LunaSeis-1: When Balanced Apollo Seismic Detection Fails Under Cross-Station Continuous Scanning
A reproducible lightweight ML research prototype for Apollo lunar-seismic waveforms.
On This Page
01 — Overview
Overview
LunaSeis-1 is Archeum Studios' first official research paper, authored by Advaith Praveen (APRK). It studies whether a compact detector trained on Apollo Passive Seismic Experiment data can generalize across stations and remain useful during long continuous scans.
The present answer is honest and deliberately bounded: not operational yet. The selected 2,761-parameter depthwise CNN is a functioning open inference prototype, but frozen continuous evaluations exposed missed events and excessive false alarms.
02 — Evidence boundary
Evidence boundary
This page is deliberately careful with claims. LunaSeis-1 is a reproducible research prototype, not an operational lunar seismic detector.
The frozen continuous evaluations are part of the point: they expose missed events, false alarms, and station-specific distribution shifts that should be solved before stronger claims are made.
03 — Links
Links
The repository includes the paper, code, outputs, and release artifacts. The PDF is the canonical paper link for reading the current version.
04 — Findings
What the paper preserves.
- 1,314 reconciled physical-event candidates from official PDS and corrected shallow-moonquake sources.
- ATT-aware timing, explicit gap preservation, and checksum-backed NASA provenance.
- Event-, station-, chronology-, and repeating-family-aware leakage controls.
- More than 5,243 station-hours across two frozen continuous tests.
- Depthwise CNN development false-trigger rate of 0.211/hour at 91.15% development event recall.
- Frozen Grade-C impact challenge: 12/63 recall with 0.868 false triggers/hour.
05 — Prototype
Selected prototype details.
Scores above the included development threshold are candidate triggers, not verified moonquakes.