You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
blocks_species text[] (computed once in lnk_barriers_unify) bakes a binary per-species blocks/doesn't onto every barrier at unify time. Two problems surfaced in #196 / #200 (see RUNBOOK.md §5):
It conflates two orthogonal axes bcfp keeps separate: natural access (per-species, gradient-typed, observation/habitat-overridden) vs. anthropogenic descriptor (dam/pscis/modelled, passability-typed). Dams get baked into access wrongly.
It's computed before the override is known, so the "fish above ⟹ passable" evidence (observations/habitat) can't subtract from it.
bcfp has no binary blocks predicate — it builds per-species access sets from ingredients (gradient class, falls, subsurface) + evidence overrides, and tracks anthropogenic barriers separately. "Nothing is black-white."
Direction (research/design)
Carry barrier ingredients on the unified table: barrier_type, gradient_class, passability (CABD status), up_passage_type (fishway), source — instead of (or alongside) the pre-baked blocks_species.
Classify access late + per-context, per species, reusing fresh's existing label / label_block gradation rather than a fresh binary.
dam-override: let dams be overridden out of the relevant set by the SAME evidence rules as natural barriers (lnk_barrier_overrides: observations / confirmed habitat / control). Many CABD dams exist on paper but are passable (decommissioned, partial, fishway, or fish demonstrably above). Reuse the rules engine; do NOT build a bespoke fishway model — the name shouldn't bake in the mechanism (fishway is one case).
This is a deliberate departure from bcfp (bcfp never overrides dams per-species) — opt-in, and it breaks the exact-reproduction correctness bar (CLAUDE.md). Sequence it AFTER the match-bcfp access fix.
Activity
PWF #196: link filed issues #200 (access-set fix) + #201 (blocks_spec…
v0.40.3: persist per-source flags in streams_access (#196) (#199)
v0.40.4: per-species access reproduces bcfp barriers_<sp> so dam-down…