Reproduce external drainage-area checks and clarify junction-count populations #232

Reconstructed the external drainage-area validation on the original six shipped GeoPackages. All comparisons with the supplied result tables match the reach IDs and checked values, including all 2,660 junction-shortfall classifications.

Main reproduced results:

• After the stated exit exclusions: 365 existing shared-area-label cases, 512 possible channel rejoins, and 552 other junctions.
• Forty possible rejoins have shortfalls above 10%; 17 have a delta-arm inflow.
• facc_total decreases: 314, including 251 tidal and 63 non-tidal.
• Delta child-sum excess: median +10.98% for pure-delta children and +41.55% for mixed delta splits; 569/669 mixed splits exceed +10%.
• Ordinary channel and distance checks pass. The 12 supplied major-river comparisons all pass ±20%, with a median absolute difference of 2.76%.

Comparing actual IDs shows the historical documentation counts are different populations: only 55 of the historical 305 cases belong to the external 512-rejoin category after exit exclusions, and only 201 of the historical 469 belong to its 552-other category. The historical product is identical to the shipped drainage values and flags; this is not caused by later candidate changes.

Two reconstructed scope conditions should be stated in documentation:

1. The 2,660 and zero-below-largest checks exclude junctions receiving from a reach that itself splits. Broader comparisons over all junctions yield 2,726 summed-area shortfalls and 46 below-largest comparisons, which require allocated incoming shares before physical interpretation.
2. The supplied 387 full-copy labels test whether the largest child is within 0.1% of the parent. Testing any child yields 461; the additional 74 have another child larger than the parent.

The attachment has result tables, not analysis code. The implementation explicitly records its reconstructed rules and numeric tolerance. Shared upstream connections and numerical shortfalls are diagnostics, not proof of independent physical contributing catchments.

Local source: scripts/analysis/reproduce_yushan_validation.py (uncommitted). Outputs, input hashes, exact source snapshot, matched reach lists and preservation verification: output/validation/yushan_reproduction_20261022/.

No scientific data or candidates were changed. Track the documentation clarification and any subsequent investigation of the matched cases here.