# Why the exceedance map changes at the KAZ–KGZ boundary

Measured by `scripts/diagnose_modflow_border.py` against the same G2 workspace shown on the map. This is a read-only E0 diagnosis; no solver, parameter change or production adoption.

The model contains an abrupt change of representation at the inherited Kyrgyz footprint. **Inference:** the drain-elevation change is a direct explanation for why the above-terrain threshold is crossed much more readily on one side. The property change can also affect the result. These simultaneous differences do not establish their separate causal contributions without a controlled comparison.

## Measured neighbouring-cell comparison

The broad arrow-region inspection rectangle is UTM 43N E 340–450 km, N 4740–4800 km. It contains 157 active cross-border neighbour pairs; this rectangle is an inspection window, not a digitization of the arrows. Median properties of those paired cells:

| Property | KAZ side | KGZ side |
|---|---:|---:|
| Drain elevation relative to ground, m | 0.0 | -2.5 |
| Drain conductance, m²/day | 4000 | 1250 |
| Upper horizontal conductivity, m/day | 10.000 | 1.000 |
| Upper vertical conductivity, m/day | 0.100 | 0.005 |
| Upper K × full thickness, m²/day | 500 | 50 |
| Mean-stress recharge, mm/day | 0.233 | 0.223 |

The upper-layer full-thickness flow-capacity contrast is not a tenfold contrast in the whole aquifer: the lower-layer medians are 750 and 850 m²/day, respectively. Saturated upper-layer capacity can be lower than K × full thickness.

There are 116 neighbouring pairs with terrain seepage on the KAZ side and a legacy background drain on the KGZ side. The KGZ outlet therefore starts removing water before head reaches terrain. Its lower conductance does not mean less discharge at the same head relative to ground: at ground-level head the typical KGZ drain formula gives 3125 m³/day, while the terrain-level KAZ outlet gives zero. This is an equation evaluation, not an observed or simulated discharge measurement.

## Connectivity and interpretation

Measured native GRB connectivity: **315/315** active upper-layer cross-border neighbour pairs are connected. There is no missing intercell face in these pairs. This checks groundwater-grid connections, not the real-world surface drainage network or the full coupled receiver ledger.

Read from `docs/records/model-domain-review-2026-09-23/production_adoption/physical_basis/qualify_governed_physics.py`, function `legacy_drainage`: background drains are restricted to `data["old"]`, at terrain minus 2.5 m. Read from `lowland_mf6_candidate/run.py`, property construction: the extension uses default properties while inherited zones keep their transferred properties. The seam follows the inherited footprint, which largely follows the border here; it is not a demonstrated hydrogeological boundary. The generic surface-seepage conductance uses vertical conductivity, cell area and half-layer thickness, not distance or connectivity to a mapped surface drain (G2 build receipt `surface_seepage` and `physical_basis/probe_surface_seepage.py`).

The original coloured layer is an **ever-exceeded threshold over the transient calendar**, not a head surface at one instant. That binary threshold can make a gradual change in head relative to terrain look like an abrupt wall. The new steady-state head and head-minus-terrain views let you inspect that distinction. Native same-date examples are in `border_diagnosis_receipt.json#same_date_examples`.

## Added map evidence and next decision

The map includes both layers' saved steady-state heads, 10 m contours, signed head-minus-terrain, applied drain types and upper-layer K × thickness. Values come from the first saved native state, verified against the STO steady-state flag. The build receipt describes calendar-mean stresses; this is a synthetic initial solution, not observed piezometry or proof of physical equilibrium.

Before treating this pattern as a physical KAZ–KGZ difference, review a consistent, evidence-based drainage conceptualization across the seam. Then isolate drain elevation/coverage and hydraulic-property effects in separate controlled comparisons. Do not simply extend the KGZ blanket drains or tune K to erase the coloured cells.

Reproduce: `PYTHONPATH=diagnostics/coupled-improvement-2026-09-14/scratch/runtime .venv/bin/python scripts/diagnose_modflow_border.py`. Inputs, extracted-head hash, paired values and example dates are in the accompanying receipt and CSVs.