diff --git a/.jules/sentinel.md b/.jules/sentinel.md
index 8b7813b..75b4a40 100644
--- a/.jules/sentinel.md
+++ b/.jules/sentinel.md
@@ -9,3 +9,11 @@
 5. DoS 완화를 위해 `return(1L)` 같은 기본 승인값을 넣을 때는 추정 기준척도, anchor/common item, true parameter 재현 계약을 우회하지 않는지 먼저 검증합니다.
 6. Fail-secure 에러 메시지는 테스트의 일부로 취급합니다. 보안 테스트는 실제 구현 메시지와 맞아야 하며, 오래된 `"Interactive prompt is not available"` 같은 별도 문구를 새로 만들지 않습니다.
 7. Prompt DoS 회귀 테스트는 모델 추정 실패에 기대지 말고, common-item confirmation guard처럼 취약한 입력 경계에서 바로 발생하는 fail-secure 에러를 검증합니다.
+
+## 2024-06-25 - Unbounded Loop in Model Retry (Infinite Loop DoS)
+**Vulnerability:** When the initial `mirt::mirt()` model estimation failed, the code utilized an unconstrained `while (!exists('model'))` loop to continually attempt re-estimation. Since R's deterministic errors inside `try()` would repetitively fail without side-effects altering the outcome, this created an infinite loop Denial of Service (DoS) in automated environments.
+**Learning:** Deterministic failure handling must never rely on unbounded loops. Relying on `try()` combined with `while (!exists(...))` assumes transient errors, which is often not true for statistical model convergence issues.
+**Prevention:**
+1. Always replace `while (!exists(...))` retries with a bounded `for` loop (e.g., `for (attempt in seq_len(3))`).
+2. Include an explicit check for the success condition inside the loop (`if (exists('model')) break`).
+3. After the loop, verify success and fail securely with an explicit error (`if (!exists('model')) stop(...)`) to prevent unhandled exceptions downstream.
diff --git a/R/aFIPC.R b/R/aFIPC.R
index d0329f2..8ff9257 100644
--- a/R/aFIPC.R
+++ b/R/aFIPC.R
@@ -215,8 +215,8 @@ autoFIPC <-
             'Estimation failed. estimating new parameters with no prior distribution using  Cai\'s (2010) Metropolis-Hastings Robbins-Monro (MHRM) algorithm. please be patient.'
           )
 
-          try(rm(oldFormModel))
-          while (!exists('oldFormModel')) {
+          try(rm(oldFormModel), silent = TRUE)
+          for (attempt in seq_len(3)) {
             try(
               oldFormModel <-
                 mirt::mirt(
@@ -230,7 +230,9 @@ autoFIPC <-
                   GenRandomPars = F
                 )
             )
+            if (exists('oldFormModel')) break
           }
+          if (!exists('oldFormModel')) stop('Failed to estimate oldFormModel with MHRM after 3 attempts')
         }
       }
 
@@ -427,8 +429,8 @@ autoFIPC <-
             'Estimation failed. estimating new parameters with no prior distribution using  Cai\'s (2010) Metropolis-Hastings Robbins-Monro (MHRM) algorithm. please be patient.'
           )
 
-          try(rm(newFormModel))
-          while (!exists('newFormModel')) {
+          try(rm(newFormModel), silent = TRUE)
+          for (attempt in seq_len(3)) {
             try(
               newFormModel <-
                 mirt::mirt(
@@ -442,7 +444,9 @@ autoFIPC <-
                   GenRandomPars = F
                 )
             )
+            if (exists('newFormModel')) break
           }
+          if (!exists('newFormModel')) stop('Failed to estimate newFormModel with MHRM after 3 attempts')
         }
       }
 
@@ -975,10 +979,12 @@ autoFIPC <-
     # if(!LinkedModel@OptimInfo$secondordertest){
     #   message('Estimation failed. estimating new parameters with no prior distribution using  Cai\'s (2010) Metropolis-Hastings Robbins-Monro (MHRM) algorithm. please be patient.')
     #
-    #   rm(LinkedModel)
-    #   while (!exists('LinkedModel')) {
+    #   try(rm(LinkedModel), silent = TRUE)
+    #   for (attempt in seq_len(3)) {
     #     try(LinkedModel <- mirt::mirt(data = newformXDataK[colnames(newFormModel@Data$data)], LinkedModelSyntax, itemtype = newFormModel@Model$itemtype, SE = T, method = 'MHRM', accelerate = 'squarem', technical = list(NCYCLES = 1e+5, MHRM_SE_draws = 200000), pars = NewScaleParms, GenRandomPars = T))
+    #     if (exists('LinkedModel')) break
     #   }
+    #   if (!exists('LinkedModel')) stop('Failed to estimate LinkedModel with MHRM after 3 attempts')
     # }
 
     # if(!LinkedModel@OptimInfo$secondordertest){