ADR-0084: CHRQIN Source Normalization and WEPP 261010
Status: Accepted Date: 2026-10-10
Context
The undeployed wepp_261009 exposed a pre-existing CHRQIN indexing error when its component interface supplied small rainfall volumes separately from return. The normalization excluded the first real sample on initialized calls, then delivered that sample with an inflated multiplier. Topanga's January 18, 1993 undisturbed outlet peak reached 707.63910 m3/s. Roger judged that physically unacceptable; observer-neutral traces and captured native-routine tests located the first inflation before channel routing.
Decision
Release the validated candidate as wepp_261010 and vendor its watershed and hillslope pair in WEPPpy. Remove the never-deployed wepp_261009 pair and their sidecars from WEPPpy's selectable binaries, retaining historical evidence.
Normalize using every delivered positive-time sample. Subtract qin0(0) only when nt0=0 and that endpoint was accumulated. Do not restrict the repair to mixed-return sources: ordinary sources must follow the same sample accounting. Retain hourly MIXPEAK, PASS v3 and accepted channel-state continuation unchanged. No model parameters, rrinit, flags, capacities, banner or general project defaults change. Deployment is a separate action, not part of this decision.
Decision Provenance
- Decision Venue: Roger/Codex workspace conversation, 2026-10-10, America/Los_Angeles; exact message time not retained.
- Participants Present: Roger Lew and Codex.
- Decision Owner: Roger Lew, explicitly requesting release as wepp_261010, vendoring and removal of undeployed wepp_261009 after the completed studies.
- Implementer: Codex.
This promotion authorization supersedes the preceding release hold. It does not erase the reported ordinary-event sediment consequences or establish their accuracy against observations.
Rationale and Alternatives
The smallest arithmetic correction fixes the demonstrated defect without changing routing architecture or inventing precision requirements. Topanga's target peak falls to 12.50683 m3/s. Mutation and disturbed hillslope results remain exactly unchanged. Cedar preserves hillslope yield and sampled outlet volume is +0.001008% versus the wepp_260803 baseline over 24 years.
Retaining the defective denominator to preserve legacy output is rejected. A reporting-only cap would leave incorrect physical source pulses. Restricting the fix to surface-return events would preserve the same error in ordinary sources. Routing reconstruction and comprehensive closure remain out of scope.
Consequences and Limits
Ordinary-event effects are real: a burned Rattlesnake peak changes 44.54780 to 20.21181 m3/s, while another event's sediment changes 140954.78 to 621811.88 kg with unchanged hillslope sediment inputs. Cedar's accounting gap is close to the original baseline but worse than 261009's. Neither event sediment accuracy nor universal conservation is established by this release.
Use fresh same-build hillslope passes for every watershed rerun. Local projects explicitly selecting 261009 must select 261010 and regenerate; do not silently alias the retired name. Historical reported results retain their true build identity. The existing paired v3 reader and combiners remain required.
Evidence
- Mechanism and native reproducer
- Four watershed comparisons
- Mutation and disturbed rankings
- Fresh Cedar comparison
Risk and Rollback Notes
Keep the bounded native regression and independently generated watershed evidence. Reassess if representative cases expose unacceptable consequences; do not compensate by tuning parameters or silently changing source eligibility. Binary rollback requires an explicit scientific/operator decision, a paired historical executable selection and fresh matching passes. Do not restore 261009 as a silent fallback. Required gates use committed resources; external watershed studies remain supplementary evidence.