vyncint · vyncint · Sep 22, 2026 · Sep 22, 2026
@@ -90,7 +90,7 @@ concurrency:
 
 env:
   PINNED_TOOLCHAIN: nightly-2026-08-28
-  RECONVERGE_VERSION: "0.6.0"
+  RECONVERGE_VERSION: "0.7.0"
 
 jobs:
   # Prune and compile on a plain runner, exactly as `stage` is designed to be
@@ -112,7 +112,7 @@ jobs:
         with:
           persist-credentials: false
           repository: NVlabs/cuda-oxide
-          ref: 26754ae52c26c097dc1c465a1e42c4c5d05a3d40
+          ref: b0f961df3af0ff140b3b006fa2b6750b71f43f62
           path: cuda-oxide
       - name: Install pinned toolchain
         working-directory: launchbound

@@ -26,9 +26,9 @@ concurrency:
 # The lockstep pin set. A pin bump updates these values (and
 # rust-toolchain.toml) in its own commit.
 env:
-  CUDA_OXIDE_PIN: 26754ae52c26c097dc1c465a1e42c4c5d05a3d40
+  CUDA_OXIDE_PIN: b0f961df3af0ff140b3b006fa2b6750b71f43f62
   CUDA_OXIDE_REPO: NVlabs/cuda-oxide
-  RECONVERGE_PIN: v0.6.0
+  RECONVERGE_PIN: v0.7.0
   RECONVERGE_REPO: vyncint/reconverge
 
 jobs:

@@ -23,7 +23,7 @@ concurrency:
 # pins.yml issue body lists every site.
 env:
   PINNED_TOOLCHAIN: nightly-2026-08-28
-  RECONVERGE_VERSION: "0.6.0"
+  RECONVERGE_VERSION: "0.7.0"
 
 jobs:
   gate:
@@ -40,7 +40,7 @@ jobs:
         with:
           persist-credentials: false
           repository: NVlabs/cuda-oxide
-          ref: 26754ae52c26c097dc1c465a1e42c4c5d05a3d40
+          ref: b0f961df3af0ff140b3b006fa2b6750b71f43f62
           path: cuda-oxide
       - name: Install pinned toolchain
         working-directory: launchbound

@@ -9,6 +9,26 @@ change measured timings are marked `bench:`.
 
 ## [Unreleased]
 
+### Changed
+
+- **The lockstep pin set moves: cuda-oxide `b0f961df`, reconverge `0.7.0`,
+  nightly unchanged** (#70). Its own commit across every site
+  `scripts/check-pins.sh` lists, because the policy is "every site or not at
+  all" and 2.0.0 did four of six.
+
+  reconverge 0.7.0 is the release that matters here rather than a routine
+  bump. Before it, a cluster- or grid-wide barrier under a guard that is only
+  *block*-uniform produced no findings at all — so the gate this product is
+  built on could not see that class, and 2.2.1 shipped with a reconverge that
+  could not.
+
+  Verified at the new set rather than assumed: `just gate` passes all three
+  gate tests — the most important in this repository — and `just prune` over
+  the whole corpus returns **the same verdicts**, including `reduce-flip`'s
+  3 clean / 8 refused, which is the known-flip split the README quotes. The
+  new analyzer changes what it can see and does not change what this corpus
+  is told.
+
 ### Added
 
 - **A dispatchable workflow that re-runs the CUDA path.** Nothing automated

@@ -190,7 +190,7 @@ The `pins.yml` workflow reports upstream movement by opening an issue; it
 never bumps anything. It runs every Monday, and reports a toolchain move as
 its own line — a channel change means reconverge must be rebuilt, which
 commit churn does not. Current pins: nightly-2026-08-28,
-cuda-oxide 26754ae5, reconverge 0.6.0 (installed from crates.io).
+cuda-oxide b0f961df, reconverge 0.7.0 (installed from crates.io).
 
 ## 10. License
 

@@ -29,7 +29,7 @@ jobs:
       - uses: actions/checkout@v4
         with:
           repository: NVlabs/cuda-oxide
-          ref: 26754ae52c26c097dc1c465a1e42c4c5d05a3d40
+          ref: b0f961df3af0ff140b3b006fa2b6750b71f43f62
           path: cuda-oxide
 
       - name: Safety gate
@@ -63,7 +63,7 @@ the gate specializes per candidate.
 | `cc` | — | target compute capability, e.g. `"8.6"` (required; verdicts do not transfer across parts) |
 | `fail-on` | `tool-error` | `never`, `refused`, or `tool-error` |
 | `version` | `latest` | launchbound-cli release to install |
-| `reconverge-version` | `0.6.0` | reconverge release from crates.io — moves in lockstep with `toolchain` |
+| `reconverge-version` | `0.7.0` | reconverge release from crates.io — moves in lockstep with `toolchain` |
 | `toolchain` | `nightly-2026-08-28` | the nightly that built that reconverge |
 | `summary` | `"true"` | write the verdict table to the job summary |
 

@@ -51,7 +51,7 @@ inputs:
       reconverge-driver). Must be built by the toolchain below — the pins
       move together.
     required: false
-    default: "0.6.0"
+    default: "0.7.0"
   toolchain:
     description: >-
       The pinned nightly that matches reconverge-version; the pair moves

@@ -34,5 +34,5 @@ Every kernel is a **standalone, device-only lib crate**:
 
 ```bash
 git clone https://github.com/NVlabs/cuda-oxide ../cuda-oxide   # sibling of this repo
-git -C ../cuda-oxide checkout 26754ae5                          # the pinned commit
+git -C ../cuda-oxide checkout b0f961df                          # the pinned commit
 ```
@@ -173,7 +173,7 @@ fn is_missing_subcommand(stderr: &str) -> bool {
 /// asserts this constant against `rust-toolchain.toml` and the workflows,
 /// because drift between recorded pins is the failure this repository keeps
 /// hitting.
-pub const CUDA_OXIDE_PIN: &str = "26754ae52c26c097dc1c465a1e42c4c5d05a3d40";
+pub const CUDA_OXIDE_PIN: &str = "b0f961df3af0ff140b3b006fa2b6750b71f43f62";
 
 /// What to say when `cargo oxide` is not installed.
 fn missing_oxide_message() -> String {

@@ -8,7 +8,7 @@ set, and the corpus decided identically ([research-baseline](research-baseline.m
 
 ## The gate inherits reconverge's limits, wholesale
 
-A clean gate is **not a proof of correctness**. `reconverge` (v0.6.0) is
+A clean gate is **not a proof of correctness**. `reconverge` (v0.7.0) is
 summary-based and interprocedural, handles reducible control flow only,
 cannot evaluate non-literal masks, and puts data races entirely out of
 scope. Its own documentation is the authority; launchbound adds no analysis

@@ -15,7 +15,7 @@ Measured 2026-08-20.
 | tier-1 guest | Apple `container` 1.2.0, **native arm64** Ubuntu 24.04 (no Docker, no Rosetta — operator requirement), CUDA toolkit 13.2 (sbsa), LLVM 21.1.8, container `cuda-oxide-dev` |
 | tier-2 box | AWS `g5.xlarge` spot @ **$0.364/hr**, us-east-2c, **NVIDIA A10G** (`sm_86`, cc 8.6), driver 595.71.05, CUDA 13.2 (V13.2.51), LLVM 21.1.8 — chosen over the T4 by the operator; barely above T4 spot ($0.335/hr) |
 | pinned toolchain | `nightly-2026-04-03` (`rustc 1.96.0-nightly (55e86c996 2026-04-02)`) |
-| reconverge | `cargo-reconverge 0.1.11` (built at `~/Projects/reconverge/target/release`) — **the version these measurements were taken with**. The gate has since moved to 0.3.0, 0.4.0, 0.5.0 and (in 2.2.0) 0.6.0; each move was verified to admit the identical set — see the two comparison sections below. This row records the conditions of *these* measurements and is deliberately not updated. |
+| reconverge | `cargo-reconverge 0.1.11` (built at `~/Projects/reconverge/target/release`) — **the version these measurements were taken with**. The gate has since moved to 0.3.0, 0.4.0, 0.5.0, (in 2.2.0) 0.6.0 and (in 2.3.0) 0.7.0; each move was verified to admit the identical set — see the two comparison sections below. This row records the conditions of *these* measurements and is deliberately not updated. |
 | cuda-oxide | checkout `50d07314eb8b7d5ec821ba02b0048a753c20dd4e` — the tree synced to the box (the box AMI's own stale clone reports `e28248c1`, but `./gpu sync` replaces the working tree and excludes `.git`, so the synced tree is what compiled) |
 | subject kernels | `s0-reduce` (device-only lib crate, dep `cuda-device` only, containing the README's known-flip reduction); cuda-oxide examples `vecadd` (small) and `tiled_gemm` (large) |
 | evidence logs | `~/Projects/cuda-oxide/.gpu-evidence/20260820T{071248,071807,071959}Z.log` |

@@ -5,8 +5,8 @@
 #
 # Current lockstep set:
 #   nightly     nightly-2026-08-28
-#   cuda-oxide  26754ae52c26c097dc1c465a1e42c4c5d05a3d40
-#   reconverge  v0.6.0
+#   cuda-oxide  b0f961df3af0ff140b3b006fa2b6750b71f43f62
+#   reconverge  v0.7.0
 [toolchain]
 channel = "nightly-2026-08-28"
 components = ["rustfmt", "clippy"]