Verify · paste your output
Run it. Paste it. See the shape.
Download mgai-meter from
releases,
run it with --json, and paste the output below. The page
decodes it locally — no upload — and shows your per-tier values
next to the canonical receipt pinned to this site. The substrate's
claim is the cascade ratio shape (L2 ≫ L1, L1 ≈ L0-closed within
an order of magnitude). Absolute joules vary by silicon.
chmod +x mgai-meter-<target>
./mgai-meter-<target> --json download · 4 targets · macOS arm64/x86_64 · Linux x86_64/arm64
Your machine vs the canonical
| primitive | V-class | your wall_ns/op | canonical wall_ns/op | ratio (you / canonical) |
|---|
cascade ratios (wall-time)
| ratio | your reading | canonical | shape match |
|---|
What "shape match" means
The substrate's claim is not "we beat your numbers" or "our silicon is faster". It's narrower: every cascade tier sits at its expected order of magnitude on commodity silicon, and the ratios between tiers are consistent. A reading where L1 ≪ L0-closed, or where L2 / L0-closed is nowhere near 10², would falsify the cascade design.
Use this page to confirm your machine produces ratios within an order of magnitude of the canonical. Different silicon, different temperatures, different background load — all of those move the absolute numbers. The ratio shape is what we publish; the ratio shape is what you can verify.
view the canonical receipt JSON src/data/mgai-meter-canonical.json
{
"receipt_kind": "mgai-meter/cost+substrate v1",
"substrate": {
"host_os": "macos",
"host_arch": "aarch64",
"vcpus": 32,
"energy_method": "tdp_estimate_landauer_floor",
"energy_detection_trail": [
"msr_rapl: absent (target_os=macos, /dev/cpu/0/msr missing)",
"powercap: absent",
"nvml: nvidia-smi absent",
"powermetrics: macOS detected (root-only; not invoked by this crate)",
"chosen: tdp_estimate + landauer_floor (live counter integration is a follow-up)"
],
"energy_methods_available": [
"tdp_estimate",
"landauer_floor",
"powermetrics"
],
"assumed_tdp_w": 8,
"kt_ln2_j": 2.87e-21
},
"primitives": [
{
"name": "l0_closed::fk_rollout",
"v_class": "l0_closed",
"source": "tool_result",
"iters": 1000000,
"wall_ns_per_op": 82.75250000000001,
"bit_ops_per_op": 2048,
"e_floor_j_per_op": 5.87776e-18,
"e_tdp_j_per_op": 6.620200000000001e-7,
"mu_apparent": 112631342552.26483
},
{
"name": "l0_retrieved::hashmap_lookup",
"v_class": "l0_retrieved",
"source": "retrieved",
"iters": 1000000,
"wall_ns_per_op": 8.109,
"bit_ops_per_op": 512,
"e_floor_j_per_op": 1.46944e-18,
"e_tdp_j_per_op": 6.4872e-8,
"mu_apparent": 44147430313.58885
},
{
"name": "l0_identifiable::hdc_cleanup",
"v_class": "l0_identifiable",
"source": "perceived",
"iters": 1000,
"wall_ns_per_op": 2793.417,
"bit_ops_per_op": 160000,
"e_floor_j_per_op": 4.591999999999999e-16,
"e_tdp_j_per_op": 0.000022347336,
"mu_apparent": 48665801393.72823
},
{
"name": "l1::cited_lookup",
"v_class": "l1",
"source": "retrieved",
"iters": 1000000,
"wall_ns_per_op": 12.298375,
"bit_ops_per_op": 1024,
"e_floor_j_per_op": 2.93888e-18,
"e_tdp_j_per_op": 9.8387e-8,
"mu_apparent": 33477719403.31011
},
{
"name": "l2::matmul_proxy",
"v_class": "l2",
"source": "ruminated",
"iters": 100,
"wall_ns_per_op": 110991.25,
"bit_ops_per_op": 524288,
"e_floor_j_per_op": 1.50470656e-15,
"e_tdp_j_per_op": 0.0008879300000000001,
"mu_apparent": 590101767084.7399
}
],
"cascade_ratios": {
"l1_over_l0_closed": 0.14861635600132925,
"l0_identifiable_over_l0_closed": 33.756285308600944,
"l2_over_l0_closed": 1341.2434669647441,
"l2_over_l1": 9024.871172004432
},
"e_measured_j": null,
"honest_note": "Time and inter-tier cascade ratios are MEASURED. Energy is bounded below by Landauer (kT·ln 2 × bit-op proxy, exact lower bound) and estimated above by TDP (wall_time × assumed_watts, labelled assumption). Absolute joules require a live silicon counter (RAPL, powercap, NVML, or powermetrics with root); when none is available `e_measured_j` is null. The cascade ratios themselves are real on commodity silicon."
}