diff --git a/image_analysis/image_preprocessing/ImagePreprocessorGPU.cu b/image_analysis/image_preprocessing/ImagePreprocessorGPU.cu
index 37c772af..1dfb2052 100644
--- a/image_analysis/image_preprocessing/ImagePreprocessorGPU.cu
+++ b/image_analysis/image_preprocessing/ImagePreprocessorGPU.cu
@@ -237,9 +237,14 @@ ImagePreprocessorGPU::ImagePreprocessorGPU(const DiffractionExperiment &experime
       cpu_stats_reg(cpu_stats) {
     // Setup mask. The same for every worker, so it is uploaded once per GPU and shared; keyed on the
     // PixelMask's own vector, which the derived table is a pure function of.
+    // Hoist the accessor and index without the bounds check: this runs once per worker over every
+    // pixel of the detector - 18 million times per engine on a 16 Mpx one, and an engine is built per
+    // worker per pass - and .at() on each of them stops the loop vectorising for a bound the loop
+    // itself already respects.
+    const std::vector<uint32_t> &mask_raw = mask.GetMask();
     std::vector<uint8_t> mask_vec(npixels);
     for (int i = 0; i < npixels; i++)
-        mask_vec[i] = (mask.GetMask().at(i) != 0);
+        mask_vec[i] = (mask_raw[i] != 0);
     gpu_mask = SharedDeviceTable(mask.GetMask().data(), npixels, mask_vec.data(), *stream);
 
     // Setup GPU settings. The current device, not device 0: workers are pinned round-robin across GPUs,
diff --git a/rugnux/Rugnux.cpp b/rugnux/Rugnux.cpp
index c063e16d..6255c59a 100644
--- a/rugnux/Rugnux.cpp
+++ b/rugnux/Rugnux.cpp
@@ -2782,7 +2782,12 @@ ProcessResult Rugnux::RunPipeline(RugnuxObserver *observer, bool write_output, b
             }
         }
 
-        if (result.consensus_cell && write_files && config_.write_merged) {
+        // Not on the geometry pre-pass. Pass 1 exists to choose the space group and post-refine the
+        // geometry; pass 2 remakes these files seconds later at the refined geometry, and that is the
+        // answer anyone reads. Writing them twice costs 0.6 s of a 15 s run - a fifth of the merged
+        // mmCIF on a large crystal - for a file that is superseded before the run ends. The pass-2
+        // quality guard is unaffected: has_merge_statistics is set well above this, at the merge.
+        if (result.consensus_cell && write_files && config_.write_merged && !geometry_prepass) {
             phase("Writing reflections");
         const ErrorModelReport em_report{
                 result.error_model_isa > 0 ? fmt::format("{:.2f}", result.error_model_isa) : "?",
@@ -2799,7 +2804,7 @@ ProcessResult Rugnux::RunPipeline(RugnuxObserver *observer, bool write_output, b
             ScalingResult(indexer->GetIntegrationOutcome()).SaveToFile(config_.output_prefix);
         }
 
-        if (result.consensus_cell && write_files && !config_.model_path.empty()) {
+        if (result.consensus_cell && write_files && !config_.model_path.empty() && !geometry_prepass) {
             phase("Validating against model");
             const auto data_sg = experiment_.GetSpaceGroupNumber();
             // With a reference MTZ the merohedral indexing was already resolved against it (rotation