release-0.3.20 plan: the app side assembled (the picks, the five unions, the rebuilt signer, the box UI tests), the isSynced hold, main's Arc rotr_var order, the fleet's case timing

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

docs/plans/release-0.3.20.md

@@ -325,3 +325,13 @@ c18-1 (RTX 3070 pod, wiped datadir, from 08:51:50Z; the Mac's reboot cut the for
 - The node lane's IBD-end read (ca3-v4-node d49c71d0): the weight-table cache (64 tables, cleared past that) made every 2,000-body IBD batch walk the window again (359,709 tables for 5,335 checkpoints, 968 s); fix on release-0.3.20-node: WEIGHT_TABLE_CACHE = 1,024, oldest-first eviction, never a clear, a unit test; bodies 16,592 s is thread time behind the finality state lock, not CPU; the next fresh join on the pod should read 60 to 70 minutes.
 
 **Main (11:5x UK):** a one-off exception to the 6 October rule, the project lead's word ("you will have to fix it"): exactly one POST to /api/cards on PC 1 from the Intel lane's scratch script (not in the tree), the 5090 and the 9070 XT to 2 identities and the Arc off; the engine restarts the workers on the change. The permanent path is the reliability lane's signed `cards` job kind (per-card enabled, identities, power_pct, through the app's own card path, read back), which replaces any runner option: 0.3.20.
+
+## 29. The app side assembled on release-0.3.20 (11:5x UK)
+
+Picked by hash onto ca1742cd, in order: miner-ui-5 810bf5a1 and b322e9fa; ui-ota 0247b065, c337f768, 10c881dc, ee09ae8b; ember-heat 7c779035, cd1034a2, 0d5fc6d2. A merge of ui-ota was tried first and aborted (14 files, its base is miner-ui-5's lineage, which this tree carries as picks). Five conflicts, each resolved by union: publish-manifest.sh keeps Ember's `activation_passed` form and takes ui-ota's `--ui`/`--no-ui` arms, UI block and the `"${UI:-}"` argument (one python call, `sys.argv[1:13]`); the Settings literal in config.rs and the SettingsState literal in engine.rs carry both heat and `ui_builtin`; app.js's export carries both lanes' names; pre-push.sh runs the ui-ota self-test and the heat gate reader. The Mac's `igneum-ota-sign` was rebuilt under the build lock (the stale binary lacked `sign-ui`, the pre-push self-test read RED). UI tests on the box: heat-region 8, notices 8, tune-line 5, ui-ota 4, update-card 4, view 30, all pass. The app gate (`cargo test --release`) runs on the box; the push follows its green.
+
+**Held in the 0.3.20 verdict (the fleet, 11:4x UK):** isSynced reads false at the tip on dc141409 (7 of 12 reads false with blocks equal to headers and no IBD session; a 17-read sample 13 false; the 0.3.17 control reads true 12 of 12). Mining, acceptance and the exec follower unaffected. Relayed to the node lane: a fix on the node line or a ruling that the kit reads the tip. The publish waits on that word.
+
+**Main (12:0x UK), the Arc B580 for 0.3.20's worker kit:** Intel's OpenCL compiler folds the kernel's `rotr_var` helper `rotate(x, (0u - n) & 31u)` into a left rotate by n, so every variable right-rotate is wrong on Intel. The fix is worker-side only: on an Intel platform the worker rewrites that one helper line to the shift form before clBuildProgram, behind the vendor check; no consensus or pack change; the vectors self-test already refuses the wrong hash. When the Intel lane reports bench d green on the Arc (vectors pass, a rate row), its host.c patch goes into the shipped igneum-worker-opencl for 0.3.20 (Windows and Linux) with a unit test that the Intel path builds the shift form and the AMD path is unchanged; the Intel vendor badge and card row from intel-arc ride the same cut. Asked of the Intel lane: the patch as one commit, the test and its command, the badge and card-row hashes. The 0.3.20 worker build holds on that; nothing else in the cut waits on it.
+
+**The fleet (12:0x UK):** the 0.3.20 cases (relay and poison against c18-1 on dc141409) start at about 11:20Z after a re-rent (relay on a 3090, poison on an A5000); the ten-member pool window keys on CASES END; prover roll 10 of 13 paired.