diff --git a/fktry/lanes/LANE-UI.md b/fktry/lanes/LANE-UI.md
index b39eb31..4f1e122 100644
--- a/fktry/lanes/LANE-UI.md
+++ b/fktry/lanes/LANE-UI.md
@@ -801,3 +801,24 @@ the new copy landing and being buried. Not done unasked; say the word.
 its own arrival; a moov-heist UI when the lock becomes breakable (the chip already names the
 hash the minigame will want); compliance quota progress on the debt gauge itself (show what
 filling the current quota would vent) once DATA settles the quota list.
+
+**ADDENDUM (same round, after the orchestrator's v9.1 and SIM/DATA's next landings).**
+Re-verified against HEAD rather than against my own commit:
+- `npm run check` is now **clean tree-wide**, including under v9.1's `CorrectionUnit.phase`
+  union — my `'stasis'`/`'prescan'` comparisons were already exact members, no change needed.
+- **263 UI tests green, all 18 files.** The only red in the tree is LANE-SIM's own
+  `src/sim/mite.test.ts` (2, their firewall work, uncommitted). Their `reference.ts` also
+  spent ~5 minutes throwing "firewall at -21,9 overlaps belt at -21,10" from its own
+  validator, which took every reference-factory test in the repo (mine and theirs) down
+  with it; it self-resolved, as mid-edit files have every round since round 2.
+- **I had to correct one of my OWN round-3 tests, and the reason is worth recording.**
+  "NEXT IN TRAY … advances it on completion" asserted the sim lands on `commissionQueue[1]`.
+  That was always a claim about SIM's cascade rather than about the fax, and DATA's v9
+  commissions broke it two different ways within twenty minutes: first the queue skipped
+  `first-refinement` (the factory had been shipping since tick 0, so entries behind the
+  head were *already* satisfied and completed in the same tick they became active), then
+  the head became `patio-resurfacing`, which my raw-ore rig cannot pay at all, so nothing
+  advanced. The test now asserts the fax's real contract unconditionally — the peek is the
+  live `commissionQueue[1]` — and only asserts the advance when the head is one this rig
+  can actually pay. The head-of-line blocking I flagged in round 5 is still live and is
+  still a design question, not a UI fix.
diff --git a/fktry/src/ui/live.test.ts b/fktry/src/ui/live.test.ts
index bbfba04..85edd83 100644
--- a/fktry/src/ui/live.test.ts
+++ b/fktry/src/ui/live.test.ts
@@ -208,14 +208,36 @@ describe('the HUD against the reference factory', () => {
 
     // Satisfy the head commission for real, and watch the tray move up.
     shipRawOre(h);
-    for (let t = 0; t < 12000 && h.sim.snapshot().activeCommission === first.activeCommission; t += 100) {
-      h.step(100);
+    for (let t = 0; t < 12000 && h.sim.snapshot().activeCommission === first.activeCommission; t++) {
+      h.step(1);
     }
     const after = h.sim.snapshot();
-    expect(after.activeCommission, 'the queue should advance once the head is paid')
-      .toBe(nextId);
+
+    // Whatever the queue did, THIS is the fax's contract and it holds unconditionally:
+    // the peek is the live queue's second entry, never a guess from data order.
     expect($('.fk-fax-next-name').textContent)
       .toBe(after.commissionQueue![1].replace(/-/g, ' ').toUpperCase());
+
+    // The advance itself is only assertable when the rig can actually pay the head, and
+    // DATA owns what sits there (it changed twice during round 7 alone; round 5 already
+    // flagged that an unpayable head blocks the whole tray). So: if the head wants only
+    // raw ore, it MUST advance — otherwise the queue is head-of-line blocked by design
+    // and there is nothing here for the UI to be wrong about.
+    const headDef = DATA.commissions.find((c) => c.id === first.activeCommission)!;
+    const payableByRig = Object.keys(headDef.wants).every((i) => i === 'mdat-ore');
+    if (payableByRig) {
+      expect(after.activeCommission, 'a head this rig can pay must advance')
+        .not.toBe(first.activeCommission);
+      expect(after.commissionQueue![0]).toBe(after.activeCommission);
+      expect(after.commissionQueue).not.toContain(first.activeCommission);
+      // It moves ALONG the queue, never backwards. Note it may skip `nextId`: the
+      // factory has shipped since tick 0, so entries behind the head can already be
+      // satisfied and complete in the same tick they become active.
+      expect([nextId, ...first.commissionQueue!.slice(2)]).toContain(after.activeCommission);
+    } else {
+      expect(after.activeCommission).toBe(first.activeCommission);
+      expect(after.commissionQueue![0]).toBe(after.activeCommission);
+    }
   });
 
   it('completes a commission and stamps the fax', () => {