Eve Count Quantum Systems · 02 · The classical baseline first. Select the architectural plate to inspect full size.

02 · The classical baseline first

If you haven't benchmarked a well-tuned classical heuristic, you cannot measure quantum value.

A recurring antipattern in vendor-led pilots is comparing a tuned quantum-inspired or hybrid solver against a strawman brute-force script. At Eve Count, our first rule of governance is Evidence Before Advantage: every candidate formulation must be benchmarked against a deterministic classical baseline on the exact same constraint set.

How greedy heuristics get trapped by coupling

Consider four discrete allocation requests competing for max_capacity = 8 units of berth/crane capacity:

• T1_FEEDER_EXPRESS yields value 16.0 at cost 5 (highest value-per-unit ratio 3.20), but conflicts with both T2_ULCV_NORTH and T3_ULCV_SOUTH.
• T2_ULCV_NORTH (value=12.0, cost=4) and T3_ULCV_SOUTH (value=11.0, cost=4) do not conflict with each other and together yield 23.0 within the exact capacity of 8.

Let's run the deterministic classical greedy baseline:

from examples.combinatorial_triage import AllocationTask, solve_classical_greedy

tasks = [
    AllocationTask("T1_FEEDER_EXPRESS", value=16.0, resource_cost=5, conflicts_with=("T2_ULCV_NORTH", "T3_ULCV_SOUTH")),
    AllocationTask("T2_ULCV_NORTH", value=12.0, resource_cost=4, conflicts_with=("T1_FEEDER_EXPRESS",)),
    AllocationTask("T3_ULCV_SOUTH", value=11.0, resource_cost=4, conflicts_with=("T1_FEEDER_EXPRESS",)),
    AllocationTask("T4_BUNKER_BARGE", value=4.0, resource_cost=3, conflicts_with=()),
]

baseline = solve_classical_greedy(tasks, max_capacity=8)
assert baseline.selected_tasks == ("T1_FEEDER_EXPRESS", "T4_BUNKER_BARGE")
assert baseline.objective_value == 20.0
print(baseline)

Because solve_classical_greedy ranks tasks locally by efficiency (16.0 / 5 = 3.2), it locks in T1_FEEDER_EXPRESS first, blocking both T2 and T3 and leaving the schedule stuck at 20.0 instead of the global optimum 23.0 (a 1,500 basis-point optimality gap).

Try with your AI coding assistant

Ask your assistant: "Write a short unit test showing when solve_classical_greedy achieves the exact global optimum (when conflicts_with is empty) versus when pairwise conflicts create a local-minimum trap."

Checkpoint

You should have a deterministic classical baseline (solve_classical_greedy, MIP, or simulated annealing) that runs in milliseconds, produces a feasible schedule, and establishes the floor objective value that any advanced solver must beat.