Separate coastal aggregates and confluence units in HydroBASINS–SWORD FACC validation #233

The adopted bifurcation correction preserves facc_total. Track reference pairing independently of channel allocation.

The read-only delta/tidal audit reconstructed one area/Pfaf-consistent HydroBASINS level-6 feature for all 575 frozen records. Original feature/pour-point join IDs were not retained in the recovered reference CSV. Of 285 misses above 20%, 252 use COAST=1 aggregate polygons. These can combine independent watersheds and multiple pour points; aggregate UP_AREA is not an individual mouth's drainage area. The 28 selected worst cases comprise 23 coastal aggregates and five branch/confluence comparison-unit conflicts.

Example: 44220100015 receives Tonle Sap and Mekong parents. The frozen reference is 88,253.8 km², and the adjacent Mekong unit is 665,658.8 km². Their combined 753,912.6 km² is close to unchanged candidate facc_total 751,289.059 km². This is a diagnostic, not a replacement reference selected by model fit.

Follow-up:

• Preserve all 6,262 original rows and raw scores.
• Attach feature identity, COAST, pour-point multiplicity and pairing evidence.
• Build a separately versioned per-river reference roster from actual pour-point catchments, with pairing rules independent of candidate area.
• Continue evaluating facc_total only; never split it or adjust SWORD widths to fit reference areas.

Local evidence: output/facc_detection/conservation_followup_20261026/delta_adoption_audit/references/ and REVIEW.txt. Related: #232. This follow-up does not block adoption of the conservative bifurcation budget correction.