diff --git a/phase-ends/CURRENT_PHASE.md b/phase-ends/CURRENT_PHASE.md
index 0fd0ce9..250f2ea 100644
--- a/phase-ends/CURRENT_PHASE.md
+++ b/phase-ends/CURRENT_PHASE.md
@@ -280,3 +280,23 @@ cluster addresses as the class's strongest members.
   global-ra family is a second library-asm class (ra in a global, no stack
   frame — the CRT-entry pattern outside the CRT).
 - Worker B continues on 14 clean tier-1 leaves in partition A.
+
+## Cycle 3 — recorded-negatives exclusion (the queue fix)
+
+- Worker B measured that recorded negatives cluster at the head of every
+  partition: 39 of the first 40 worklist rows and 57/271 tier-1 rows (21%)
+  were attempted-and-recorded addresses. sf3_triage never read the negatives
+  index, so closed negatives returned to the queue head every cycle.
+- Fix (committed a38aa26): sf3_triage reads config/near_match_negatives.tsv
+  (new --negatives input, tracked) and excludes those rows as
+  excluded_recorded_negative=69. Tracked index grown 38→94 by importing the
+  phase's 56 unregistered negatives (workers A/B/C + coordinator absorption).
+  Worklist regenerated to 1,490 rows (tier 1=214, tier 2=1274, tier 3=2);
+  3 new synthetic tests. Partitions re-proved disjoint (497/497/496).
+- Worker B cycle-3: 0 matches / 12 attempts, causes separated (queue defect
+  fixed; class-blocked call-shape rows; two self-inflicted incl. the
+  lui-immediate misread lesson — cookbook material). B's handoff paid off:
+  0x80101838 and 0x80036A9C are registered by worker C's late run.
+- Coordinator absorption this segment: 0x800266A8 (3B, two fresh re-spellings),
+  0x80042D88 (strength-reduce, 2 spellings) recorded as negatives. 310
+  regions / 301 distinct bodies; all gates green; SHA-1 stable.