Scientific validation: what can we justify?

Environmental data differ in their meaning as well as their file formats. Existing semantic systems describe different parts of those meanings. Bounded AI can help navigate the choices, while tested operations execute only reviewed plans. We can now measure failures and identify decisions that still need scientists.

This 22 October 2026 pass measured three real vocabulary graphs, specified 25 canonical observations, investigated 18 candidate metadata records and their 153 pairs, and completed 100 live Verde calls. These objects have different evidence scopes; their counts do not establish ecological validity.

• Sentinel expert package retains ten issue records and every historical veto.
• Semantic and dataset experiments separate graph measurements from provisional scope assessments.
• NEON + Sentinel case proposes environmental context, not measurement equivalence.
• Verde constraint experiment shows why valid JSON is not enough.

Read the separate 20-page scientific validation report and rebuild its 18 figures and evidence packages. Repository links require publication of the local changes; no deployment is claimed.

No real scientific release was approved. No new raster or NEON observation files were downloaded. Small public vocabulary graphs and product/site metadata were retrieved explicitly and retain attribution, versions where declared, and hashes. The older pipeline-state report remains a dated historical snapshot; its unmeasured claims describe the earlier audit.