Update model card framing for slab benchmark artifact

README.md

@@ -1,55 +1,50 @@
 ---
 library_name: pytorch
 tags:
-  - physics-informed-ml
-  - surrogate-model
+  - research-artifact
+  - computational-photonics
   - helmholtz
-  - photonics
+  - surrogate-model
 ---
 
-# Neon Slab Models
-
-This repository contains the current benchmark-facing slab-family checkpoints from the Neon research codebase.
+# Neon: Scalar Slab Benchmark Research Artifact
 
-## Included assets
+This repository contains benchmark-specific research artifacts from the Neon codebase. It is not a general-purpose photonics model and it should not be treated as a deployable surrogate.
 
-- `model_c_benchmark_best.pt`: best single Model C checkpoint
-- `training_summary.json`: single-model training summary
-- `slab_hybrid_train_enhanced_benchmark.json`: training config used for the benchmark-facing model
-- `ensemble/member_00.pt` to `ensemble/member_04.pt`: 5-member Model C ensemble checkpoints
-- `ensemble/ensemble_manifest.json`: relative manifest for the ensemble files
-- `ensemble/ensemble_evaluation_summary.json`: benchmark-facing ensemble evaluation summary
+## What is included
 
-## Benchmark scope
+The main checkpoint is Model C, a reduced hybrid model trained on Neon’s current 2D scalar normal-incidence rectangular dielectric slab benchmark. A 5-member ensemble and evaluation summaries are also included.
 
-These checkpoints are for the current Neon scalar normal-incidence dielectric slab benchmark only.
-The scalar targets are:
+Inputs:
+- slab thickness
+- slab relative permittivity real part
+- wavelength
 
+Outputs:
 - `benchmark_normalized_transmission`
 - `benchmark_normalized_reflection`
 - `normalized_peak_intensity`
+- a cropped 145-point complex centerline field
 
-## Important limitations
-
-- This is a reduced scalar Helmholtz benchmark model, not a full-vector Maxwell model.
-- The hybrid model predicts scalar targets and a cropped centerline field, not a full 2D field.
-- Ensemble spread is a useful heuristic diagnostic, not a calibrated uncertainty guarantee.
-- Direct Neon reevaluation remains mandatory for reported inverse-design candidates.
-
-## Current reported single-model result
-
-From `training_summary.json`:
+Training data:
+- 18 slab designs x 6 wavelengths = 108 total samples
+- 72 train, 18 validation, 18 test
+- one fixed benchmark geometry, source, grid, monitor layout, and absorber configuration from the Neon slab base case
 
-- test mean MAE: `0.1145`
+## Intended use
 
-## Current reported ensemble result
+Use these files only to reproduce or inspect the slab-benchmark results reported by Neon.
 
-From `ensemble/ensemble_evaluation_summary.json`:
+## Limitations
 
-- test mean MAE: `0.1155`
-- OOD mean MAE: `0.1293`
+- This model was trained only on the current scalar normal-incidence rectangular dielectric slab benchmark.
+- It cannot be used as a general model for metasurfaces, waveguides, photonic crystals, arbitrary multilayer structures, arbitrary source conditions, or full-vector Maxwell problems.
+- It does not predict a full 2D field; it predicts slab-response scalars and a cropped centerline field only.
+- Any uncertainty values here are ensemble disagreement, not calibrated uncertainty.
+- Direct Neon reevaluation remains mandatory for any reported inverse-design candidate.
+- The trained weights are a byproduct of an investigation into simulation-driven ML workflows on a controlled benchmark, not a deployable tool.
+- The research contribution of Neon is the methodology, validation discipline, and benchmark findings, not the model weights themselves.
 
-## Provenance
+## Research contribution
 
-Generated from a local checkout of the Neon research repository on 2026-03-22.
-This bundle corresponds to the benchmark-facing scalar dielectric-slab milestone described in the repository README and companion docs.
+Neon’s contribution is the benchmark methodology and the findings about target definition, reduced physics-informed losses, uncertainty diagnostics, active learning, and direct-solver-verified screening on a controlled slab problem.