[NewGraphEnvironment]

Summary

• Lake connection lines are reported apart from rearing km. The compare rollups (lnk_compare_rollup() / lnk_compare_wsg()) report FWA lake connectors (edge 1450, "Construction line, connection", in a lake or reservoir polygon) as a new rearing_lake_connection km slice and leave them out of rearing and rearing_lake km. fresh's rearing flag still carries them, so cluster_rearing still joins inlet rearing to the lake. Rules are unchanged and nothing re-classifies.
• One NULL-safe predicate, both sides. .lnk_sql_lake_connection() is exposed as a connection alias in lnk_rollup_wsg() and applied to the link side and the bcfishpass reference alike, because bcfishpass counts these lines in BT and SK rearing too. lnk_rollup_wsg()'s default rearing_km stays the flag total, since the validator's cost and the parity scripts compare the flag itself.
• Measured on Default lake and wetland rearing: connected to spawning, centrelines kept, km and ha broken out #310's ADMS / NATR snapshots (data-raw/logs/lake_connection_317/). ADMS against bcfishpass: CH +91.1 % → +38.9 %, CO +75.0 % → +28.7 %, BT +16.4 % → +2.0 %. SK rearing km fall 55–95 % on both sides, so SK parity stays 0.0 % on six bcfishpass-config WSGs. Write-up: research/habitat_thresholds.md, "Lake connection lines".

Related Issues

• Closes Lake km: should 1450 connection lines count as lake rearing? #317
• Relates to NewGraphEnvironment/sred#24

Test plan

• devtools::test(): 2585 pass / 0 fail / 16 warnings (baseline)
• New tests pin the predicate, the connection alias, and the flag-total default. They also pin the compare metrics and the link-side and bcfishpass-side SQL, plus the 9-row long format.
• Live: the link-side rollup matches a flag-level measure within 0.008 km (rollup_check.csv); bcfishpass-config parity on 9 taxonomy-pinned WSGs (parity_bcfishpass.csv)
• lintr: no new lints in R/ (lnk_compare_wsg.R carries 18 pre-existing, same as main)
• /code-check: four rounds. Round 1 found the primitive-default leak, fixed by keeping the flag default; round 2 clean; round 3 found two doc errors, fixed; round 4 clean.

Notes

• Connectors, not construction lines. The first cut also took out 1400. 1400 is "other flow / inferred connection", a construction flow line, so the predicate was narrowed to 1450 only (operator, 2026-10-08). Lake 1400 is at most 1.4 km of rearing, so only NATR BT and RB moved: BT rearing 3,709.5 → 3,710.4 km, against bcfishpass +24.3 % → +24.4 %. Wetland construction lines are Wetland construction lines (1200 / 1400) reach no rear rule in default #316.
• Output change: compare rollups gain a 9th habitat type, rearing_lake_connection, and rearing / rearing_lake km shrink. data-raw/compare_rollups.R stops on a directory that mixes pre- and post-Lake km: should 1450 connection lines count as lake rearing? #317 rollups.
• Departure from the approved plan: the plan changed lnk_rollup_wsg()'s default too. Review showed that change would break the validator's pairing of capture and cost, and the parity proof, so it was kept as the flag total.
• Follow-up rearing_km means two things: lnk_rollup_wsg() keeps lake connection lines, the compare rollups do not #319: lnk_rollup_wsg()'s default rearing_km still counts connection lines, so rearing_km means two things until rearing_km means two things: lnk_rollup_wsg() keeps lake connection lines, the compare rollups do not #319 makes it one meaning everywhere, with the validator's capture staying on the flag.
• ADMS CH / CO remain well above bcfishpass, which has no CH / CO lake rearing there. The residual is lake flow lines, Default lake and wetland rearing: connected to spawning, centrelines kept, km and ha broken out #310's position, not the line type.

🤖 Generated with Claude Code

https://claude.ai/code/session_01WJE2szSCkeV9R3qKhuSwyx