Three steps

1. Save the script: reproduce.py.
2. Run it: python3 reproduce.py (or python3 reproduce.py --slug a100-80gb-neocloud-od for another series).
3. Compare the recomputed median against the value published on the index page. They should match to four decimal places. If they don't, that's a bug — see below.

What the script does

1. Fetches /api/indices/{slug} from this site (returns the same JSON document as /data/{slug}.json).
2. Reads the contributing_providers_latest array — one row per contributing provider on the latest assessment date.
3. Computes the equal-weighted median of those provider prices.
4. Asserts the recomputed median equals the published value within four decimal places.

Sample output

$ python3 reproduce.py
    Pulse — reproduce one published index value
    ========================================================
    Index:        Pulse H100 SXM Hyperscaler OD
    Methodology:  https://pulsebenchmarks.com/methodology/h100-sxm-hyperscaler-od/v1-0/

    Step 2 — recompute the median locally
      AWS                     $ 6.8800   from 2026-04-24
      Oracle Cloud            $10.0000   from 2026-04-24
      GCP                     $10.9791   from 2026-04-24
      Azure                   $12.2900   from 2026-04-24

      Equal-weighted median of contributing providers: $10.4896

    Step 3 — compare to the published value
      Published:  $10.4896
      Recomputed: $10.4896

      ✓ Match. The published value is reproducible from the published inputs.

Going further

The script verifies the latest published value end-to-end from public inputs. To reproduce historical values you'll need per-provider history per date. Pulse does not redistribute provider snapshots in bulk — fetch the underlying numbers from each provider's public source. Source URLs and SKU filters are listed in the methodology page for each series.

The calculation code (and the rest of the collection pipeline) is open at github.com/pulsebenchmarks; release tags match methodology versions.

Disclosure

Pulse's redistribution policy is graded by source. The published series outputs (median, percentiles, dispersion) are released under CC-BY 4.0. A redistributable subset of sample data is also CC-BY 4.0. Provider-origin raw snapshots are not implicitly covered — provider terms vary, and where redistribution rights are unclear, Pulse links to the original public source so anyone can independently verify. See what is open and free for the full scope.

Verified by

Independent reproductions will be acknowledged here. If your number matches the published value, write to [email protected] — your reproduction will be added to this list. Mismatches are bugs worth reporting.