Delegation overhead and ownership

Status: evidence-backed problem, not a universal framework ranking. Reviewed 2026-09-19 against local tests and published findings; see those pages for the exact dependency pins and measured configurations.

Symptom and mechanism

A multi-agent workflow costs more even when it produces the same answer.

Delegation can advertise schemas repeatedly, copy context, or add model turns. Orchestration shape changes the bill.

Diagnose and reproduce

Separate calls for handoff, worker execution, context forwarding, and task tools.

From the repository root, with the relevant optional adapters installed:

python -m pytest tests/test_delegation_depth.py tests/test_delegation_forwarding.py -q

The test is the executable contract; published evidence explains scoped findings. Missing adapters may skip comparisons; inspect the test summary before generalizing.

Alternatives and tradeoffs

Use a single loop when roles add no verified benefit. Isolated workers reduce context sharing but can duplicate calls.

Verification and open question

Hold task and scripted responses fixed; vary depth and forwarded evidence while measuring the wire.

Open question: does the same outcome hold for your actual provider, workload, configuration, and framework version? Existing scripted findings alone cannot answer that.

Implication for a new harness

Parent task/actor IDs, propagated budgets, and explicit ownership of worker failure.

Return to the problem register.