Reader guide
The interface turns a saved N-body validation artifact and a saved cardiac signal-processing artifact into an inspectable public surface. The terms below are operational labels for the evidence packet.
Why Modal should care
Modal is named here as a candidate remote batch-compute layer for repeatable orbit jobs and public inspection workflows. This page does not state or imply Modal approval.
| Variable | Common field reference | Current MendozaLab rung | Delta Modal can amplify |
|---|---|---|---|
| Scale Catalog breadth | Literature catalogs range from tens to thousands of known periodic three-body families. | 60-row pass gate and 101-row trajectory rung from source-verifiable numeric rows. | Scale curated validation ladders from 101 toward larger literature catalogs with reproducible job logs. |
| Rigor Reference backend | High-accuracy N-body work expects trusted integrators and invariant diagnostics. | Remote REBOUND/IAS15 jobs with energy, angular momentum, center-of-mass, and period-return checks. | Make backend parity visible across many independent jobs, not just local notebooks. |
| Batch Parallel execution | Independent orbit integrations are naturally batch-parallel but often published as static tables. | Modal Function.map runs independent known-orbit jobs; no detached sweep used in this gate. | Move from validation batches to scheduled, auditable sweeps without changing the artifact contract. |
| Artifact Trajectory tensors | Tables and plots are usually separated from reproducible simulation artifacts. | 101 trajectories x 360 frames x 3 bodies emitted as a sidecar and rendered in the browser. | Turn compute output into inspection-ready media for reviewers, collaborators, and grant panels. |
| Audit Negative evidence | Failed orbit returns often disappear into tuning notes. | 101 rung preserves 8 diagnostic failures while still passing the 90% catalog gate. | Modal logs make failure rows auditable instead of hidden, improving trust in the benchmark. |
| Kernel Scaling backend | Python orchestration is convenient; compiled kernels are preferred for sustained throughput. | REBOUND remains the reference; Rust is the next parity-and-scale path, not a replacement claim. | Use Modal to compare Python reference, REBOUND, and Rust kernels under the same registry. |
| Surface Audience translation | Scientific simulations rarely become customer-readable evidence products. | N-body and cardio panels share one artifact-backed interface. | Show Modal as the compute layer behind evidence people can inspect, not just a backend bill. |
Cosmos / N-body
The catalog rung uses source-verifiable initial conditions, remote REBOUND/IAS15 integration, and invariant thresholds. The browser view reads the trajectory sidecar, so the visual is tied to the run artifact.
Cardio proof panel
No clinical endorsement language. This panel only exposes saved validation facts from the Rust/Python parity full-pass artifact.
The two panels are intentionally asymmetric: N-body shows run reproducibility and trajectory hashes; Cardio shows saved classifier readouts. They share artifact discipline, not identical evidence types.
Claim filter: known-orbit validation and artifact-backed cardiac signal-processing readout only. No discovery assertion, no clinical-use assertion, no institutional endorsement, no formal-math overclaim.