diff --git a/cmd/server/config.go b/cmd/server/config.go
index 98a92e22..3a93dee6 100644
--- a/cmd/server/config.go
+++ b/cmd/server/config.go
@@ -8,6 +8,7 @@ import (
 	"path/filepath"
 	"strings"
 	"sync"
+	"sync/atomic"
 	"time"
 
 	"github.com/meshcore-analyzer/dbconfig"
@@ -47,9 +48,18 @@ type Config struct {
 	// operator refuses to fix.
 	NodeBlacklist []string `json:"nodeBlacklist"`
 
-	// blacklistSetCached is the lazily-built set version of NodeBlacklist.
-	blacklistSetCached map[string]bool
-	blacklistOnce      sync.Once
+	// blacklistSetPtr holds the active lookup set as an atomic pointer.
+	// Read path is a single atomic load — no mutex, no sync.Once. Writers
+	// always replace the whole map; readers see either the old or the new
+	// map as a single value, never a partially-built one.
+	blacklistSetPtr atomic.Pointer[map[string]bool]
+
+	// blacklistGen is a monotonic generation counter bumped every time the
+	// blacklist mutates via SetNodeBlacklist. Callers that cache responses
+	// keyed by pubkey (e.g. /api/nodes/{pubkey}/reach, #1629) include this
+	// generation in their cache key so any blacklist change naturally
+	// invalidates prior entries on the next request.
+	blacklistGen atomic.Uint64
 
 	Branding   map[string]interface{} `json:"branding"`
 	Theme      map[string]interface{} `json:"theme"`
@@ -636,31 +646,78 @@ func (c *Config) LiveMapMaxNodes() int {
 	return v
 }
 
-// blacklistSet lazily builds and caches the nodeBlacklist as a set for O(1) lookups.
-// Uses sync.Once to eliminate the data race on first concurrent access.
-func (c *Config) blacklistSet() map[string]bool {
-	c.blacklistOnce.Do(func() {
-		if len(c.NodeBlacklist) == 0 {
-			return
+// buildBlacklistSet recomputes the lookup set from pks and returns it.
+// Empty/whitespace-only entries are skipped. Keys are lowercased + trimmed.
+// Returns nil for an empty effective set so callers can `len(m) == 0` short-circuit.
+func buildBlacklistSet(pks []string) map[string]bool {
+	if len(pks) == 0 {
+		return nil
+	}
+	m := make(map[string]bool, len(pks))
+	for _, pk := range pks {
+		trimmed := strings.ToLower(strings.TrimSpace(pk))
+		if trimmed != "" {
+			m[trimmed] = true
 		}
-		m := make(map[string]bool, len(c.NodeBlacklist))
-		for _, pk := range c.NodeBlacklist {
-			trimmed := strings.ToLower(strings.TrimSpace(pk))
-			if trimmed != "" {
-				m[trimmed] = true
-			}
-		}
-		c.blacklistSetCached = m
-	})
-	return c.blacklistSetCached
+	}
+	if len(m) == 0 {
+		return nil
+	}
+	return m
+}
+
+// SetNodeBlacklist atomically replaces NodeBlacklist with pks, rebuilds the
+// lookup set, and bumps the generation counter so any cache keyed on the
+// generation invalidates on the next request (#1629). Safe for concurrent
+// use with IsBlacklisted / BlacklistGeneration.
+func (c *Config) SetNodeBlacklist(pks []string) {
+	if c == nil {
+		return
+	}
+	// Copy so callers can mutate their slice without affecting us.
+	cp := make([]string, len(pks))
+	copy(cp, pks)
+	c.NodeBlacklist = cp
+	m := buildBlacklistSet(cp)
+	c.blacklistSetPtr.Store(&m)
+	c.blacklistGen.Add(1)
+}
+
+// BlacklistGeneration returns a monotonic counter that increments on every
+// SetNodeBlacklist call. Response caches keyed per-pubkey embed this value
+// in their cache key so any blacklist mutation invalidates prior entries on
+// the next request (#1629).
+func (c *Config) BlacklistGeneration() uint64 {
+	if c == nil {
+		return 0
+	}
+	return c.blacklistGen.Load()
 }
 
 // IsBlacklisted returns true if the given public key is in the nodeBlacklist.
+// Hot read path: a single atomic pointer load + map lookup. No locks, no
+// sync.Once. The in-memory set is populated either via SetNodeBlacklist or
+// lazily on first read from c.NodeBlacklist (covering the JSON-load path
+// where the setter was never called).
 func (c *Config) IsBlacklisted(pubkey string) bool {
-	if c == nil || len(c.NodeBlacklist) == 0 {
+	if c == nil {
 		return false
 	}
-	return c.blacklistSet()[strings.ToLower(strings.TrimSpace(pubkey))]
+	mp := c.blacklistSetPtr.Load()
+	if mp == nil {
+		// Lazy first-read materialisation from the JSON-loaded slice.
+		// CAS-style: if another goroutine wins the race, drop ours.
+		built := buildBlacklistSet(c.NodeBlacklist)
+		if c.blacklistSetPtr.CompareAndSwap(nil, &built) {
+			mp = &built
+		} else {
+			mp = c.blacklistSetPtr.Load()
+		}
+	}
+	if mp == nil || len(*mp) == 0 {
+		return false
+	}
+	return (*mp)[strings.ToLower(strings.TrimSpace(pubkey))]
 }
 
 // SaveGeoFilter writes the geo_filter section back to config.json on disk.
diff --git a/cmd/server/node_reach.go b/cmd/server/node_reach.go
index 77ade07a..8b04840c 100644
--- a/cmd/server/node_reach.go
+++ b/cmd/server/node_reach.go
@@ -10,6 +10,7 @@ import (
 	"strconv"
 	"strings"
 	"sync"
+	"sync/atomic"
 	"time"
 
 	"github.com/gorilla/mux"
@@ -283,6 +284,13 @@ type reachState struct {
 	// simultaneous callers run the scan + attribution once, not N times.
 	sf singleflight.Group
 
+	// lastSeenBlacklistGen is the BlacklistGeneration() value that the cache
+	// was last reconciled with. When the live generation moves past this
+	// value, the cache is purged wholesale on the next request to prevent
+	// prior-gen entries from accumulating until their TTL expires (#1629
+	// round-2, adversarial #5).
+	lastSeenBlacklistGen atomic.Uint64
+
 	degreeMu   sync.Mutex
 	degreeSnap *degreeSnapshot
 }
@@ -300,6 +308,33 @@ func (s *Server) reachCacheGet(key string) ([]byte, bool) {
 	return e.raw, true
 }
 
+// reachCacheLen returns the current entry count in the reach response cache.
+// Test helper — exposes the size without leaking the internal mutex/map.
+func (s *Server) reachCacheLen() int {
+	s.reach.cacheMu.RLock()
+	defer s.reach.cacheMu.RUnlock()
+	return len(s.reach.cache)
+}
+
+// reachPurgeIfBlacklistGenChanged drops every cached entry when the live
+// blacklist generation has advanced past the cache's last-seen value. CAS
+// gates the purge so concurrent callers only do the work once per gen bump
+// (#1629 round-2, adversarial #5).
+func (s *Server) reachPurgeIfBlacklistGenChanged(gen uint64) {
+	seen := s.reach.lastSeenBlacklistGen.Load()
+	if gen == seen {
+		return
+	}
+	// CAS gates the actual purge to a single winner on a given gen bump.
+	if !s.reach.lastSeenBlacklistGen.CompareAndSwap(seen, gen) {
+		// Another goroutine already advanced (and purged). Done.
+		return
+	}
+	s.reach.cacheMu.Lock()
+	s.reach.cache = nil
+	s.reach.cacheMu.Unlock()
+}
+
 // isHexPubkey reports whether s is a full 64-char lowercase-hex public key.
 // The handler lowercases input first, so we only accept [0-9a-f].
 func isHexPubkey(s string) bool {
@@ -373,7 +408,21 @@ func (s *Server) handleNodeReach(w http.ResponseWriter, r *http.Request) {
 	}
 	days = clampDays(days)
 
-	cacheKey := pubkey + "|" + strconv.Itoa(days)
+	// cacheKey includes the blacklist generation so any mutation via
+	// SetNodeBlacklist invalidates all prior reach cache entries on the
+	// next request (#1629). Without the generation suffix a node added
+	// to the blacklist post-warm would keep being served the cached
+	// non-blacklisted response until the TTL expires.
+	var gen uint64
+	if s.cfg != nil {
+		gen = s.cfg.BlacklistGeneration()
+	}
+	// Purge prior-gen entries wholesale when the generation advances so a
+	// steady stream of operator blacklist edits cannot leak cache entries
+	// up to the TTL. Cheap: one map reset under the cache mutex, only when
+	// the gen actually moved (#1629 round-2, adversarial #5).
+	s.reachPurgeIfBlacklistGenChanged(gen)
+	cacheKey := pubkey + "|" + strconv.Itoa(days) + "|g" + strconv.FormatUint(gen, 10)
 	if raw, ok := s.reachCacheGet(cacheKey); ok {
 		w.Header().Set("Content-Type", "application/json")
 		w.Write(raw)
diff --git a/cmd/server/node_reach_blacklist_cache_test.go b/cmd/server/node_reach_blacklist_cache_test.go
new file mode 100644
index 00000000..88fef603
--- /dev/null
+++ b/cmd/server/node_reach_blacklist_cache_test.go
@@ -0,0 +1,124 @@
+package main
+
+import (
+	"net/http"
+	"testing"
+)
+
+// TestNodeReach_BlacklistMutationBustsCache reproduces #1629.
+//
+// Scenario:
+//  1. Warm the reach response cache with a non-blacklisted pubkey (200 OK).
+//  2. Operator blacklists that pubkey via SetNodeBlacklist (the legitimate
+//     mutation entry point — config reload, admin call, etc.).
+//  3. The very next /reach request for that pubkey MUST return 404 (the
+//     blacklist response), not the cached 200 payload.
+//
+// Pre-fix the blacklist set is locked in by sync.Once at first read, so
+// IsBlacklisted keeps returning false after the mutation; the cache then
+// re-serves the prior reach body and the assertion fails.
+func TestNodeReach_BlacklistMutationBustsCache(t *testing.T) {
+	resetReachState(t)
+	db, n := newReachIntegrationDB(t, `["AABB","01FA","CCDD"]`)
+	defer db.conn.Close()
+
+	// Start with a non-empty blacklist (some unrelated decoy pubkey) so the
+	// blacklist set is materialised on the first IsBlacklisted call below.
+	// This is the realistic state: a deployment running with a populated
+	// blacklist where the operator later ADDS a new entry.
+	decoy := pk64("dec0")
+	cfg := &Config{NodeBlacklist: []string{decoy}}
+	srv := &Server{store: newTestStoreWithDB(t, db, cfg), db: db, cfg: cfg, perfStats: NewPerfStats()}
+
+	// 1. Warm cache (must 200 and populate cache).
+	rr := serveReach(srv, "/api/nodes/"+n+"/reach?days=30")
+	if rr.Code != http.StatusOK {
+		t.Fatalf("warm-up: status=%d want 200 (body=%s)", rr.Code, rr.Body.String())
+	}
+	if srv.reachCacheLen() == 0 {
+		t.Fatalf("warm-up did not populate reach cache")
+	}
+
+	// 2. Operator adds the target node to the blacklist via the public setter.
+	cfg.SetNodeBlacklist([]string{decoy, n})
+
+	// 3. Next request MUST return 404. With the bug, the sync.Once-cached
+	// empty blacklist set makes IsBlacklisted return false, the response
+	// cache hits, and the prior 200 body is re-served.
+	rr2 := serveReach(srv, "/api/nodes/"+n+"/reach?days=30")
+	if rr2.Code != http.StatusNotFound {
+		t.Fatalf("post-blacklist mutation: status=%d want 404 (cached payload was served — #1629)", rr2.Code)
+	}
+}
+
+// TestConfig_BlacklistGenerationIncrements asserts that every SetNodeBlacklist
+// call bumps the generation counter by exactly 1, regardless of whether the
+// content changed. The /reach cache key embeds this generation, so the
+// monotonic-bump contract is part of the public API of the package
+// (adversarial #4 from round-1 polish).
+func TestConfig_BlacklistGenerationIncrements(t *testing.T) {
+	cfg := &Config{}
+	g0 := cfg.BlacklistGeneration()
+	cfg.SetNodeBlacklist([]string{"aa"})
+	g1 := cfg.BlacklistGeneration()
+	if g1 != g0+1 {
+		t.Fatalf("first SetNodeBlacklist: gen %d -> %d (want +1)", g0, g1)
+	}
+	// Identical content — generation MUST still bump. Callers rely on
+	// "any call invalidates" rather than "content-diff invalidates."
+	cfg.SetNodeBlacklist([]string{"aa"})
+	g2 := cfg.BlacklistGeneration()
+	if g2 != g1+1 {
+		t.Fatalf("second SetNodeBlacklist (same content): gen %d -> %d (want +1)", g1, g2)
+	}
+	// Empty mutation also bumps.
+	cfg.SetNodeBlacklist(nil)
+	g3 := cfg.BlacklistGeneration()
+	if g3 != g2+1 {
+		t.Fatalf("nil SetNodeBlacklist: gen %d -> %d (want +1)", g2, g3)
+	}
+}
+
+// TestNodeReach_BlacklistMutationPurgesCache asserts that a blacklist
+// mutation evicts ALL prior reach cache entries (not just the affected
+// pubkey) on the next /reach request. Per adversarial #5, the previous
+// gen-suffix-only design left every prior cached entry stranded until TTL,
+// growing the cache by N entries per operator edit. The current design
+// purges on generation bump (detected on the next handler invocation) so a
+// steady stream of edits cannot leak entries unboundedly.
+func TestNodeReach_BlacklistMutationPurgesCache(t *testing.T) {
+	resetReachState(t)
+	db, n := newReachIntegrationDB(t, `["AABB","01FA","CCDD"]`)
+	defer db.conn.Close()
+
+	cfg := &Config{}
+	srv := &Server{store: newTestStoreWithDB(t, db, cfg), db: db, cfg: cfg, perfStats: NewPerfStats()}
+
+	// Warm cache with two distinct keys (different days param).
+	for _, days := range []string{"30", "7"} {
+		rr := serveReach(srv, "/api/nodes/"+n+"/reach?days="+days)
+		if rr.Code != http.StatusOK {
+			t.Fatalf("warm-up days=%s: status=%d want 200", days, rr.Code)
+		}
+	}
+	before := srv.reachCacheLen()
+	if before < 2 {
+		t.Fatalf("warm-up populated %d entries, want >=2", before)
+	}
+
+	// Unrelated blacklist mutation. The cached pubkey is not in the
+	// blacklist, but prior entries are now keyed under a stale generation
+	// and would otherwise sit until TTL.
+	cfg.SetNodeBlacklist([]string{pk64("dead")})
+
+	// Next /reach request triggers the purge inside the reach path.
+	rr := serveReach(srv, "/api/nodes/"+n+"/reach?days=30")
+	if rr.Code != http.StatusOK {
+		t.Fatalf("post-mutation request: status=%d want 200", rr.Code)
+	}
+	// After the purge + this single re-populate we expect exactly 1 entry,
+	// not the 2 stale + 1 new = 3 that the leaky design would leave behind.
+	if got := srv.reachCacheLen(); got != 1 {
+		t.Fatalf("post-mutation cache len = %d, want 1 (prior entries leaked — adv #5)", got)
+	}
+}
diff --git a/cmd/server/node_reach_endpoint_test.go b/cmd/server/node_reach_endpoint_test.go
index 866fb886..6094077b 100644
--- a/cmd/server/node_reach_endpoint_test.go
+++ b/cmd/server/node_reach_endpoint_test.go
@@ -5,6 +5,7 @@ import (
 	"encoding/json"
 	"net/http"
 	"net/http/httptest"
+	"strconv"
 	"strings"
 	"testing"
 	"time"
@@ -178,9 +179,11 @@ func TestNodeReach_AttributionAndCacheHit(t *testing.T) {
 		t.Errorf("expected they_hear≥1 for neighbour B, links=%+v", resp.Links)
 	}
 
-	// Cache hit: the key must now be populated and a second request must 200.
-	if _, ok := srv.reachCacheGet(n + "|30"); !ok {
-		t.Fatalf("expected reach response to be cached under %q", n+"|30")
+	// Cache hit: the key (now generation-suffixed, #1629) must be populated
+	// and a second request must 200.
+	wantKey := n + "|30|g" + strconv.FormatUint(srv.cfg.BlacklistGeneration(), 10)
+	if _, ok := srv.reachCacheGet(wantKey); !ok {
+		t.Fatalf("expected reach response to be cached under %q", wantKey)
 	}
 	rr2 := serveReach(srv, "/api/nodes/"+n+"/reach?days=30")
 	if rr2.Code != http.StatusOK || rr2.Body.String() != rr.Body.String() {