wip: snapshot das alterações não commitadas (2026-10-20)

Origem: D:/Documents/Github/OllaHub
Arquivos omitidos (segredos, bancos de dados, logs, >50 MB): 0

lib/content-condenser.ts

@@ -3,6 +3,7 @@ import type { ResearchEntry } from '@/hooks/use-deep-research';
 export interface Chunk {
   content: string;
   score: number;
+  relevanceScore?: number;
   sourceIndex: number;
   startIndex: number;
   endIndex: number;
@@ -45,17 +46,17 @@ export function summarizeMarkdown(content: string, maxLen: number = 2000): strin
   // Dividir por headings (###, ##, #)
   const headingRegex = /^(#{1,3})\s+(.+)$/gm;
   const sections: Array<{ level: number; title: string; content: string; startIndex: number }> = [];
-  
+
   let lastIndex = 0;
   let match;
   const contentLines = content.split('\n');
-  
+
   // Encontrar todas as seções
   while ((match = headingRegex.exec(content)) !== null) {
     const level = match[1].length;
     const title = match[2].trim();
     const startIndex = content.substring(0, match.index).split('\n').length - 1;
-    
+
     // Encontrar conteúdo da seção (até próximo heading ou fim)
     let endIndex = contentLines.length;
     const nextHeadingRegex = /^(#{1,3})\s+/gm;
@@ -64,24 +65,24 @@ export function summarizeMarkdown(content: string, maxLen: number = 2000): strin
     if (nextMatch) {
       endIndex = content.substring(0, nextMatch.index).split('\n').length - 1;
     }
-    
+
     const sectionContent = contentLines.slice(startIndex + 1, endIndex).join('\n').trim();
-    
+
     sections.push({
       level,
       title,
       content: sectionContent,
       startIndex,
     });
-    
+
     lastIndex = endIndex;
   }
-  
+
   // Se não encontrou headings, dividir por parágrafos
   if (sections.length === 0) {
     const paragraphs = content.split(/\n\s*\n/).filter(p => p.trim().length > 0);
     let summary = '';
-    
+
     for (const para of paragraphs) {
       if (summary.length + para.length + 2 <= maxLen) {
         summary += para.trim() + '\n\n';
@@ -94,17 +95,17 @@ export function summarizeMarkdown(content: string, maxLen: number = 2000): strin
         break;
       }
     }
-    
+
     return summary.trim() || content.substring(0, maxLen);
   }
-  
+
   // Construir sumário priorizando seções de nível superior
   let summary = '';
   const sortedSections = sections.sort((a, b) => a.level - b.level);
-  
+
   for (const section of sortedSections) {
     const sectionText = `## ${section.title}\n\n${section.content}\n\n`;
-    
+
     if (summary.length + sectionText.length <= maxLen) {
       summary += sectionText;
     } else {
@@ -117,7 +118,7 @@ export function summarizeMarkdown(content: string, maxLen: number = 2000): strin
       break;
     }
   }
-  
+
   return summary.trim() || content.substring(0, maxLen);
 }
 
@@ -138,18 +139,18 @@ export function chunkBySections(
 
   const chunks: Chunk[] = [];
   const lines = markdown.split('\n');
-  
+
   // Dividir por headings primeiro
   const headingRegex = /^(#{1,3})\s+(.+)$/;
   let currentChunk: string[] = [];
   let currentHeading = '';
   let startIndex = 0;
   let chunkIndex = 0;
-  
+
   for (let i = 0; i < lines.length; i++) {
     const line = lines[i];
     const headingMatch = line.match(headingRegex);
-    
+
     if (headingMatch) {
       // Se já temos um chunk, salvá-lo
       if (currentChunk.length > 0) {
@@ -165,14 +166,14 @@ export function chunkBySections(
           chunkIndex++;
         }
       }
-      
+
       // Iniciar novo chunk
       currentHeading = line;
       currentChunk = [line];
       startIndex = i;
     } else {
       currentChunk.push(line);
-      
+
       // Se chunk ficou muito grande, dividir por parágrafos
       const currentContent = currentChunk.join('\n');
       if (currentContent.length > maxChunkSize) {
@@ -188,14 +189,14 @@ export function chunkBySections(
           });
           chunkIndex++;
         }
-        
+
         // Iniciar novo chunk com a linha atual
         currentChunk = [currentHeading, line].filter(Boolean);
         startIndex = i;
       }
     }
   }
-  
+
   // Adicionar último chunk
   if (currentChunk.length > 0) {
     const chunkContent = currentChunk.join('\n');
@@ -209,16 +210,16 @@ export function chunkBySections(
       });
     }
   }
-  
+
   // Se não encontrou headings, dividir por parágrafos
   if (chunks.length === 0) {
     const paragraphs = markdown.split(/\n\s*\n/).filter(p => p.trim().length > 0);
     let currentChunkContent = '';
     let paraStartIndex = 0;
-    
+
     for (let i = 0; i < paragraphs.length; i++) {
       const para = paragraphs[i].trim();
-      
+
       if (currentChunkContent.length + para.length + 2 <= maxChunkSize) {
         currentChunkContent += (currentChunkContent ? '\n\n' : '') + para;
       } else {
@@ -232,13 +233,13 @@ export function chunkBySections(
             endIndex: i - 1,
           });
         }
-        
+
         // Iniciar novo chunk
         currentChunkContent = para;
         paraStartIndex = i;
       }
     }
-    
+
     // Adicionar último chunk
     if (currentChunkContent.length > 0) {
       chunks.push({
@@ -250,7 +251,7 @@ export function chunkBySections(
       });
     }
   }
-  
+
   return chunks;
 }
 
@@ -267,7 +268,7 @@ export function extractKeyFacts(markdown: string): string[] {
 
   const facts: string[] = [];
   const normalized = markdown.toLowerCase();
-  
+
   // Extrair datas (formato brasileiro: DD/MM/YYYY ou DD-MM-YYYY)
   const dateRegex = /\b(\d{1,2}[\/\-]\d{1,2}[\/\-]\d{2,4})\b/g;
   const dates = [...markdown.matchAll(dateRegex)];
@@ -276,7 +277,7 @@ export function extractKeyFacts(markdown: string): string[] {
       facts.push(match[0]);
     }
   });
-  
+
   // Extrair anos (4 dígitos entre 1900-2100)
   const yearRegex = /\b(19\d{2}|20[0-2]\d)\b/g;
   const years = [...markdown.matchAll(yearRegex)];
@@ -285,7 +286,7 @@ export function extractKeyFacts(markdown: string): string[] {
       facts.push(match[0]);
     }
   });
-  
+
   // Extrair números relevantes (valores, estatísticas)
   // Padrão: número seguido de unidade ou contexto
   const numberPatterns = [
@@ -293,7 +294,7 @@ export function extractKeyFacts(markdown: string): string[] {
     /\bR\$\s*(\d{1,3}(?:\.\d{3})*(?:,\d+)?)/gi,
     /\b(\d{1,3}(?:\.\d{3})*(?:,\d+)?)\s*%/g,
   ];
-  
+
   numberPatterns.forEach(pattern => {
     const matches = [...markdown.matchAll(pattern)];
     matches.forEach(match => {
@@ -303,13 +304,13 @@ export function extractKeyFacts(markdown: string): string[] {
       }
     });
   });
-  
+
   // Extrair locais (padrões comuns: "em [Local]", "de [Local]", "para [Local]")
   const locationPatterns = [
     /\b(?:em|de|para|no|na|nos|nas)\s+([A-ZÁÉÍÓÚÂÊÔ][a-záéíóúâêôãõç]+(?:\s+[A-ZÁÉÍÓÚÂÊÔ][a-záéíóúâêôãõç]+)*)/g,
     /\b([A-ZÁÉÍÓÚÂÊÔ][a-záéíóúâêôãõç]+\s+(?:de\s+)?[A-ZÁÉÍÓÚÂÊÔ][a-záéíóúâêôãõç]+)\b/g,
   ];
-  
+
   locationPatterns.forEach(pattern => {
     const matches = [...markdown.matchAll(pattern)];
     matches.forEach(match => {
@@ -323,7 +324,7 @@ export function extractKeyFacts(markdown: string): string[] {
       }
     });
   });
-  
+
   // Limitar a 20 fatos mais relevantes
   return facts.slice(0, 20);
 }
@@ -342,7 +343,7 @@ export function processContent(
   options: ProcessingOptions = {}
 ): ProcessedContent {
   const opts = { ...DEFAULT_OPTIONS, ...options };
-  
+
   if (!content || content.length === 0) {
     return {
       original: content,
@@ -351,30 +352,30 @@ export function processContent(
       totalTokens: 0,
     };
   }
-  
+
   // Sumarizar se muito longo
   let processedContent = content;
   let summarized: string | undefined;
-  
+
   if (content.length > opts.summarizeThreshold) {
     summarized = summarizeMarkdown(content, opts.maxSummaryLength);
     processedContent = summarized;
   }
-  
+
   // Extrair fatos-chave
   const keyFacts = extractKeyFacts(content);
-  
+
   // Fazer chunking
   const chunks = chunkBySections(processedContent, opts.maxChunkSize);
-  
+
   // Atribuir sourceIndex aos chunks (todos do mesmo source por enquanto)
   chunks.forEach((chunk, idx) => {
     chunk.sourceIndex = 0;
   });
-  
+
   // Calcular tokens totais (aproximação: 1 token ≈ 4 caracteres)
   const totalTokens = Math.ceil(processedContent.length / 4);
-  
+
   return {
     original: content,
     summarized,