ADR-0111: Retain RAP production correction notice for future impact assessment
Status: Accepted (reference record; no runtime change)
Date: 2026-10-29
Context
The Rangeland Analysis Platform (RAP) published a production-data correction notice on September 28, 2026 (reviewed October 29, 2026). It reports a processing error affecting 2024, 2025, and provisional 2026 production data. Effects varied geographically; annual forb/grass production was generally too high and perennial forb/grass production too low. Corrected data replaced these Earth Engine assets:
projects/rap-data-365417/assets/npp-partitioned-16day-v3projects/rap-data-365417/assets/npp-partitioned-16day-v3-provisionalprojects/rap-data-365417/assets/npp-partitioned-v3projects/rap-data-365417/assets/npp-partitioned-v3-masked
The notice's comparison figures use incorrect minus corrected values. It does not identify vegetation-cover products as affected.
Decision
Retain this notice as a discoverable reference for future RAP production/NPP integration, provenance questions, or discrepancies involving the affected years. Defer remediation until a WEPPpy consumer or retained artifact is shown to use affected data.
This record authorizes no code, parameterization, cache invalidation, or existing-run rebuild. It does not establish that any current WEPPcloud run is affected.
Decision provenance
Repository operator/Codex conversation, 2026-10-29: the operator supplied the notice and requested an ADR in case it becomes relevant. The accepted decision is to retain the evidence and a conditional assessment path.
Rationale and local evidence
Replacing data under existing asset names creates a reproducibility concern: an asset identifier alone cannot distinguish retained earlier values from corrected values. This is an inference from the upstream replacement notice, not a measured WEPPpy discrepancy.
The inspected RAP download helper uses rap-vegetation-cover GeoTIFFs. A search of repository Python sources found no literal references to npp-partitioned or rap-data-365417. These limited checks do not audit deployed services, external data mirrors, or historical run artifacts. The existing RAP_TS cover normalization decision addresses a separate cover-processing concern.
Alternatives considered
- Immediate refresh or rebuild: deferred because there is no confirmed affected consumer or run.
- Treat every RAP product as affected: unsupported by the notice and the inspected cover-data path.
- Keep only a conversation bookmark: rejected because future investigations need a durable, searchable reference.
Consequences and implementation notes
Revisit this ADR when adding a production/NPP consumer or investigating a matching data discrepancy. At that point:
- Identify the exact product, asset, years, acquisition date, local cache/mirror, and downstream consumers. Distinguish cover from production and provisional from finalized data.
- Preserve available original inputs and record retrieval metadata and content digests for comparisons. Do not infer correction status solely from an unchanged asset name or a local file timestamp.
- Compare retained and corrected inputs over the relevant geography; verify propagation into generated model inputs and consumed outputs before claiming an impact or successful repair.
- Document the affected-run scope, compatibility/regression plan, and operator-approved remediation separately. Keep historical outputs attributable to their original inputs.
Risk and rollback notes
This documentation-only decision introduces no runtime rollback requirement. The unresolved risk is unknown historical provenance if an affected production dataset is discovered later. No affected-run inventory or quantitative downstream impact has been established.
Review is event-driven by the triggers above; no scheduled implementation is implied.