible `BUS_W` tile contributions

VSB: common physical context

VSB does not travel through a tile chain. Within a tick, the same eight-lane vector is available to every ACTIVE tile. Tiles hear it differently because of their weights and accumulated history.

This is the hydraulic model: one shared pressure manifold acts on many local chambers simultaneously.

BUS16: readout, not tile input

During WRITE, eligible tiles contribute row_out:

bus_raw[lane] = sum(row_out[t][lane] for eligible BUS_W tiles)
BUS16[lane] = clamp15(bus_raw[lane])
BUS_CLIP[lane] = bus_raw[lane] > 15

The upper bound of bus_raw depends on the actual number of active writers. The old 256 x 15 value is not a v3 invariant: the runtime supports fabrics up to 4096 tiles.

BUS_R and BUS_W

BUS_R is a historical flag name. In the current runtime it makes a tile an ACTIVE graph seed; the tile does not read BUS16 data.
BUS_W allows a tile to contribute to the common output when its branch has a locked source.

A neighbour edge also carries no Level16 data. It only permits the descendant to apply its own matrix to the common VSB on the next tick.

Why the split matters

One data stream is presented to the whole fabric while topology selects which parts of the personality may respond. Branch depth can therefore express a temporal sequence without copying market, audio, or sensor data between tiles.