bump-hrx: hold the mirror for the release — the orchestrator's decision on #1945 (Fixes #1945)

config/llama-mirror-hold.config

@@ -23,11 +23,12 @@
 # is out of date (the mirror is no longer the named commit) or no longer needed (AMD's pin contains the
 # named commit again), so a hold cannot outlive the state it describes.
 #
-# No commit is named below, so nothing is held. Naming one is the owner's decision. The state it would
-# record, as of 2026-11-11 (engine#1945):
+# HELD 2026-11-11 for the release (orchestrator decision on engine#1945, the owner having delegated technical
+# calls): the engine's llama.cpp pin stays where it is, and the merge of AMD's rebased line is post-release work.
+# The state this records:
 #   mirror  1bit/hrx-vulkan           44ee5710d19f7f3af79b99bbcbe5fd13979d1f67   (AMD's line before its rebase)
 #   AMD pin                           9bf5871310a8d7de74f4cb0c57009835fbf91c27   (the rebased line, 1777 upstream commits newer)
 #   merge into 1bit/hrx-vulkan-patched, against the last AMD commit our line took (43b0ec2c41cc):
 #                                     149 conflicted files, 538 hunks (65 files / 83 hunks in the HRX backend)
-# To hold, put the mirror's commit on a line of its own:
-#   44ee5710d19f7f3af79b99bbcbe5fd13979d1f67
+# Lift it by deleting the line below, once AMD's line has been merged into ours.
+44ee5710d19f7f3af79b99bbcbe5fd13979d1f67

docs/hrx.md

@@ -102,12 +102,13 @@ fails a pin that leaves a gate model's node off HRX.
   nothing to do but fail, every day, until somebody does the merge by hand. While
   [`config/llama-mirror-hold.config`](../config/llama-mirror-hold.config) names the mirror's commit, the run
   instead reports the hold in its summary (`mirror 1bit/hrx-vulkan HELD at …; AMD's pin is …, on a line that
-  does not contain it`), moves nothing and succeeds; the engine's pin still follows our own line. The hold
+  does not contain it`), moves nothing and succeeds. The engine's pin does not move in a held run either: moving
+  it along our own line is a reviewed PR of its own. The hold
   describes one state and fails the run when that state ends, so it cannot be forgotten: **stale** when the
   mirror is no longer the named commit, **unneeded** when AMD's pin contains the named commit again (an ordinary
   sync would work). `scripts/bump-hrx-sync.sh llama-hold <repo> <mirror> <pin>` prints the decision alone
-  (`none`, `held`, `stale`, `unneeded`). The file ships naming no commit, so it holds nothing; naming one is
-  the owner's decision. It was added for AMD's 2026-11-09 rebase (engine#1945): taking that pin is a merge of
+  (`none`, `held`, `stale`, `unneeded`). **Held since 2026-11-11** for the release (engine#1945): the file names
+  the mirror's commit `44ee5710d19f`, and the merge of AMD's rebased line is post-release work. It was added for AMD's 2026-11-09 rebase (engine#1945): taking that pin is a merge of
   149 conflicted files (538 hunks) into `1bit/hrx-vulkan-patched`, measured against the last AMD commit our
   line took.
 - **The hold.** While [`config/hrx-system-hold.config`](../config/hrx-system-hold.config)

scripts/bump-hrx-sync.sh

@@ -169,10 +169,12 @@ sync_llama() {
     held)
       note "1bit/hrx-vulkan is held at ${mirror:0:12} by $LLAMA_HOLD_FILE: AMD's pin ${pin:0:12} does not descend from it (AMD rewrote its line). Nothing was moved; the merge is tracked where the hold file says."
       summary "- llama.cpp: mirror \`1bit/hrx-vulkan\` HELD at \`${mirror:0:12}\` ($LLAMA_HOLD_FILE); AMD's pin is \`${pin:0:12}\`, on a line that does not contain it. Not merged."
-      output "rebased=$patched"
+      # The engine's pin does not move while the mirror is held: `rebased` is what the workflow would pin, and
+      # naming the engine's own pin makes its last step say "nothing to move" (our line may be ahead of the pin;
+      # moving the pin along it is a reviewed PR of its own, not something a held run does).
+      output "rebased=${OURS_LLAMA:-$patched}"
       output "llama_held=1bit/hrx-vulkan HELD at \`${mirror:0:12}\` by \`$LLAMA_HOLD_FILE\` (AMD's pin is \`${pin:0:12}\`)."
       output "patches<<EOF
-$(git log --format='- `%h` %s' "${OURS_LLAMA:-$patched}..$patched" 2>/dev/null || true)
 EOF"
       return 0 ;;
   esac

tests/bump_hrx_sync_test.sh

@@ -124,11 +124,14 @@ expect "the commit line may carry a trailing note; a line that is not a commit i
 printf '%s\n' "${mirror:0:12}" > "$holdf"
 expect "a short hash is not a hold: forty hex digits or nothing" "$(hold "$mirror" "$pin")" "none"
 
-# The file shipped in the tree names no commit, so merging it holds nothing.
+# The file shipped in the tree: whatever it names must be one full commit hash or nothing, and what it means for a
+# run is decided by the same function as above (a hold on a commit the mirror has left is stale, and so on).
 shipped=$(cd "$(dirname "$script")/.." && pwd)/config/llama-mirror-hold.config
 if [ -f "$shipped" ]; then
-  expect "the shipped hold file holds nothing until a commit line is added" \
-    "$(LLAMA_HOLD_FILE=$shipped bash "$script" llama-hold "$repo" "$mirror" "$pin")" "none"
+  n=$(grep -v '^[[:space:]]*#' "$shipped" | grep -cE '^[0-9a-f]{40}' || true)
+  if [ "$n" -le 1 ]; then ok "the shipped hold file names at most one commit"; else fail "the shipped hold file names at most one commit (found $n)"; fi
+  junk=$(grep -v '^[[:space:]]*#' "$shipped" | grep -v '^[[:space:]]*$' | grep -vcE '^[0-9a-f]{40}([[:space:]].*)?$' || true)
+  expect "every line of the shipped hold file is a comment, blank, or a full commit hash" "$junk" "0"
 else
   fail "config/llama-mirror-hold.config is in the tree"
 fi