# EVT audit: present throughout, but discharge is conditional

Measured by `PYTHONPATH=diagnostics/coupled-improvement-2026-09-14/scratch/runtime .venv/bin/python docs/records/domain-seepage-investigation-2026-09-28/outlet_evidence/probe_evt.py`. Inputs and retained control outputs are hashed in `evt_receipt.json`; no native model inputs were changed.

**EVT is configured over the complete active footprint through the upper layer: 16,689 unique cells, 10,219 KAZ and 6,470 KGZ, with no missing active cells.** The native name file includes the EVT6 package. Configuration alone does not imply discharge at every cell on every day.

The native input uses exactly the model-top elevation as its evaporation surface everywhere, an extinction depth of **3 m** everywhere, and the same segmented depth-response everywhere. The potential-rate fractions are 1 at the surface, 0.5443 at one-third depth, 0.1925 at two-thirds depth, and 0 at full extinction depth. Above-surface heads do not shut EVT off; they allow the prescribed maximum rate. These rules describe this input deck, not a universal evaporation law.

All 2,637 time-series groups have a positive potential rate at some time in the retained 6,211-row table. Zero-rate days also occur; the package cannot remove more than its prescribed atmospheric-demand ceiling. Spatially averaged 2017 prescribed ceilings are **1.721 mm/day in KAZ** and **1.165 mm/day in KGZ**. These are potential rates before head limitation, not actual evapotranspiration or a country calibration.

## Executed discharge versus head shutoff

The retained control first-cycle 2017 native budget reports positive groundwater-ET loss on **365 days**, totalling **1047.807 million m³**. This is simulated discharge, not measured field ET. Independently applying the segmented input law to the retained month-end heads reproduces the executed total on all sampled dates, with maximum absolute difference **3.73e-08 m³/day**. Country flow splits are calculated from that reconciled per-cell formula, not separate native country budget records.

| Date | Country | Head too deep for EVT: cells | Zero prescribed potential: cells | Cells with discharge |
|---|---|---:|---:|---:|
| 2017-04-30 | KAZ | 6706 | 8399 | 637 |
| 2017-04-30 | KGZ | 3837 | 5787 | 252 |
| 2017-07-31 | KAZ | 7099 | 0 | 3120 |
| 2017-07-31 | KGZ | 3889 | 125 | 2579 |
| 2017-09-30 | KAZ | 7528 | 1164 | 2196 |
| 2017-09-30 | KGZ | 4078 | 2182 | 1435 |

Deep-head and zero-potential counts overlap and must not be added. These snapshots are the original control's month ends, not the latest adjusted candidate and not a year-round active/inactive mask. `evt_month_end_activity.csv` retains all sampled dates.

## Border and forcing provenance

All **315** active cross-country neighbouring pairs have identical extinction depth and the same surface-relative-to-top rule; **148** pairs use the very same rate group. The mean prescribed rates vary spatially; the maximum absolute neighbouring-pair mean difference is **1.629 mm/day**. There is no country switch in the audited depth-law or coverage parameters. This does not certify every transferred atmospheric-demand rate.

Read producer code: `docs/records/model-domain-review-2026-09-23/production_adoption/mf6/residual_et.py:39–44` assigns the nearest SWAT HRU to active cells; lines 90–105 compute `max(PET - actual ET, 0)` and save rate groups; line 120 explicitly acknowledges transfer to uncovered areas. `segmented_evt.py:23–55` maps these groups and the common depth law into EVT. This is a spatial-transfer assumption that merits physical qualification across the entire model, not evidence of a binary political-border selector. Measured array comparison also finds exact equality between the written native daily/steady rates and cell groups and the retained `forcing_feedback2_dec206_precise/native_residual_et/residual_et.npz` producer arrays (hash in receipt). The audit verifies the written native rates and their executed head response; it does not rerun or independently validate the SWAT PET/ET producer.

The residual-demand design subtracts modelled land ET before providing groundwater demand. Neither this code inspection nor the native package totals establishes double counting. A claim about double counting would require tracing shared water sources, land ET and groundwater ET together. Conversely, subtraction alone does not establish correct capillary return from groundwater into soil.

**Practical conclusion:** absent EVT is not the explanation for the border seam or the high-head cells. EVT is operating; its finite demand and groundwater-depth limitation mean it cannot be expected to eliminate flooding. Aquifer regionalization, soil recharge feedback and outlet assumptions remain separate questions.

Files: `evt_cell_parameters.csv`, `evt_border_pairs.csv`, `evt_month_end_activity.csv`, `evt_budget_reconciliation.csv`, `evt_receipt.json`, and reproducible `probe_evt.py`.