From 55de02f5e392b759e730daf75e71badc386ecabf Mon Sep 17 00:00:00 2001
From: Julian Freeman <freemanjulian372@gmail.com>
Date: Wed, 25 Feb 2026 11:38:18 -0400
Subject: [PATCH] improve ui and temp

---
 src/App.vue            | 14 ++++++------
 src/stores/settings.ts | 51 +++++++++++++++++++++++++-----------------
 2 files changed, 37 insertions(+), 28 deletions(-)

diff --git a/src/App.vue b/src/App.vue
index e496934..486ea4e 100644
--- a/src/App.vue
+++ b/src/App.vue
@@ -324,11 +324,6 @@ const translate = async () => {
       targetText.value = response;
     }
     settings.addLog('response', 'Translation completed');
-
-    // Trigger evaluation if enabled
-    if (settings.enableEvaluation) {
-      await evaluateTranslation();
-    }
   } catch (err: any) {
     const errorMsg = String(err);
     settings.addLog('error', errorMsg);
@@ -336,6 +331,11 @@ const translate = async () => {
   } finally {
     isTranslating.value = false;
   }
+
+  // Trigger evaluation if enabled
+  if (settings.enableEvaluation) {
+    await evaluateTranslation();
+  }
 };
 </script>
 
@@ -458,7 +458,7 @@ const translate = async () => {
           
                     <div class="p-4 border-t dark:border-slate-800 bg-slate-50/30 dark:bg-transparent flex justify-end shrink-0">            <button 
               @click="translate"
-              :disabled="isTranslating || !sourceText.trim()"
+              :disabled="isTranslating || isEvaluating || !sourceText.trim()"
               class="bg-blue-600 hover:bg-blue-700 disabled:bg-blue-300 dark:disabled:bg-blue-900/40 text-white px-6 py-2.5 rounded-lg font-medium transition-all flex items-center gap-2 shadow-sm"
             >
               <Loader2 v-if="isTranslating" class="w-4 h-4 animate-spin" />
@@ -658,7 +658,7 @@ const translate = async () => {
             <button 
               @click="evaluateTranslation"
               :disabled="isEvaluating || isTranslating || !targetText.trim()"
-              class="bg-slate-200 hover:bg-slate-300 dark:bg-slate-800 dark:hover:bg-slate-700 disabled:opacity-50 text-slate-700 dark:text-slate-200 px-6 py-2.5 rounded-lg font-medium transition-all flex items-center gap-2 shadow-sm border border-slate-300/50 dark:border-slate-700/50"
+              class="bg-slate-200 enabled:hover:bg-slate-300 dark:bg-slate-800 dark:enabled:hover:bg-slate-700 disabled:opacity-50 text-slate-700 dark:text-slate-200 px-6 py-2.5 rounded-lg font-medium transition-all flex items-center gap-2 shadow-sm border border-slate-300/50 dark:border-slate-700/50"
             >
               <Loader2 v-if="isEvaluating" class="w-4 h-4 animate-spin" />
               <Check v-else class="w-4 h-4" />
diff --git a/src/stores/settings.ts b/src/stores/settings.ts
index c1587ff..176c2ab 100644
--- a/src/stores/settings.ts
+++ b/src/stores/settings.ts
@@ -51,31 +51,40 @@ export const DEFAULT_TEMPLATE = `You are a professional {SOURCE_LANG} ({SOURCE_C
 3. Produce ONLY the {TARGET_LANG} translation, without any additional explanations, notes, or commentary.
 4. If [Context] is provided, use it strictly to disambiguate polysemous words. DO NOT add any factual information or descriptive details from the [Context] that are not present in the [Text to Translate].`;
 
-export const DEFAULT_EVALUATION_TEMPLATE = `You are an expert translation auditor proficient in {SOURCE_LANG} and {TARGET_LANG}. 
-Your task is to critically evaluate the accuracy and quality of a translation.
+export const DEFAULT_EVALUATION_TEMPLATE = `# Role
+You are an **Objective Translation Auditor**. Your task is to evaluate translation quality based on accuracy, grammar, and fundamental readability. Avoid pedantic nitpicking over synonyms, but do point out issues that hinder professional quality.
 
-[Context Info]
-- Source Language: {SOURCE_LANG}
-- Target Language: {TARGET_LANG}
-- Speaker Identity: {SPEAKER_IDENTITY}
-- Intended Tone/Register: {TONE_REGISTER}
-- Context: {CONTEXT}
+# Context Info
+- **Source Language**: {SOURCE_LANG}
+- **Target Language**: {TARGET_LANG}
+- **Speaker Identity**: {SPEAKER_IDENTITY}
+- **Intended Tone/Register**: {TONE_REGISTER}
+- **Context**: {CONTEXT}
 
-[Input]
-- Source Text: {SOURCE_TEXT}
-- Translated Text: {TRANSLATED_TEXT}
+# Input
+- **Source Text**: {SOURCE_TEXT}
+- **Translated Text**: {TRANSLATED_TEXT}
 
-[Instructions]
-1. Compare the [Source Text] and [Translated Text] meticulously.
-2. Check if the translation respects the [Context Info].
-3. Assign an "Accuracy Score" from 0 to 100. 
-   - Give 0 if there are fatal semantic errors, complete hallucinations, or if the meaning is reversed.
-   - Deduct points for minor inaccuracies, unnatural phrasing, or tone mismatches.
-4. Provide a concise "Analysis" of why you gave that score.
-5. (Optional) Provide "Improvements" for a more accurate/natural translation.
+# Audit Criteria
+Only penalize and provide improvements if the translation meets one of these criteria:
+1. **Semantic Error**: Objective misalignment with the source meaning or complete hallucinations.
+2. **Grammatical Error**: Clear violations of target language grammar or syntax rules.
+3. **Tone Failure**: A tone that is the opposite or significantly different from the [Intended Tone/Register].
+4. **Poor Readability**: The phrasing is so stiff, unnatural, or convoluted that it hinders smooth comprehension (e.g., obvious "translationese" or broken logic).
 
-[Output Format]
-You MUST respond in JSON format with the following keys. The values for "analysis" and "improvements" MUST be written in Simplified Chinese (简体中文), except when quoting the source or target text:
+**Note**: Do NOT penalize if the translation is simply "not the most elegant" or if there's a subjective preference for a different synonym. If it's natural enough for a native speaker to understand without effort, it's acceptable.
+
+# Instructions
+1. **Evaluation**: Compare the [Source Text] and [Translated Text] based on the [Audit Criteria].
+2. **Scoring Strategy**: 
+   - **90-100**: Accurate, grammatically sound, and flows naturally.
+   - **75-89**: Accurate meaning, but suffers from "stiff" phrasing or minor flow issues that need adjustment.
+   - **Below 75**: Contains semantic errors, severe grammar issues, or tone mismatches.
+3. **Analysis**: Provide a concise explanation in Simplified Chinese. Focus on *why* the error matters (e.g., "meaning is reversed" or "too awkward to read").
+4. **Improvements**: Provide suggestions in Simplified Chinese. If the score is 95+, this field can be an empty string "".
+
+# Output Format
+Respond ONLY in JSON format. The "analysis" and "improvements" MUST be in Simplified Chinese:
 {
   "score": number,
   "analysis": "string",