Synthetic evidence-planning benchmark

Python 3.14.6; SciPy 1.17.1; macOS-26.2-arm64-arm-64bit-Mach-O.
Three fresh-process samples per row. The table reports median elapsed time.
Time excludes module imports and prior incremental cache construction.
Python allocation peaks exclude native solver allocations.
Process RSS includes imports, timing, and memory calls. It is an absolute process high-water mark.
Cached prefixes remain resident in incremental process RSS.
Every incumbent passes an independent exact integer feasibility and cost check.
Numerical solver status and dual bounds remain separate from exact reference results.

case | method | status | cost | exact gap | median ms | Python peak bytes | process RSS bytes
recognition_complement | exact_dp | optimal | 7 | 0 | 0.020 | 824 | 27508736
recognition_complement | prefix_full | optimal | 7 | 0 | 0.021 | 1536 | 27623424
recognition_complement | greedy | feasible_heuristic | 7 | 0 | 0.039 | 592 | 27557888
recognition_complement | milp | feasible_solver | 7 | 0 | 7.049 | 9817 | 81461248
greedy_gap | exact_dp | optimal | 5 | 0 | 0.020 | 888 | 27344896
greedy_gap | prefix_full | optimal | 5 | 0 | 0.018 | 1536 | 27312128
greedy_gap | greedy | feasible_heuristic | 6 | 1 | 0.036 | 592 | 27590656
greedy_gap | milp | feasible_solver | 5 | 0 | 6.965 | 9841 | 81674240
uncovered | exact_dp | infeasible | None | None | 0.011 | 800 | 27688960
uncovered | prefix_full | infeasible | None | None | 0.011 | 1088 | 27574272
uncovered | greedy | infeasible | None | None | 0.014 | 448 | 27607040
uncovered | milp | unknown | None | None | 0.968 | 9721 | 80969728
existing_and_free | exact_dp | optimal | 2 | 0 | 0.021 | 888 | 27656192
existing_and_free | prefix_full | optimal | 2 | 0 | 0.021 | 1312 | 27656192
existing_and_free | greedy | feasible_heuristic | 2 | 0 | 0.031 | 544 | 27705344
existing_and_free | milp | feasible_solver | 2 | 0 | 1.186 | 9769 | 80871424
seed7_d6_m12 | exact_dp | optimal | 22 | 0 | 0.090 | 2072 | 27688960
seed7_d6_m12 | prefix_full | optimal | 22 | 0 | 0.040 | 9000 | 27672576
seed7_d6_m12 | greedy | feasible_heuristic | 22 | 0 | 0.051 | 912 | 27574272
seed7_d6_m12 | milp | feasible_solver | 22 | 0 | 1.238 | 10665 | 80642048
seed19_d9_m18 | exact_dp | optimal | 15 | 0 | 0.830 | 9240 | 27574272
seed19_d9_m18 | prefix_full | optimal | 15 | 0 | 0.245 | 91632 | 27377664
seed19_d9_m18 | greedy | feasible_heuristic | 16 | 1 | 0.069 | 1128 | 27525120
seed19_d9_m18 | milp | feasible_solver | 15 | 0 | 5.159 | 12141 | 81821696
seed31_d12_m24 | exact_dp | optimal | 33 | 0 | 9.845 | 249528 | 28033024
seed31_d12_m24 | prefix_full | optimal | 33 | 0 | 3.241 | 818608 | 29130752
seed31_d12_m24 | greedy | feasible_heuristic | 35 | 2 | 0.081 | 2936 | 27443200
seed31_d12_m24 | milp | feasible_solver | 33 | 0 | 8.929 | 15667 | 81969152
joint_transport | exact_search | optimal | 3 | 0 | 0.185 | 46056 | 27688960
atomic_transport_only | exact_search | infeasible | None | None | 0.111 | 33856 | 27672576
conflicting_support | exact_search | optimal | 5 | 0 | 0.187 | 40992 | 27623424
duplicate_origins | exact_search | optimal | 5 | 0 | 0.455 | 115208 | 27688960
duplicate_only | exact_search | infeasible | None | None | 0.177 | 47272 | 27639808
local_refuse | exact_search | infeasible | None | None | 0.121 | 1840 | 27574272
local_await | exact_search | infeasible | None | None | 0.118 | 1840 | 27557888
local_cap | exact_search | infeasible | None | None | 0.179 | 41544 | 27377664
induced_duties | exact_search | optimal | 3 | 0 | 0.182 | 40992 | 27820032
revoked_origin | exact_search | infeasible | None | None | 0.182 | 47760 | 27590656
joint_transport | bounded_search | unknown | None | None | 0.010 | 816 | 27295744
joint_transport | bounded_search | feasible_incomplete | 3 | 0 | 0.075 | 14920 | 27508736
update_revocation | prefix_full | optimal | 15 | 0 | 0.262 | 112176 | 27361280
update_revocation | prefix_incremental | optimal | 15 | 0 | 0.031 | 21544 | 27656192
update_cheaper_offer | prefix_full | optimal | 2 | 0 | 0.332 | 130744 | 27688960
update_cheaper_offer | prefix_incremental | optimal | 2 | 0 | 0.073 | 21600 | 27623424

Incremental comparisons:
update_revocation: full 2210 transitions; incremental 512 transitions; reused 18 offer layers; recomputed 1 layer; full 0.262 ms; incremental 0.031 ms.
update_cheaper_offer: full 2722 transitions; incremental 512 transitions; reused 19 offer layers; recomputed 1 layer; full 0.332 ms; incremental 0.073 ms.

Manuscript paragraph:
The deterministic synthetic set contains seven coverage instances and ten semantic fixtures. Exact dynamic programming supplies coverage optima. Finite enumeration checks the semantic fixtures. All returned incumbents pass exact integer feasibility and cost checks. The integer-programming incumbents match the dynamic-programming reference costs. The constructed greedy example costs 6 against an optimum of 5. The joint-transport fixture reaches its target at cost 3, while its separate-transport catalogue is infeasible. Revocation and a cheaper new offer preserve agreement between incremental and complete recomputation. A zero-work limit returns Unknown. A 16-pair limit returns Feasible-incomplete after finding a checked incumbent. These measurements concern the declared finite instances and provide no deployment performance claim.
