TaskQ does not “route” like a proxy. It solves placement as a constrained decision across policy, jurisdiction, custody, attestation, price, health, capacity, and proof.
Load balancing asks which server is least busy. TaskQ asks which worker is allowed, trusted, healthy, priced within ceiling, jurisdiction-compatible, custody-compatible, and cheapest after every hard constraint is enforced.
If the customer requires Canada-only execution, the US does not become acceptable because it is cheaper, faster, or available.
The engine finds the lowest-cost worker inside the compliant set. It does not compare compliant and non-compliant workers as if they are equal.
Policy hash, registry snapshot, eligible-set commitment, chosen worker attributes, lease, and price basis are written into proof.
Every filter removes workers that would make the request cheaper, faster, or easier — if they violate the policy, they are not candidates. Empty the pool and TaskQ refuses with proof.
customer.default.must_run_in = ["CA"] request.must_run_in = ["CA-AB"] result = allowed / tighter customer.default.must_not_run_in = ["US"] request.must_not_run_in = [] result = 422 POLICY_LOOSENING_FORBIDDEN
TaskQ commits to the policy, registry snapshot, eligible set, chosen worker attributes, price basis, lease ID, and selection reason. That is the difference between a router and an execution control plane.
| Field | Purpose | Why it matters |
|---|---|---|
| policy_hash | sha256:... | Proves the constraints that were applied. |
| registry_snapshot | merkle:... | Shows the worker universe at decision time. |
| eligible_set | commitment | Supports replay without leaking the entire network. |
| selection_reason | lowest_price_within_hard_constraints | Explains why the worker won. |
| lease_id | lease_... | Prevents solve-to-dispatch race conditions. |
Hard constraints, cheapest-compliant selection, deterministic tie-breakers, and a proof object for every decision.