diff --git a/docs/equipment/inline-optical-qc.md b/docs/equipment/inline-optical-qc.md
index a9c1791..4a9effe 100644
--- a/docs/equipment/inline-optical-qc.md
+++ b/docs/equipment/inline-optical-qc.md
@@ -424,8 +424,42 @@ October 2026).
   measured in hours. The racks hold about one day of output: 2,800 to
   3,100 records at the launch plant (the 850k target to the ~0.94M
   capacity in the [calculator](../fleet-calculator.html), over 300 days).
+- **Stacks and cooling plates** (decided October 2026).
+  - **50 records a stack,** about 15 to 25 minutes of one press. Cascade
+    Record Pressing pulls the spindle at 50
+    ([VRTX](https://vrtxmag.com/articles/how-to-press-vinyl-locally)).
+    Fifty 180 g records weigh 9 kg; a day of output is about 58 stacks.
+  - **A flat aluminium cooling plate every 10 records** to start with, as
+    at MY45 ([oberflaeche.de](https://oberflaeche.de/en/knowledge/topics/electro-plating/a-vinyl-renaissance/the-great-comeback-of-the-record-3));
+    Mobineko lays one under every record
+    ([Mobineko](https://mobineko.com/vinyl-quality-standards)). The
+    Classic places the plates itself ("automatic cooling plate handling",
+    [Newbilt](https://vinyl-pressing-plants.com/blog/newbilt-classic-12-inch-vinyl-record-press));
+    at the Duplex the operator lays one by hand. Full flat discs, not ring
+    spacers: a warm record held only round the label sags into a dish.
+  - **Plates cut locally.** 5052 aluminium (5083 if they are anodised),
+    3 mm thick, with the diameter and centre hole of Newbilt's plate so
+    the Classic's magazine takes them; thickness and alloy are our
+    estimate until Newbilt's drawing arrives. A 1220 × 2440 mm sheet nests
+    28 discs of about 305 mm, and a 3 mm sheet costs about THB 8,300
+    before VAT
+    ([OneStockHome](https://www.onestockhome.com/th/product_categories/aluminum-sheet-thickness-3-mm),
+    grade 3003, October 2026): about THB 300 of metal a plate, THB 350 to
+    450 with cutting and deburring (estimate). Cut on a fibre laser or a
+    waterjet. Chiang Mai has several laser shops (SK Laser Cut and Laser
+    CNX on Rattanakosin Road, among others), but many cut only acrylic and
+    wood, so ask for 3 mm aluminium. Edges and the centre hole are
+    deburred and rounded, because a burr marks the record. Every plate is
+    checked on a surface plate (no gap over 0.1 mm, our target).
+  - **How many:** one plate per 10 records over the full 24-hour queue
+    needs about 350 plates (13 sheets, THB 120,000 to 160,000, estimate);
+    an 8-hour queue about 100 to 130. Two sheets come first, for the
+    plate trial (next steps).
+  - **Clean plates:** dust on a plate ends up on the record, so the plates
+    are wiped on their way back to the presses and kept covered.
 - **At the station.** The finishing operator takes each record off the
-  cooled stack and puts it on the infeed. Good records leave the outfeed
+  cooled stack and puts it on the infeed, and lifts each cooling plate
+  onto the plate cart that goes back to the presses. Good records leave the outfeed
   into sleeving; a reject goes into the bin, and the HMI shows the reason.
   Finishing handles every record anyway, so the station adds no handling
   step.
@@ -459,6 +493,24 @@ October 2026).
   minute). Past that, add a second station or move to stations at the
   presses.
 
+**Considered, not taken: a spiral cooling tower.** Holding the whole
+queue on a spiral is out of proportion. Eight hours of output lying
+single is about 2,900 records, some 300 m of 1 m belt three records
+across, roughly USD 75k to 120k at USD 250 to 300 per metre for Chinese
+spiral coolers
+([listing](https://www.sourcingai.com/smart-products/acebakebreadmachine/flexible-screw-cooling-spiral-tower-conveyor-tower-with-advanced-cooling-technology-for-bakery-BUWkKSqORktM);
+our arithmetic), for a job the racks do for a few thousand. Bakery
+spirals are built for 30 to 60 minutes
+([I.J. White](https://www.ijwhite.com/blog/ultra-series-spiral-cooler-at-klosterman-baking/)).
+Open on a belt for hours, fresh records also collect dust on both faces,
+and a record lying alone has no stack weight holding it flat. A short
+spiral of 30 to 60 minutes (20 to 40 m of belt, about USD 5k to 16k
+ex-works, estimate) stays a later option: it would hand cooled records
+to the station one by one and give each its exact pressing time, but
+warp would then be read after an hour instead of on settled records.
+The plate trial checks whether the early reading predicts the settled
+one.
+
 **Later option: a station at each press.** It finds a defect streak
 within a cycle or two, at the price of one station per cell, and it reads
 warm records, so release on warp still needs cooled samples.
@@ -546,19 +598,26 @@ vision share is not yet confirmed.
 2. **Keyence test.** Ask Keyence Thailand for a test report on the same
    records.
 3. **Newbilt.** Ask how the Classic hands over its spindle stacks and
-   cooling plates (records per stack). Also ask for the cycle counter over
+   cooling plates (records per stack, records per plate), and for the
+   drawing of its plate (diameter, thickness, alloy, centre hole) so the
+   locally cut plates fit its magazine. Also ask for the cycle counter over
    OPC UA, so every stack label carries the record key (stamper, press,
    cycle, side; see [digital factory](../digital-factory.md)).
 4. **Warp sensors.** Choose between two point lasers and a line profiler
    once the trials show whether a thickness map is worth the extra cost.
-5. **EtherCAT bench.** Before the frame is built, run one A6-EC kit, the
+5. **Plate and settling trial.** Cut two sheets of plates (56). Press one
+   job with a plate every 5, 10 and 25 records, and read edge run-out and
+   dish on the station after 1 hour and after 24 hours. That sets the
+   records per plate (black, colour and specials apart) and shows whether
+   a 1-hour reading predicts the settled one.
+6. **EtherCAT bench.** Before the frame is built, run one A6-EC kit, the
    EK1101 with its terminals and the spindle encoder from the mini PC
    with qitech_lib on EtherCrab: eight tests and pass marks are in the
    [control plan](inline-qc-control-plan.md#7-bench-test-before-the-frame-is-built)
    (about THB 56,000 of parts, all reused in the station). If the
    EtherCrab stack falls short on the drive, repeat the test with the
    IgH master.
-6. **Acceptance test.** Write it before ordering. It should cover the
+7. **Acceptance test.** Write it before ordering. It should cover the
    detection rate per defect class on the trial set, the false-reject
    rate on good records (black and specials apart), and every result
    matched to its record key.