Explicit node coordinates (x/y positions) not fully respected in Sankey diagram

[DanielYang59]

We (@janosh) are trying to generate a Sankey diagram but for some reason the explicit coordinates are not fully respected. The "right_a" node has a coordinate of (0.9, 0.05) (top right), but for some reason shows at bottom left.

plotly 6.3.1, macos 26.0.1

import plotly.graph_objects as go


labels = ["left_a", "left_b", "left_c", "left_d", "left_e", "left_f", "left_g", "left_h", "left_i", "left_j", "left_k", "left_l", "left_m",
    "left_n", "left_o", "left_p", "left_q", "left_r",
    "right_a", "right_b", "right_c", "right_d", "right_e", "right_f", "right_g", "right_h", "right_i", "right_j", "right_k", "right_l"]

node_pos = {
    "left_a": (0.1, 0.1), "left_b": (0.1, 0.14), "left_c": (0.1, 0.18), "left_d": (0.1, 0.23),
    "left_e": (0.1, 0.27), "left_f": (0.1, 0.31), "left_g": (0.1, 0.35),
    "left_h": (0.1, 0.40), "left_i": (0.1, 0.44), "left_j": (0.1, 0.48),
    "left_k": (0.1, 0.52), "left_l": (0.1, 0.56), "left_m": (0.1, 0.61),
    "left_n": (0.1, 0.65), "left_o": (0.1, 0.69), "left_p": (0.1, 0.73),
    "left_q": (0.1, 0.77), "left_r": (0.1, 0.82),
    "right_a": (0.9, 0.05),
    "right_b": (0.9, 0.15), "right_c": (0.9, 0.25), "right_d": (0.9, 0.35), "right_e": (0.9, 0.45), "right_f": (0.9, 0.55), "right_g": (0.9, 0.65),
    "right_h": (0.9, 0.75), "right_i": (0.9, 0.85), "right_j": (0.9, 0.93), "right_k": (0.9, 0.97), "right_l": (0.9, 0.99)}

sources = [0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 3, 3, 3, 3, 4, 4, 4, 5, 5, 5, 5, 5, 6, 6, 6, 6, 7, 7, 7, 7, 7, 8, 9, 9, 9, 9, 9, 9, 9, 9, 10,
    10, 10, 11, 12, 13, 13, 13, 13, 13, 13, 14, 14, 14, 14, 14, 14, 14, 14, 14, 15, 15, 15, 15, 15, 15, 16, 16, 16, 18, 18, 18, 19, 19, 19, 19, ]

targets = [23, 21, 20, 26, 30, 20, 24, 22, 21, 20, 20, 25, 29, 22, 22, 29, 20, 20, 22, 25, 24, 29, 20, 22, 29, 21, 21, 20, 24, 30, 22,
    20, 20, 21, 25, 22, 30, 28, 24, 29, 20, 21, 24, 20, 20, 22, 20, 23, 21, 24, 31, 25, 21, 22, 20, 23, 30, 27, 26, 24, 21, 20, 22,
    23, 30, 25, 20, 21, 22, 20, 21, 25, 20, 22, 21, 24, ]
values = [1 for _ in targets]
assert len(sources) == len(targets) == len(values)

x_positions: list[float] = [node_pos[lbl][0] for lbl in labels]
y_positions: list[float] = [node_pos[lbl][1] for lbl in labels]

fig = go.Figure(
    go.Sankey(
        arrangement="fixed",
        node=dict(
            label=labels,
            x=x_positions,
            y=y_positions,
        ),
        link=dict(
            source=sources,
            target=targets,
            value=values,
        ),
    )
)

# fig.show()

fig.write_image("debug.svg", format="svg")

[FredericoMendes10]

I'm working on it!

[FredericoMendes10]

After digging into this issue, I found two related root causes that explain why x/y positions are misapplied in Sankey diagrams with arrangement='fixed':

1. Isolated nodes: Plotly.js excludes nodes with no valid links from its internal b.nodes array, then applies i.node.x[p] to b.nodes[p] assuming a 1:1 index correspondence. When isolated nodes appear before connected nodes in the original array, this correspondence breaks and positions are applied to the wrong nodes.

2. Phantom nodes: Out-of-bounds source/target indices cause Plotly.js to compute _ = max(sources + targets) + 1, inflating b.nodes with unlabelled entries that further misalign the index mapping.

I submitted PR plotly/plotly.py#5548 which works around both issues on the plotly.py side by reordering per-node arrays before rendering (so isolated nodes appear last), but as noted in the PR review, this only fixes static image export via Kaleido. The underlying fix should live in plotly.js itself.

I'd be happy to explore a plotly.js fix at some point, though I can't take it on right now.