Note: This is simply a test of the Wolfram MCP integration with AI, nothing more than this.
With the Wolfram MCP server, we can do something more complex other than the coding.
There is an interesting post “Don’t Study Large Earthquakes with Mathematica’s EarthquakeData” by from Charles J. Ammon’s at 2017. And nine years later on Wolfram Language 15.0.1, is it still the case. Let’s revisit this post with AI.
The blog post is tested in Claude, ChatGPI, and verified by Meta’s muse-spark model. I won’t go into any details, and the reports are listed in the table.
The conclusion: the main claim from Ammon’s post in 2017 still holds on WL 15.0.1 today: if you need to work with earthquake magnitude data in Mathematica, import it from USGS, ISC, or GCMT directly rather than relying on EarthquakeData.
| File | What it is |
|---|---|
blogpost-followup-claude-sonnet-5.md | Follow-up A (Claude): tight literal re-run of Ammon’s 9 claims on WL 15.0.1, with Wolfram vs reference bar charts |
blogpost-followup-GPT5.6-sol.md | Follow-up B (ChatGPT): same core + seismological “why” — magnitude-type mixing, 1964 Alaska duplicate (8.5 + 8.4), 178 vs 100 count gap since 1900, live 2025 Kamchatka failure (8.0 Mi vs 8.8 Mww) |
blogpost-independent-verification-muse-spark-1.2.md | Independent live verification (muse-spark-1.2, Sept 4, 2026): local wolframscript + USGS FDSN checks; head-to-head of A vs B — both correct on essentials, B is a strict superset |
blogpost-independent-verification-muse-spark-1.3.md | Re-run of the verification under muse-spark-1.3: fresh wolframscript + USGS curl log; reproduces the 1.2 pass exactly |
blogpost.webarchive | Archived copy of Ammon’s original 2017 post |