Bundle review table

04-water-retention

README.md

# 04-water-retention

**Claim.** Soil water retention is set by the ratio of grain size to

**Verdict: Supported**

porosity alone R2=0.212; texture adds dR2=0.375; median within-bin p90/p10 of theta33 = 2.5x.

Files: `hypothesis.md` (posted claim + contract), `run.py` (reproducible),
`results.json` (numbers), `figures/` (plots).

hypothesis.md

# Hypothesis 04 — Water Retention Follows Grain Size

**Source.** Incoprea Research hypothesis notebook, `frontier-theories-16 #37`
(identifier H291). Reproduced from the posted entry.

**Claim.** Soil water retention is set by the ratio of grain size to
pore-throat size rather than by total porosity, so two soils with equal
porosity can hold very different amounts of water.

**Proposed mechanism.** Retention is governed by the pore-size distribution
rather than by total pore volume; texture sets the pore-throat sizes.

**Predictions.**
- P1 — Texture terms add a large R² gain over porosity alone.
- P2 — Within equal-porosity bins, retained water varies several-fold
  (2–5×), so porosity alone badly mispredicts retention.

**Refutation condition.** Porosity alone explains most of the variance
(R² > 0.8) and equal-porosity soils differ by less than ~1.3×.

## Test contract (pre-specified)

Data: USDA KSSL `lab_physical_properties` (Soil Data Access REST), 100,000
lab horizons with measured θ at −33 kPa.

- Outcome: `water_retention_third_bar` (θ33, % vol), `water_retention_15_bar`
  (θ1500, % vol).
- Porosity: `1 − bulk_density_oven_dry / particle_density`
  (particle density from `particle_density_less_than_2mm` when present, else
  2.65 g/cm³). The directly measured `water_retention_0_bar_sieve` is 99.9 %
  missing, so porosity is computed from bulk density — this deviation from the
  original note is flagged.
- Texture: clay, silt, and the sand sub-fractions (fine, very fine, medium,
  coarse, very coarse).
- Regress θ33 on porosity alone vs porosity + texture; report R² gain and AIC.
- Within 0.02-wide porosity bins, report the p90/p10 ratio of θ33 and θ1500.

**Support if** texture adds a large R² gain and equal-porosity bins show 2–5×
spread. **Refute if** porosity alone reaches R² > 0.8 and bins differ < 1.3×.

Included code and results

• tests/04-water-retention/results.json
• tests/04-water-retention/run.py