[NewGraphEnvironment]

Summary

PCEA's +1.87 % BT rearing against bcfishpass (bcfishpass config, after #319) is explained and annotated. No package code changes.

• Main cause: where streams break. link breaks at every 15 / 20 % gradient position for every configured species, whatever is present (BT/ST accessible_km over-credit: access segmentation drops the accessibility frontier #223). bcfishpass builds its 15 / 20 % models only where those species are present, and breaks at minimal positions.
  • In groups with none of CH, CM, CO, PK, SK, ST or WCT (PCEA, FINA, PARS), link's segments are finer.
  • On those finer segments, spawn and rear gradient tests flip both ways, and the flips join or cut rearing clusters.
• Two rule differences sit beside it.
  • bcfishpass rears order-1 lines with parent order ≥ 5 whose known width is below the minimum. This is rearing_km means two things: lnk_rollup_wsg() keeps lake connection lines, the compare rollups do not #319's −8.2 km on connection lines and −24.8 km in PARS. The bundle sets rear_stream_order_bypass = no.
  • bcfishpass spawns only where width > 2.0 m; fresh uses ≥ 2.0 m.
• What the branch adds:
  • data-raw/habitat_diff_segments.R: a segment-level diff joined on measure overlap. It asserts that it reconciles.
  • data-raw/logs/rearing_diff_324/: attribution (mechanism.R, with net tables for the flag, the rearing row and the rearing_stream slice) and a break census (breaks.R).
  • Taxonomy class SEGMENTATION plus two class-B entries.
  • research/bcfp_segmentation_parity.md, and a CLAUDE.md status entry.

rearing row, link − bcfp km	segmentation net	moved in both directions	row net
PCEA	+26.22	70.8	+30.33
FINA	−13.04	73.4	−12.88
PARS	+9.81	84.9	−13.83 (width exception −24.78)
LKEL (salmon present)	+1.41	1.4	+1.65

Related Issues

• Closes PCEA: link (bcfishpass config) flags different BT rearing segments than bcfishpass (−8.2 km connection, +30.3 km other) #324
• Relates to NewGraphEnvironment/sred#24
• Follow-ups filed: Break streams only where a present species' access changes (both configs); bcfishpass config takes the order-1 exception #326 (break only where a present species' access changes, both configs; order-1 exception in the bcfishpass config), default config: rear order-1 streams off larger rivers on a measured width? #327 (default's order-1 position). research/bcfp_segmentation_parity.md now lines up bcfishpass's five access scripts against link's break set, with the history.

Test plan

• No R/, tests/ or inst/ change. test-lnk_parity_annotate.R passes 55 / 0 against the edited taxonomy, and the YAML parses.
• lnk_parity_annotate() tags the PCEA / FINA / PARS rearing, spawning and connection-line rows with the new entries. LKEL comes out WITHIN_TOLERANCE.
• The diff script reproduces itself: a re-run gives a byte-identical pcea_bt_pieces.csv. Coverage and flag-total gaps are 0.0 m on all four groups.
• The rearing and rearing_stream scopes were recomputed from the DB (lnk_rollup_wsg() against .lnk_sql_lake_connection() on the view) and match the CSVs to 0.01 km.
• /code-check branch: six rounds. Rounds 2–5 each found a defect inside the previous fix. The loop ended by enumeration: every claim is in planning/archive/2026-10-issue-324-pcea-bt-segmentation/review-enumeration.md, and round 6 swept all of them.

Notes

• Upstream SQL was read at smnorris/bcfishpass@620dbe4. The cited access and habitat files changed on or before 2026-04-16, before the local fresh.streams_vw_bcfp snapshot.
• Open, not done here. A follow-up issue is drafted for review, not filed. It asks:
  • should the bcfishpass config break where bcfishpass does? That would re-segment every group.
  • should the bundle take the order-1 bypass? fresh's frs_order_child() has no width test, so as written it would also admit NULL-width lines that bcfishpass leaves out.
  • should spawning use width > 2.0 m, as bcfishpass does?
  • the 1.8 km PCEA access residual.
• The taxonomy's hors-class-stream-order-bypass still reads INTENTIONAL_FRESH_DEFERRED. fresh#158 is closed and the bundle declines the bypass. Its status is left for that follow-up.

🤖 Generated with Claude Code

https://claude.ai/code/session_01WJE2szSCkeV9R3qKhuSwyx