Reassess all FACC flags, quality and source metadata for current NetCDF candidate #239

Completed full reassessment of drainage-area metadata and a new six-region NetCDF candidate. Prior candidates, numerical areas, geometry, routing, source admissions, adopted selector, databases and published releases are preserved.

Final candidate: data/exports/v17c_beta/candidates/bifurcation_unsplit_total_reassessed_20261027_yushan/; six data files plus the plain-language README are in netcdf/ (16.35 GB). Reassessment code/evidence: output/facc_detection/variants/facc_metadata_reassessed_20261027/.

Every one of 248,673 reaches was reassessed, and every one of 11,112,454 node labels was regenerated by exact reach_id join. Fresh source-fill and topology-classification replays passed in all 60 basins. The current source/class values agree with the numeric model; old facc_flag and facc_quality never select a new label.

facc_flag now uses current evidence, with unresolved checks prioritized as suspect, followed by shared-source warnings, unvalidated delta shares/capped estimates, other width shares, fills, and no triggered screen (ok). facc_quality is regenerated as rederived_v3:; it is explicitly a source-method label, not measured accuracy. New facc_review lists all applicable processing/review reasons on reaches and nodes. All related variable/global attributes, source/class counts and dictionaries are current.

Current reach flags:

• ok: 214,249
• suspect: 6,282
• same_river: 1,328
• delta_arm: 8,718
• braid_share: 12,562
• filled: 5,534

Primary flags change at 9,513 reaches and 418,659 nodes; every quality label is regenerated. Unchanged labels were also reassessed. All 1,428 current original-scope tail-sum shortfalls have non-ok flags: 1,421 suspect and seven shared-area warnings. All 29 existing source-placement cases, both direction-conflict reaches and the weak outlets are explicitly marked suspect. The 41 zero-allocation context steps affect 23 receivers, which now carry zero_transfer_total_step warnings. Nothing was silently converted to physical certainty.

Width-share provenance is tracked per actual channel source across frozen fractions, aliases and explicit source boundaries. A cap alone cannot create a width-share flag; a complete rejoin removes the split-share warning for that source. Current same-river/native-path warnings, upstream warning chains, width/area screens, numerical caps, fills, headwater priors and current junction diagnostics were recalculated. Width/area screens do not treat a partitioned share as an independent basin.

Verification: 21 rule tests pass. An independent reverse-source calculation checked 285,691 source-at-reach cases in 5,617 reaches across all 60 basins, including all alias receivers and complete rejoins. Independent saved-file readback checks every current metadata value, every node-parent label join, current primary reasons, all numerical and non-metadata preservation, all original 18,241 junctions, exact global counts, code/product/file hashes, and 599 protected inputs before/after. All checks pass. Prior candidate and database metadata are unchanged; no database was opened.

Reference predictions were recalculated from the unchanged numerical values and rejoined to current flags. All 17,320 paired reference rows (two area fields for 6,262 HydroBASINS, 2,297 SVS and 101 USGS rows) retain identical predictions. Whole-area within-20% counts remain 5,310, 2,119 and 98. New current-flag score tables are included. Frozen historical cohort labels are explicitly separated from current flags.

The package includes current flag transitions, the 1,428 junctions with current labels and new/worsened status, full current reference records, reason definitions/counts, the field dictionary, rule code/tests, independent evidence verification, and final NetCDF verification. The upload README is plain language, distinguishes standard SWORD facc from our added fields, and states this is our modified v17c test candidate. Remaining physical matching/direction/source questions remain warnings; ok is not a guarantee of physical accuracy.

The new files have not been sent, uploaded or published by this task. Existing follow-ups #233–#235 remain open. Related numeric candidate: #238.