Research tool Model P1 Open source

Observed VNIR.
Candidate SWIR.

Abstract BridgeCheck retrieves a physics-grounded SWIR candidate from an absolute-reflectance VNIR leaf spectrum. It is designed for paired-measurement study planning—not sensor replacement or diagnosis.

Try an example Read the evidence

Input: 400–1000 nm measured VNIR
Output: 1052–2400 nm model-derived candidate
Compute: Verified locally in your browser

Figure 01 Spectral bridge schema

The 1000–1052 nm gap and the two data origins remain explicit.

Run the frozen bridge

Start with a documented example, or provide your own measured VNIR point spectrum.

Verifying frozen artifact…

Example spectra

One click loads and runs locally. Labels describe spectral geometry—not biology.

5 deterministic examples

Generated-bank examples are exact by construction; zero context error is not evidence of real-world accuracy.

Input / VNIR

VNIR point spectrum

Provided input

Coverage: 400–1000 nm
Minimum: 100 bands
Gap: ≤ 10 nm
Unit: decimal fraction

Built-in test data CABO measured VNIR sample 151 bands · anonymized training-group median · not validation

Paste CSV spectrum wavelength_nm,reflectance

Strictly increasing wavelengths. Fractions, not percentages. Values are never clipped.

Use a local file Read on this device; never transmitted.

No spectrum loaded

Ctrl + Enter

Output / inferred

Candidate spectrum

No candidate generated

Select an example or provide a contract-valid measured VNIR spectrum.

Nearest state: —
Context RMSE: —
Reference support: —

Reference support is descriptive distance—not confidence or validation.

Spectral view

Neighbor range is descriptive—not calibrated uncertainty.

Input VNIR Model-derived 5-neighbor range

Swipe or use arrow keys to inspect the full spectrum

Origin	Bands	Provided to model
Input VNIR	— input bands	`true`
Candidate SWIR	— derived bands	`false`

Interpretation boundary

The useful result and the failed gate

Reconstruction improved across three sources. Camera validity and downstream utility still failed.

Study	Evaluation unit	Reconstruction result	Decision
EXP-0123CABO	468 biological groups	61.98% lower MAE	Development mechanism pass
EXP-0124NASA FFT	9 external sites	55.28% lower MAE	External point-spectrum pass
EXP-0125HySpex cubes	46 plants · 230 cubes	49.83% lower aggregate MAE	Camera + utility gate failed

Relative MAE reductions are reconstruction results, not gains in diagnosis, biology, or downstream task performance.

From candidate to evidence

01
Generate
Explore a separately typed candidate curve.
02
Measure
Acquire same-unit VNIR and real SWIR.
03
Audit
Compare against strong baselines by group.
04
Test value
Evaluate measured SWIR on untouched endpoints.

Separate companion / Linux CLI

Challenge the declared spectral route before external validation.

BandTrace challenges a declared sensor-to-model route, measures required-band output dependence on adequate probes, and compares routed response functions with declared training support. It keeps executable, spectral and biological evidence separate.

Open BandTrace release Browse immutable source

BandTrace does not run on this page. BridgeCheck sends it no inputs or generated bands, and it does not change BridgeCheck's evidence, output types or claim ceiling.

X3 Dependence observed

On frozen probes for the selected numeric output—not global causal use.

S3 Declared SRF support

Routed response functions fit declared support—not camera calibration.

T0 Biology not evaluated

No transport, accuracy, diagnosis, certificate or deployment approval.

X3 + S3 + T0 is the instrument-controlled reference result, not a promise that a submitted model or sensor will pass.