[NewGraphEnvironment]

Summary

• The balance's monthly shares now come from donor gauges. Each watershed takes the inverse-distance mean of the observed monthly shares of the five calibration gauges most like it, matched on elevation, location, snowmelt timing and peak SWE. These replace Snow predictors for the monthly shares #16's twelve regressions.
  • The rule fixed every candidate setting and was committed in scripts/wb_validate.R's header (a8ca570) before any candidate code existed.
  • Median per-gauge share NSE, 315 gauges: regression 0.8735, donor 0.9389, blend 0.9187 under blocked CV; leak-free 0.8735 / 0.9353 / 0.9209.
  • On 110 short-record test gauges no fit uses: 0.772 / 0.837 / 0.812.
• Code:
  • wet_share_fit(method = "regression" | "donor" | "blend").
  • wet_donor_mean(), shared with Sub-monthly flow per segment: a daily climatology for both sources, with weeks and windows derived from it #70's daily shape; the shape's output is identical, and tested.
  • The scoring code and every consumer read share_method.
  • scripts/wb_share_donor.R breaks the result down.
• Run on m4 and published (live build 8ca2dee):
  • The annual is identical in all ten AET variants, and cfu still ships. Month 0 is byte-identical in every basin, and no pick moved.
  • Every publish value check passes. The BUILD after the upload is 67 of 67 identical to the live build.
• The monthly-flow article is rewritten on the new build. On a calibration gauge's segment the balance now carries the gauge's own monthly pattern. At 08EE003, an ungauged site, it is several times the gauge in the late-summer windows; the article states that error as ours.

Related Issues

• Fixes Monthly shares from donor gauges' day-of-year records: #70 measured 0.939 against the shipped 0.874 #102
• Relates to NewGraphEnvironment/sred#40
• Leaves Publish segment_daily: the balance's daily shape and routed PCIC's day-of-year climatology per segment #99 (publishing the daily data) open. On the new months, Sub-monthly flow per segment: a daily climatology for both sources, with weeks and windows derived from it #70's within-month shape matches the donor's free 365-day curve (0.909), so "months only, daily later" costs nothing.

Test plan

• devtools::test(): 1227 passed, 0 failed (after the R changes; nothing under R/ changed since)
• Both changed articles (monthly-flow, segment-discharge) render from the regenerated bundles
• The article's new window check was shown to fail when its annual term is dropped
• Plan review, then four /code-check rounds (records in the archive)
• Issue links spot-checked (Snow predictors for the monthly shares #16, Sub-monthly flow per segment: a daily climatology for both sources, with weeks and windows derived from it #70, Publish segment_daily: the balance's daily shape and routed PCIC's day-of-year climatology per segment #99, Routed PCIC over 1981-2010: switch the pcic run and republish #103, sred#40)

Notes

• Owed after the merge:
  • Move data/logs/102/ from m4's wet-102 worktree into the main checkout, then remove the worktree.
  • Update the m4 and m1 memories.
  • Write the daily-shape outcome into Publish segment_daily: the balance's daily shape and routed PCIC's day-of-year climatology per segment #99's body.
• Record: research/water_balance_method.md §0, "Monthly shares from donor gauges (Monthly shares from donor gauges' day-of-year records: #70 measured 0.939 against the shipped 0.874 #102)"; archive planning/archive/2026-10-issue-102-donor-monthly-shares/.

🤖 Generated with Claude Code

https://claude.ai/code/session_01TaHJkAqd9xHDZQmhpQoVbW