diff --git a/recipes/glm-5-3-flash-exl3-3-05bpw-on-2x-cmp-170hx-dflash2-k7-q8-kv-384k/recipe.yaml b/recipes/glm-5-3-flash-exl3-3-05bpw-on-2x-cmp-170hx-dflash2-k7-q8-kv-384k/recipe.yaml
index 1dc36b957..43202e53a 100644
--- a/recipes/glm-5-3-flash-exl3-3-05bpw-on-2x-cmp-170hx-dflash2-k7-q8-kv-384k/recipe.yaml
+++ b/recipes/glm-5-3-flash-exl3-3-05bpw-on-2x-cmp-170hx-dflash2-k7-q8-kv-384k/recipe.yaml
@@ -101,3 +101,4 @@ provenance:
     - 'https://github.com/Flun/glm53-flash-cmp170hx-exl3'
     - 'https://www.reddit.com/r/LocalLLaMA/comments/1x0b1ws/'
   notes: "Submitted 2026-10-08 from lar-scout ingest card t_eb5f504f. Repo @ 5933112fdb858c960dfa1a68f507c3a98a9f1201 (HEAD verified 2026-10-08, 2 commits, MIT license). Launch argv token-verified against scripts/serve.sh + scripts/setup.sh + glm/glm_api.py argparse gates at that sha (profile MEM_ARGS/SPEC_ARGS case blocks, DFlash2-vs-MTP parser guard, Q8-options-require-cq8 guard). HF pins verified live via HF API (target 332ab457 resolves 29 files; drafter bf582e4e matches serve.sh exactly; drafter license cc-by-nc-nd-4.0 confirmed on the model card). Reddit thread hard-blocked (403) from every access lane at ingest time — the ~90/110-120 tok/s figures are carried as submitter-T0 claims per the digest card's provenance, not re-verified; performance section says so. First exllamav3 engine row on the site (schema enum extended by the paired code commit; facet auto-creates per WAVE-3 SGLang precedent). Hardware profile cmp-170hx-x2 pre-existing (id 774); cap-mod PCIe note lives in config.runtime.notes + tags (cap-modded, no-p2p) per the DUFRelic modded-card precedent — same silicon profile, mod documented not profiled."
+max_context_len: 393216