# Drainage seam: demonstrated cause and reconciliation boundary

**Conclusion.** The KAZ–KGZ pattern is strongly controlled by a change in represented drainage at the inherited model footprint. The implementation carries forward below-ground background drains on the old footprint and uses terrain-level seepage on the extension. The source selector is model history, not a demonstrated change in drainage geology at the political border.

**Measured** by `scripts/probe_modflow_drainage_seam.py` and `scripts/verify_modflow_drainage_replays.py`: an isolated replay of **1 April–1 May 2017** reproduces the original G2 daily heads exactly (maximum absolute difference 0 m). All perturbations keep that same saved starting state, daily recharge, pumping, mountain inputs, river-stage and ET series, terrain, lower-layer properties and solver. Input-file difference checks pass. No source-model inputs were changed.

## Causal result

**Measured**, recomputed into this record by `scripts/record_modflow_drainage_findings.py` from `replay_results.json`; counts refer to upper-layer cells crossing terrain +0.01 m at least once in the window, within E 340–450 km / N 4740–4800 km (UTM 43N):

| Isolated change | KAZ affected cells | KGZ affected cells |
|---|---:|---:|
| Unchanged G2 replay | 1,267 | 398 |
| KGZ background drain level moved to terrain | 1,270 | 1,479 |
| KAZ seepage level lowered by 2.5 m | 51 | 396 |
| KGZ background replaced by terrain-seepage recipe | 1,270 | 1,514 |
| KAZ upper horizontal K reduced from 10 to 1 m/day in inspection window | 1,267 | 398 |


Changing only KAZ drain elevation removes **96.0%** of the affected KAZ cells in this short replay. Changing only upper horizontal conductivity leaves the KAZ affected-cell count unchanged, although individual heads and the durations of exceedance do change. Conversely, lifting only the old background-drain elevation increases the KGZ affected population. Together these tests identify a strong drainage-elevation mechanism under the actual forcing; they do not establish that the lowered KAZ drains are physically justified.

**Measured:** replacing the old background with the complete terrain-seepage recipe produces a median final KGZ head rise of 10.57 m in the inspection window. That arm changes both lip and conductance: it cannot isolate conductance alone. In the inherited plain-zone cells, the default vertical-resistance formula gives a much weaker outlet than the old background treatment. Simply replacing all old drains with terrain seepage is therefore not a validated remedy either.

The exact same-date cell comparisons are in `same_date_cell_comparisons.csv`; package discharges, storage and ET are retained in `replay_package_flows.csv` and `mean_package_flows.csv`. Both groundwater layers balance separately: maximum relative compartment residual across all arms is 4.12e-13; maximum total relative residual is 1.08e-13. These are native MODFLOW balances, not coupled surface-water closure.

## Why the seam exists

**Read — explicit implementation:**

1. `docs/records/model-domain-review-2026-09-23/production_adoption/physical_basis/qualify_governed_physics.py:37` restricts background-drain eligibility to `data["old"]`. Line 45 places these drains 2.5 m below terrain. Its opening docstring explicitly says that legacy drainage is not extended into new terrain.
2. `docs/records/model-domain-review-2026-09-23/production_adoption/physical_basis/probe_surface_seepage.py:152` excludes existing river/drain cells from generic seepage; line 168 uses Kv × cell area / half-layer thickness for conductance. This combination makes the old footprint select one drainage regime and the extension select the other. Generic seepage is not a mapped channel connected by a measured lateral drainage distance.
3. The owner ruling governing the earlier footprint is **DEC-195**, in the sole decision register `docs/chu_swatplus_decision_register.md`. That ruling is not evidence for a hydrogeological discontinuity at the country border and is not amended here.
4. `docs/phase2/chu_groundwater_fd_gw5_behavioral_calibration_2026-08-11.md:440–516` describes the older two-tier rationale: mapped karasu drains plus a background representation for unresolved channels, then adjustment of its lip and conductance range against discharge-partition criteria. It is a regional approximation, not a measured drain in every grid cell.

**Read — original source checked visually:** the scanned *Hydrogeology of the USSR, vol. XL, Kyrgyz SSR* (1971), printed/PDF p. 52, describes gently inclined alluvial/proluvial loamy fan-margin plains with a dense network of shallow erosion incisions, commonly spring-fed karasu. The stated incision range is generally up to 2–5 m, with rarer deeper examples. The passage distinguishes landforms. It does not prescribe a drain under every cell, assign a conductance, or justify changing the process at a political border. Local file: `data/raw/chu_reference_literature/gidrogeologiya_sssr_t40_kirgizskaya_ssr_1971.pdf`.

**Read — coverage limitation:** `docs/records/model-domain-review-2026-09-23/lowland_mf6_candidate/findings.md:3–30` records the corrected explicit-grid KAZ terrain/winter-water inventory. These are screening arcs and possible receivers, not qualified spring/drain connections. They cannot yet justify a blanket extension of the old drain treatment. Regional cross-border aquifer context also does not supply local drain stages or hydraulic properties.

**Measured in the preceding border probe:** all active upper-layer cross-country neighbour pairs remain connected (`reports/figures/modflow_problem_cells_2026-09-28/border_diagnosis_receipt.json`). This investigation identifies a discontinuity of assumptions, not missing groundwater-grid faces. Actual surface drainage and the coupling receiver ledger require their own evidence.

## Reconcile by physical setting, not country

Apply one evidence rule across both countries: identify comparable fan-margin plains, floodplains, incised spring-fed channels, artificial collectors and undrained terrain. Use observed/mapped drain elevations and receiving pathways where supported; represent unresolved drainage explicitly with a common, tested rule only where the corresponding landform and observations support it. Keep river exchange distinct from drains and avoid duplicate discharge pathways. Preserve real geological contrasts in K/thickness where evidence supports them.

First review the corrected KAZ terrain/winter-water arcs alongside the KGZ karasu evidence using the same spatial-support criteria, then qualify drain elevations, conductances and receivers. Declare an evidence-based cross-border candidate before a long-term, coupled comparison. Do not add drains merely to eliminate coloured cells, erase genuine geology, or treat either diagnostic intervention as adopted.

## Limits and reproducibility

This is **E0 causal reconnaissance**, not production correction or independent scientific acceptance. It covers one wet month and uses the inherited state before the intervention. Slow property responses, other seasons, the full historical exceedance envelope, physical head validation and changed groundwater-return feedback remain untested. The lower layer and vertical conductivity were held fixed in the K-only test. Numerical balance does not establish physical realism.

Commands (repository root):

- `PYTHONPATH=diagnostics/coupled-improvement-2026-09-14/scratch/runtime .venv/bin/python scripts/probe_modflow_drainage_seam.py`
- `PYTHONPATH=diagnostics/coupled-improvement-2026-09-14/scratch/runtime .venv/bin/python scripts/verify_modflow_drainage_replays.py`
- `.venv/bin/python scripts/record_modflow_drainage_findings.py`

The runner refuses to overwrite an existing workspace. Archive/remove only its owned temporary replay directories before a fresh execution. Durable inputs are identified by the replay input receipt; perturbations, head summaries, native execution logs, package fluxes and verification are retained here. Raw replay head/budget output is disposable after those checks.