robyn_allocator() allocating zero spend to effective channels

[sendizou]

I’m having trouble with robyn_allocator() and would appreciate advice on what to check.

Robyn MMM itself looks OK: saturation/response curves are generated and make sense.

The issue is specifically with allocator output: some channels that appear effective in MMM (non-zero contribution/coef) end up with 0 optimized spend, while other channels that show near-zero response can receive budget.

This behavior persists even after trying different allocator settings (e.g., constr_mode variants and other knobs).

It seems that, under constraints, certain channels become effectively “inactive” in the allocator optimization space (e.g., bounded points/marginal response collapsing to 0/NA), causing unexpected skipping/zero allocation.

Questions

Under what conditions can allocator treat a channel as inactive (skipped / bounded points collapse) even when MMM indicates non-zero effectiveness?

What are the recommended diagnostics to pinpoint why bounded points or marginal response become invalid (which objects/fields should we inspect)?

Any known pitfalls with weekly data / low-spend periods / spend↔exposure pairing that can cause allocator to behave like this?