[skills] fix: preserve secrets baseline anchors

Signed-off-by: yaoyu-33 <yaoyu.094@gmail.com>

Author: yaoyu-33

skills/nemo-mbridge-perf-activation-recompute/card.yaml

@@ -1,6 +1,5 @@
 title: activation_recompute
 validated_on: "2026-08-12"
-guidance_updated_on: "2026-08-25"
 summary: >
   Selective activation recompute trades replayed forward work for lower retained
   activation memory. The correct checkpoint boundary is architecture- and
@@ -17,8 +16,6 @@ validation_status:
     - code_verified  # transformer_config.py and transformer_block.py
   recompute_modules_enum:
     - code_verified  # transformer_config.py
-  qwen35_gdn_boundary:
-    - code_verified  # pinned MCore gdn_norm_out boundary and current Bridge recipe use
   llama3_70b_sft_fp8_cs_experiment:
     - measured  # historical PR #3107; contextual, not a cross-module ranking
   moonlight_16b_bf16_experiment:
@@ -121,7 +118,6 @@ recommended_path:
   standard_attention: test core_attn, but compare it with an empty list under fused or Flash Attention
   mla: test mla_up_proj first when expanded Q/K/V tensors dominate; add core_attn only with evidence
   grouped_moe: test moe_act and then layernorm when their outputs drive the peak
-  qwen35_gdn: test gdn_norm_out on the exact revision; core_attn alone does not cover the GDN normalization output
   dense_ffn: test mlp only when its broader replay cost is acceptable
   broader_moe: reserve whole moe for cases where narrower MoE boundaries do not fit
   last_resort: use full recompute with recompute_method and recompute_num_layers
@@ -339,3 +335,10 @@ follow_up_validation:
   - Add matched GDN measurements.
   - Measure context-parallel communication replay for attention and GDN boundaries.
   - Validate representative selective boundaries under the supported CUDA-graph scopes and FP8 recipes.
+
+guidance_updated_on: "2026-08-25"
+operational_guidance:
+  qwen35_gdn: >
+    Test gdn_norm_out on the exact revision; core_attn alone does not cover the
+    GDN normalization output. Use full-layer recompute when targeted boundaries
+    still do not make the complete optimizer step fit.