Boat node — looking down
Drone node — looking up
Dead zone wedge
Fish both nodes caught
Fish only the drone caught
Fish nothing caught
Modelled target — verified
Modelled target — one node only
Bed structure — wreck, crib, rock
Marker — click to zoom in

300 ft

Every detected fish gets a square — green means both nodes confirmed it. Click one to fly in. Get back out with Release, Wide, a double-click, or just by zooming out. Wheel, pinch, + / − all zoom.

Two sounders, one water column

Each ping is a 1D line of echoes. The volume you see is built from motion: the boat drags the pair along track while the drone hangs below. The second sounder is not what makes it 3D — towing it is.

What the drone buys you is the bottom dead zone. It only pays when it flies below that band, so Fly at is the control that matters. Try Narrow beam: the blind band shrinks to inches and the drone earns nothing.

Echoes are then peak-picked into targets. Anything both nodes report at the same place in the water is verified, and gets a fish drawn at its estimated length. Pick one from the list to fly the camera in.

The fish are real paintings — public-domain plates from the US Fish and Wildlife Service, NOAA and a 19th-century state fisheries report. Which one gets drawn is picked from estimated length alone, because no echo can tell you the species. Solid fish are verified; the faint ones were seen by a single node.

Rig

Fly at 4 ft Boat 6 mph Boat beam 40° Drone beam 34° Fish 90

Season

Month Aug

Species are ranked from the water temperature at the target's depth rather than from depth alone, because depth is only where the right temperature happens to be that month. Drag through the year and the same depth fills with different fish — in April a lake trout is shallow, in August it is under the thermocline.

Show

Targets

Model size ×4

Before you keep anything

The seasons and limits on the target card are orientation figures, not the regulations. Michigan rewrites the guide every year, the Great Lakes and inland waters differ, and individual lakes and management units carry their own rules on top.

On top of that the species itself is a ranked guess from length, depth and temperature. Acting on it could mean an undersized or out-of-season fish — and lake sturgeon are protected on almost all Michigan water. Check the current guide: Michigan DNR fishing.

View

Eye split 3.0%

Stereo 3D renders side-by-side, which is the format the Beast accepts over DisplayPort alt mode. Eye split is scaled to the orbit distance, not to a human skull — at 100 ft a real 2.5″ separation would show no depth at all.

Geometry right now

Detection score

Counted as fish pass astern. Fish that swam wide of both beams are left out, so the gap you see is the blind band, not poor coverage.

Length accuracy — sim check

Only possible because this is a simulation. A fish out at the edge of the cone reflects a weaker echo and reads short; being seen from two sides means at least one look was close to on-axis.

About the fish models

Every fish you see is built from a detection, never from the simulated truth. Depth is measured, and length comes from echo amplitude: backscatter scales with the square of length, so a range-compensated amplitude is roughly linear in it. Pose and which side of the boat are invented, and the species is a ranked guess from length and depth, never a measurement. Solid fish were confirmed by both nodes, ghosted ones by one.

The roster is 39 Great Lakes species, painted from public-domain plates (see fish/CREDITS.md). Ranking uses estimated length, depth and the water temperature at that depth for the month. Open a target to see the runner-up species and how much of the confidence they take — on a big fish the spread is wide, because amplitude saturates near 36″ and everything above that reads the same.

Only a few targets verify, and that is the honest result. The drone's narrow beam covers a slice of the boat's swath, so most fish are only ever seen once. Widen Drone beam and more will pair — then watch the merge rate climb and the length estimates fall apart.

Why the line, and how it breaks

Single-node targets sit in a straight line down the track, and that is not a drawing shortcut — it is the honest answer. One ping measures range, never bearing, so all that beam can say is "somewhere on this cone", and the axis is the conventional place to plot it. That line is the limitation.

Motion breaks it. Tow a node past a fish and its range bottoms out at closest approach; that minimum is the perpendicular distance to the track line, and the ping it happens on fixes the along-track position. Each node therefore knows the fish lies on a circle in the cross-section. Two nodes at different depths give two circles, and two circles cross at a point.

Pick a verified target and you can see both circles and their crossing. Those fish get placed out in the water where they really are. What is left is a mirror: both nodes are towed down the centreline, so port and starboard are indistinguishable, and the faint second fish is the solution the evidence cannot rule out.

Notice how few targets earn it. The body flies a few feet off the bottom and most fish hug the bottom too, so it is looking up at them from ten or twenty feet away — and a cone that close spans only a few feet of water. Watch Both nodes — placeable in the scoreboard: the drone contributes a narrow ribbon down the track, not a swath. Everything outside it stays on the axis because one beam genuinely cannot do better.

Which sets up the real trade. Baseline length buys accuracy; overlap buys coverage, and this rig cannot have both. The drone's baseline is the whole water column, so a foot of range error barely moves the answer — but it only overlaps a ribbon. Two transducers a few feet apart on the transom overlap almost perfectly, but the baseline is so short that cross-track has to come from phase across the face rather than from differencing two envelope ranges. That is precisely what a split-beam transducer is, and why it costs what it does.

What is honest here

Beam spreading, the dead-zone formula, cone footprint, ringdown blanking and downrigger blowback are modelled, and the drone is held to a real altitude rather than teleported. Echo amplitude, fish target strength and bottom hardness are plausible fiction — good enough to judge geometry, not to calibrate against.