# ADR-0053: Offline Staley M3 soil thickness evaluation

Date: 2026-09-08

## Status

Accepted for offline evaluation, with subsequent owner decision on 2026-09-09:
SSURGO primary and original STATSGO THICK fallback. Weathered-material treatment,
partial-coverage acceptance, and fallback granularity remain unresolved;
production integration is not implemented.

## Decision provenance

- Venue: user/Codex repository session, 2026-09-08 Pacific / 2026-09-09 UTC.
- Participants: requesting user and Codex.
- Decision owner: user for scientific source/availability; Codex for bounded
  reproducible study implementation under the supplied ExecPlan.
- Implementer: Codex.
- Change: no prior M3 soil helper; add strict raw-interval soil candidate and
  labeled all-recorded-layers sensitivity. Mean cm / 254 supplies S. No existing
  production parameterization or WEPP soil behavior changes.

## Decision and rationale

Follow the [offline contract](../../wepppy/nodb/mods/postfire_debris_flow/docs/m3_soil_thickness.md)
and [predeclared protocol](../work-packages/20260908_staley_m3_soils/artifacts/study_protocol.md).
Require valid zero-start contiguous intervals for selected component thickness;
retain sum/union/endpoint diagnostics on rejected records. Deduplicate stable
IDs separately from depth ranges. Exclude explicit R in the strict candidate;
reject weathered/mixed/unknown material rather than invent scientific treatment.
Compare retaining all recorded layers as a separate sensitivity.

The original USGS SAS sums intervals and renormalizes over nonmissing component
weights without a bedrock filter. SSURGO has different source semantics and
vintage. Expose coverage instead of equating known support to full coverage.
Use complete-only full-catchment outputs and partial diagnostic means, with no
approved numeric cutoff. Soil support is independent of dNBR policy.

## Alternatives

Deepest endpoint fills gaps; summing cumulative bottoms double-counts depth;
union hides alternate overlapping records; generated WEPP profiles can be
clipped or extended. None supplies the selected strict estimate. Fixed-depth,
zero-fill and automatic source fallback are rejected. Finer mapping alone is
insufficient to accept a substitute predictor.

## Evidence, risks and rollback

Evidence and measured findings live in the
[work package](../work-packages/20260908_staley_m3_soils/package.md).
The strict candidate may withhold useful observations; all-layer sensitivity
may include bedrock. Neither demonstrates debris-flow predictive validation.
Survey endpoint censoring and original survey material inclusion remain open.
Retire the offline helper/artifacts if their semantics prove unsuitable; no
persisted production state or deployment rollback is involved.

## Measured outcome

The paired 12-outlet study recommends retaining original STATSGO pending a
scientific source decision. Strict SSURGO support is 0–99.25%; all-layer support
is 84.24–99.25%, so neither supplies complete full-catchment coverage. Maximum
common-support probability effects are 19.35 and 11.52 percentage points,
respectively. These are diagnostic effects, not published acceptance limits.
See the [decision report](../work-packages/20260908_staley_m3_soils/artifacts/soil_decision.md).
This historical recommendation was superseded by the owner decision below.
Weathered-material treatment and coverage rules remain unresolved.

## Subsequent owner decision: source priority

- Venue/date: user/Codex repository conversation, 2026-09-09 UTC.
- Participants and decision owner: requesting user; Codex records the decision.
- Implementer: Codex for documentation; runtime implementation pending.
- Change: select SSURGO as primary and original STATSGO THICK as fallback,
  superseding the study recommendation to retain STATSGO as primary.
- Rationale: prefer SSURGO's more detailed soil information and use STATSGO for
  broad coverage. Source differences do not demonstrate inferior SSURGO accuracy.
- Alternatives rejected: making original-source agreement the acceptance test;
  imposing complete component support on SSURGO while equating nonmissing
  STATSGO raster cells with complete source observations.
- Scope: approve source priority, not the strict study material filter, 100%
  coverage gate, fallback granularity, or an arbitrary numeric threshold.
- Evidence: original STATSGO processing itself renormalizes over nonmissing
  components; the paired study measures source sensitivity, not outcome accuracy.
- Risks and recovery: define usability and fallback boundaries before runtime
  integration; expose source contributions and invalid/unavailable support.
  Preserve version 1 offline results for reproducibility. No runtime change or
  live project mutation is made by this amendment.

The detailed source contract is maintained in the canonical postfire
specification and soil document; the closed package remains immutable history.
