diff --git a/ui/src/components/BridgeView.jsx b/ui/src/components/BridgeView.jsx
index c2a14b7..c959db8 100644
--- a/ui/src/components/BridgeView.jsx
+++ b/ui/src/components/BridgeView.jsx
@@ -12,12 +12,6 @@
 
 import { useState, useEffect, useRef, useCallback } from 'react'
 
-// pf_segment / pf_bucket are computed server-side but are ordinary columns in
-// the loaded table, so a slice keyed on one of them still filters directly.
-// Mirrors the copy in Forecast.jsx -- dropping them here is what let a
-// bucket-keyed slice bridge more rows than the operation it was drawn from.
-const COMPUTED_SLICE_COLS = new Set(['pf_segment', 'pf_bucket'])
-
 const UP      = '#2a78d6'   // increase
 const DOWN    = '#e34948'   // decrease
 const ANCHOR  = '#6b7280'   // baseline / current — neutral, 4.83:1 on white
@@ -290,6 +284,14 @@ export default function BridgeView({
         // The union of the selected slices — the same reach an operation would
         // have. Perspective view filters are AND-only, so each slice needs its own
         // view; rows matching more than one slice are counted once.
+        //
+        // pf_bucket/pf_segment are deliberately left out here, even though a
+        // click can carry one (e.g. the column axis split by bucket): the walk
+        // needs rows from every bucket, not just the one the slice landed on,
+        // or there is nothing left for "Compare to" to compare against and
+        // every basis collapses to the one bucket fetched. Which bucket the
+        // walk starts and ends on is decided inside buildSteps, from
+        // forecastBucket, not by filtering it out of the fetch.
         const dimNames  = new Set(colMeta.filter(c => c.role === 'dimension').map(c => c.cname))
         const dateNames = new Set(colMeta.filter(c => c.role === 'date').map(c => c.cname))
         const schema = await tableRef.current.schema()
@@ -301,18 +303,33 @@ export default function BridgeView({
             default: return String(val)
           }
         }
+        // The pivot's own filter (e.g. sseas_e <= 2027) is not part of a
+        // clicked slice — perspective-click reports only the cell's own
+        // dimensions — so it has to be read off the viewer and applied
+        // alongside every per-slice view, the same reason Forecast.jsx's
+        // ledger totals do this. Without it, a grid scoped narrower than the
+        // whole version bridges rows the grid is hiding.
+        let viewFilter = []
+        let expressions = {}
+        if (viewerRef?.current) {
+          const cfg = await viewerRef.current.save()
+          viewFilter = (cfg.filter || []).filter(f => Array.isArray(f) && f.length >= 2)
+          expressions = cfg.expressions || {}
+        }
         const seen = new Set()
         rows = []
         for (const sl of slices) {
-          const f = Object.entries(sl)
-            .filter(([c]) => COMPUTED_SLICE_COLS.has(c) || dimNames.has(c) || dateNames.has(c))
+          const sliceFilters = Object.entries(sl)
+            .filter(([c]) => (dimNames.has(c) || dateNames.has(c)) && !viewFilter.some(vf => vf[0] === c))
             // A null is filtered with `is null`, not `== null`: the engine
             // matches on type, and a null threshold against a string column is
             // the same dropped predicate that widens a slice back toward its
             // parent group — see totalsFor in Forecast.jsx for the same fix.
             .map(([c, v]) => (v === null ? [c, 'is null'] : [c, '==', typed(c, v)]))
-          if (!f.length) continue
-          const view = await tableRef.current.view({ filter: f })
+          if (!sliceFilters.length) continue
+          const viewCfg = { filter: [...viewFilter, ...sliceFilters] }
+          if (Object.keys(expressions).length) viewCfg.expressions = expressions
+          const view = await tableRef.current.view(viewCfg)
           const part = await view.to_json()
           await view.delete()
           for (const r of part) {