What does a successful result establish?¶
Keep four questions separate when evaluating a SpectralBridge run.
| Evidence level | What to inspect | What it establishes | What it does not establish |
|---|---|---|---|
| Execution | Process status, exceptions and logs | The requested code ran or stopped | Required products exist |
| Artifact completion | Resolved configuration, required readable files and provenance | The requested output contract was fulfilled | Numerical or scientific quality |
| Internal QA | Recorded checks, warnings, thresholds, spectra and maps | Evaluated properties met or failed the recorded rules | Agreement with independent observations |
| Independent scientific comparison | Suitable external observations, alignment/aggregation, uncertainty and study design | Agreement for the stated comparison and support | Universal sensor equivalence or transferability |
Passing a level does not establish the next. A warn can require interpretation
without being a crash; a missing report is not a pass. High R² among products
constructed from the same source spectrum does not independently validate a
sensor calibration. Configured thresholds are review aids, not universal
scientific approval criteria.
Software correctness is a separate prerequisite: unit/integration tests and CI exercise numerical contracts, deterministic outputs and restart behavior against their test inputs. See tests and contribution and the release gates. A passing test suite is not independent scientific validation. Workflow evidence comes from readable real-run products, provenance and internal QA; scientific agreement requires a suitable independent comparison with stated spatial support, acquisition alignment and uncertainty.
Start with stage QA and the R10C real-run walkthrough. The evidence index records the actual campaigns and their source revision/date, including synthetic contract checks. The technical supplement is a publication evidence record with historical run accounting, provenance and limitations. It is not an alternative tutorial or a live validation certificate.
For work with external Landsat observations, consult the explicit comparison scope in the drone/translation guide. Spatial support, acquisition conditions and independent reference quality remain part of the scientific study design.
The six-level scientific validation map connects numerical correctness, pipeline contracts, biological associations, synthetic sensor reduction, the separate real-Landsat pilot and incomplete ecological abundance validation.