feat: add is_article LLM check + remove use_llm_for_source_links option

The LLM now determines if scraped content is a real article during
classify (zero extra cost). The separate LLM link extraction option
is removed — heuristic extraction is sufficient.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

CLAUDE.md

@@ -117,7 +117,7 @@ cd frontend && npx tsc --noEmit
 - `GET /api/v1/admin/users` — user list
 - `PUT /api/v1/admin/users/:id/role` — role management
 
-## Database (25 migrations)
+## Database (26 migrations)
 Tables: `users`, `sessions`, `magic_link_tokens`, `user_settings`, `sources`, `syntheses`, `admin_providers`, `admin_rate_limits`, `user_api_keys`, `audit_log`
 
 ## Environment Variables

backend/migrations/20260326000026_remove_use_llm_for_source_links.sql

@@ -0,0 +1 @@
+ALTER TABLE settings DROP COLUMN use_llm_for_source_links;

backend/src/db/settings.rs

@@ -18,7 +18,6 @@ struct SettingsRow {
     categories: serde_json::Value,
     max_items_per_category: i32,
     max_articles_per_source: i32,
-    use_llm_for_source_links: bool,
     use_brave_search: bool,
     article_history_days: i32,
     batch_size: i32,
@@ -48,7 +47,6 @@ impl TryFrom<SettingsRow> for UserSettings {
             categories,
             max_items_per_category: row.max_items_per_category,
             max_articles_per_source: row.max_articles_per_source,
-            use_llm_for_source_links: row.use_llm_for_source_links,
             use_brave_search: row.use_brave_search,
             article_history_days: row.article_history_days,
             batch_size: row.batch_size,
@@ -80,10 +78,10 @@ pub async fn get_or_create_default(
 
     let row = sqlx::query_as::<_, SettingsRow>(
         r#"
-        INSERT INTO settings (user_id, theme, max_age_days, categories, max_items_per_category, search_agent_behavior, ai_provider, ai_model, ai_model_websearch, rate_limit_max_requests, rate_limit_time_window_seconds, max_articles_per_source, use_llm_for_source_links, use_brave_search, article_history_days, batch_size, summary_length, source_extraction_window)
-        VALUES ($1, $2, $3, $4, $5, $6, $7, $8, $9, $10, $11, $12, $13, $14, $15, $16, $17, $18)
+        INSERT INTO settings (user_id, theme, max_age_days, categories, max_items_per_category, search_agent_behavior, ai_provider, ai_model, ai_model_websearch, rate_limit_max_requests, rate_limit_time_window_seconds, max_articles_per_source, use_brave_search, article_history_days, batch_size, summary_length, source_extraction_window)
+        VALUES ($1, $2, $3, $4, $5, $6, $7, $8, $9, $10, $11, $12, $13, $14, $15, $16, $17)
         ON CONFLICT (user_id) DO UPDATE SET user_id = settings.user_id
-        RETURNING user_id, theme, max_age_days, categories, max_items_per_category, search_agent_behavior, ai_provider, ai_model, ai_model_websearch, rate_limit_max_requests, rate_limit_time_window_seconds, max_articles_per_source, use_llm_for_source_links, use_brave_search, article_history_days, batch_size, summary_length, source_extraction_window, updated_at
+        RETURNING user_id, theme, max_age_days, categories, max_items_per_category, search_agent_behavior, ai_provider, ai_model, ai_model_websearch, rate_limit_max_requests, rate_limit_time_window_seconds, max_articles_per_source, use_brave_search, article_history_days, batch_size, summary_length, source_extraction_window, updated_at
         "#,
     )
     .bind(user_id)
@@ -98,7 +96,6 @@ pub async fn get_or_create_default(
     .bind(defaults.rate_limit_max_requests)
     .bind(defaults.rate_limit_time_window_seconds)
     .bind(defaults.max_articles_per_source)
-    .bind(defaults.use_llm_for_source_links)
     .bind(defaults.use_brave_search)
     .bind(defaults.article_history_days)
     .bind(defaults.batch_size)
@@ -122,8 +119,8 @@ pub async fn upsert(
 
     let row = sqlx::query_as::<_, SettingsRow>(
         r#"
-        INSERT INTO settings (user_id, theme, max_age_days, categories, max_items_per_category, search_agent_behavior, ai_provider, ai_model, ai_model_websearch, rate_limit_max_requests, rate_limit_time_window_seconds, max_articles_per_source, use_llm_for_source_links, use_brave_search, article_history_days, batch_size, summary_length, source_extraction_window)
-        VALUES ($1, $2, $3, $4, $5, $6, $7, $8, $9, $10, $11, $12, $13, $14, $15, $16, $17, $18)
+        INSERT INTO settings (user_id, theme, max_age_days, categories, max_items_per_category, search_agent_behavior, ai_provider, ai_model, ai_model_websearch, rate_limit_max_requests, rate_limit_time_window_seconds, max_articles_per_source, use_brave_search, article_history_days, batch_size, summary_length, source_extraction_window)
+        VALUES ($1, $2, $3, $4, $5, $6, $7, $8, $9, $10, $11, $12, $13, $14, $15, $16, $17)
         ON CONFLICT (user_id) DO UPDATE SET
             theme = EXCLUDED.theme,
             max_age_days = EXCLUDED.max_age_days,
@@ -136,14 +133,13 @@ pub async fn upsert(
             rate_limit_max_requests = EXCLUDED.rate_limit_max_requests,
             rate_limit_time_window_seconds = EXCLUDED.rate_limit_time_window_seconds,
             max_articles_per_source = EXCLUDED.max_articles_per_source,
-            use_llm_for_source_links = EXCLUDED.use_llm_for_source_links,
             use_brave_search = EXCLUDED.use_brave_search,
             article_history_days = EXCLUDED.article_history_days,
             batch_size = EXCLUDED.batch_size,
             summary_length = EXCLUDED.summary_length,
             source_extraction_window = EXCLUDED.source_extraction_window,
             updated_at = now()
-        RETURNING user_id, theme, max_age_days, categories, max_items_per_category, search_agent_behavior, ai_provider, ai_model, ai_model_websearch, rate_limit_max_requests, rate_limit_time_window_seconds, max_articles_per_source, use_llm_for_source_links, use_brave_search, article_history_days, batch_size, summary_length, source_extraction_window, updated_at
+        RETURNING user_id, theme, max_age_days, categories, max_items_per_category, search_agent_behavior, ai_provider, ai_model, ai_model_websearch, rate_limit_max_requests, rate_limit_time_window_seconds, max_articles_per_source, use_brave_search, article_history_days, batch_size, summary_length, source_extraction_window, updated_at
         "#,
     )
     .bind(user_id)
@@ -158,7 +154,6 @@ pub async fn upsert(
     .bind(req.rate_limit_max_requests)
     .bind(req.rate_limit_time_window_seconds)
     .bind(req.max_articles_per_source)
-    .bind(req.use_llm_for_source_links)
     .bind(req.use_brave_search)
     .bind(req.article_history_days)
     .bind(req.batch_size)

backend/src/models/settings.rs

@@ -14,7 +14,7 @@ pub struct UserSettings {
     pub categories: Vec<String>,
     pub max_items_per_category: i32,
     pub max_articles_per_source: i32,
-    pub use_llm_for_source_links: bool,
+
     pub use_brave_search: bool,
     pub article_history_days: i32,
     pub batch_size: i32,
@@ -38,7 +38,7 @@ pub struct UpdateSettingsRequest {
     pub categories: Vec<String>,
     pub max_items_per_category: i32,
     pub max_articles_per_source: i32,
-    pub use_llm_for_source_links: bool,
+
     pub use_brave_search: bool,
     pub article_history_days: i32,
     pub batch_size: i32,
@@ -144,7 +144,7 @@ impl Default for UserSettings {
             ],
             max_items_per_category: 4,
             max_articles_per_source: 3,
-            use_llm_for_source_links: false,
+
             use_brave_search: false,
             article_history_days: 90,
             batch_size: 5,
@@ -173,7 +173,7 @@ mod tests {
             categories: vec!["Category 1".into(), "Category 2".into()],
             max_items_per_category: 4,
             max_articles_per_source: 3,
-            use_llm_for_source_links: false,
+
             use_brave_search: false,
             article_history_days: 90,
             batch_size: 5,

backend/src/services/llm/mock.rs

@@ -10,7 +10,6 @@ use super::LlmProvider;
 pub struct MockLlmProvider {
     default_category: String,
     search_urls: Vec<String>,
-    link_urls: Vec<String>,
 }
 
 impl MockLlmProvider {
@@ -18,7 +17,6 @@ impl MockLlmProvider {
         Self {
             default_category: "Autre".to_string(),
             search_urls: Vec::new(),
-            link_urls: Vec::new(),
         }
     }
 
@@ -32,11 +30,6 @@ impl MockLlmProvider {
         self
     }
 
-    pub fn with_link_urls(mut self, urls: Vec<String>) -> Self {
-        self.link_urls = urls;
-        self
-    }
-
     pub fn into_arc(self) -> Arc<dyn LlmProvider> {
         Arc::new(self)
     }
@@ -69,14 +62,10 @@ impl LlmProvider for MockLlmProvider {
                 "title": title,
                 "summary": format!("Mock summary for: {}", title),
                 "category": self.default_category,
+                "is_article": true,
             }));
         }
 
-        // Link extraction call
-        if sys_lower.contains("liens") {
-            return Ok(json!({ "urls": self.link_urls }));
-        }
-
         // Search call
         if sys_lower.contains("precis") {
             let items: Vec<Value> = self.search_urls.iter().map(|url| {
@@ -106,6 +95,7 @@ mod tests {
             .unwrap();
         assert_eq!(result["title"], "GPT-7");
         assert_eq!(result["category"], "AI News");
+        assert_eq!(result["is_article"], true);
     }
 
     #[tokio::test]
@@ -119,16 +109,4 @@ mod tests {
         let items = result["category_0"].as_array().unwrap();
         assert_eq!(items.len(), 1);
     }
-
-    #[tokio::test]
-    async fn mock_provider_returns_link_extraction() {
-        let provider = MockLlmProvider::new()
-            .with_link_urls(vec!["http://example.com/post-1".into()]);
-        let result = provider
-            .call_llm("model", "Tu dois identifier les liens", "Links...", &json!({}))
-            .await
-            .unwrap();
-        let urls = result["urls"].as_array().unwrap();
-        assert_eq!(urls.len(), 1);
-    }
 }

backend/src/services/llm/schema.rs

@@ -90,27 +90,14 @@ pub fn build_article_classify_schema() -> Value {
             "title": { "type": "string", "description": "Article title" },
             "summary": { "type": "string", "description": "Summary of the article" },
             "category": { "type": "string", "description": "Category name from the provided list" },
-            "date": { "type": "string", "description": "Publication date in YYYY-MM-DD format, or empty string if unknown" }
+            "date": { "type": "string", "description": "Publication date in YYYY-MM-DD format, or empty string if unknown" },
+            "is_article": { "type": "boolean", "description": "true if this is a real news article, false if it is a navigation page, contact page, terms of service, etc." }
         },
-        "required": ["title", "summary", "category", "date"],
+        "required": ["title", "summary", "category", "date", "is_article"],
         "additionalProperties": false
     })
 }
 
-/// Build a JSON Schema for LLM link extraction response.
-pub fn build_link_extraction_schema() -> Value {
-    serde_json::json!({
-        "type": "object",
-        "properties": {
-            "urls": {
-                "type": "array",
-                "items": { "type": "string" }
-            }
-        },
-        "required": ["urls"],
-        "additionalProperties": false
-    })
-}
 
 #[cfg(test)]
 mod tests {
@@ -302,11 +289,4 @@ mod tests {
         assert_eq!(schema["additionalProperties"], false);
     }
 
-    #[test]
-    fn link_extraction_schema_has_urls_array() {
-        let schema = build_link_extraction_schema();
-        assert_eq!(schema["properties"]["urls"]["type"], "array");
-        assert_eq!(schema["additionalProperties"], false);
-    }
-
 }

backend/src/services/prompts.rs

@@ -119,30 +119,6 @@ pub fn build_search_prompt(
     (system_prompt, user_prompt)
 }
 
-/// Build a prompt for LLM-assisted link extraction from a source page.
-///
-/// Receives a pre-formatted list of (href, anchor_text) pairs, not raw HTML.
-pub fn build_link_extraction_prompt(links_text: &str) -> (String, String) {
-    let system_prompt =
-        "Tu es un assistant qui analyse des listes de liens. \
-         Tu dois identifier les liens vers des articles d'actualite. \
-         Reponds uniquement au format JSON demande."
-            .to_string();
-
-    let user_prompt = format!(
-        "Voici une liste de liens extraits d'une page de blog ou de site d'actualites.\n\n\
-         {links}\n\n\
-         Selectionne UNIQUEMENT les URLs qui pointent vers des articles \
-         (pas les liens de navigation, tags, categories, login, pages statiques, topics, \
-         archive, companies, events, company, event, collections, etc.).\n\
-         Retourne les URLs completes, sans les modifier, dans le format JSON demande. \
-         Ne change jamais les URLs retournees, et ne les tronque jamais.",
-        links = links_text,
-    );
-
-    (system_prompt, user_prompt)
-}
-
 /// Build a prompt for per-article classification and summarization.
 ///
 /// The LLM classifies the article into a category and generates a title + summary.
@@ -180,7 +156,10 @@ pub fn build_article_classify_prompt(
          {summary_instruction}\n\
          Si le titre fourni est vide, genere un titre a partir du contenu.\n\
          Extrais la date de publication de l'article au format YYYY-MM-DD. \
-         Si la date n'est pas disponible, retourne une chaine vide.",
+         Si la date n'est pas disponible, retourne une chaine vide.\n\
+         Determine si ce contenu est un veritable article d'actualite. \
+         Retourne is_article=true pour un article, false pour une page de contact, \
+         mentions legales, page de navigation, FAQ, etc.",
         title = if title.is_empty() { "(pas de titre)" } else { title },
         body = body_snippet,
         categories = categories_list,
@@ -207,7 +186,6 @@ mod tests {
             ],
             max_items_per_category: 4,
             max_articles_per_source: 3,
-            use_llm_for_source_links: false,
             use_brave_search: false,
             article_history_days: 90,
             batch_size: 5,
@@ -365,21 +343,6 @@ mod tests {
         assert!(user_prompt.contains("exactement"));
     }
 
-    #[test]
-    fn link_extraction_prompt_includes_links() {
-        let links = "- https://example.com/post-1 | \"Breaking News\"\n- https://example.com/post-2 | \"Update\"";
-        let (sys, user) = build_link_extraction_prompt(links);
-        assert!(user.contains("https://example.com/post-1"));
-        assert!(user.contains("Breaking News"));
-        assert!(sys.contains("liens"));
-    }
-
-    #[test]
-    fn link_extraction_prompt_empty_links() {
-        let (_, user) = build_link_extraction_prompt("");
-        assert!(user.contains("articles"));
-    }
-
     #[test]
     fn article_classify_prompt_includes_content() {
         let (sys, user) = build_article_classify_prompt(

backend/src/services/source_scraper.rs

@@ -3,11 +3,8 @@
 //! Used in Phase 1 of the generation pipeline to discover articles
 //! from user-configured sources before falling back to LLM web search.
 
-use std::sync::{Arc, LazyLock};
+use std::sync::LazyLock;
 use crate::errors::AppError;
-use crate::services::llm::LlmProvider;
-use crate::services::llm::schema::build_link_extraction_schema;
-use crate::services::prompts::build_link_extraction_prompt;
 use scraper::{Html, Selector};
 use url::Url;
 
@@ -125,167 +122,6 @@ pub fn extract_links_from_html(
     links
 }
 
-/// Extract all links from HTML as (href, anchor_text) pairs for LLM analysis.
-///
-/// Minimal filtering: same-domain, http/https, non-empty path.
-/// No article-pattern filtering — the LLM decides which are articles.
-pub fn extract_links_as_pairs(
-    html: &str,
-    base_url: &Url,
-) -> Vec<(String, String)> {
-    let base_domain = base_url.host_str().unwrap_or("").to_lowercase();
-    let document = Html::parse_document(html);
-    let mut pairs = Vec::new();
-
-    for element in document.select(&ANCHOR_SELECTOR) {
-        if let Some(href) = element.value().attr("href") {
-            let resolved = match base_url.join(href) {
-                Ok(u) => u,
-                Err(_) => continue,
-            };
-
-            if resolved.scheme() != "http" && resolved.scheme() != "https" {
-                continue;
-            }
-
-            let link_domain = resolved.host_str().unwrap_or("").to_lowercase();
-            if link_domain != base_domain {
-                continue;
-            }
-
-            let path = resolved.path();
-            if path.is_empty() || path == "/" {
-                continue;
-            }
-
-            let anchor_text: String = element.text().collect::<Vec<_>>().join(" ");
-            let anchor_text = anchor_text.trim().to_string();
-
-            pairs.push((resolved.to_string(), anchor_text));
-        }
-    }
-
-    pairs
-}
-
-/// Format link pairs as a text list for the LLM prompt.
-/// Caps at 200 links to limit token usage.
-fn format_links_for_llm(pairs: &[(String, String)]) -> String {
-    pairs
-        .iter()
-        .take(200)
-        .map(|(href, text)| {
-            if text.is_empty() {
-                format!("- {}", href)
-            } else {
-                format!("- {} | \"{}\"", href, text)
-            }
-        })
-        .collect::<Vec<_>>()
-        .join("\n")
-}
-
-/// Extract article links using LLM analysis of the page HTML.
-///
-/// Falls back to heuristic extraction if the LLM call fails or returns empty.
-#[allow(clippy::too_many_arguments)]
-pub async fn extract_article_links_with_llm(
-    http_client: &reqwest::Client,
-    source_url: &str,
-    max_links: usize,
-    provider: &Arc<dyn LlmProvider>,
-    model: &str,
-    pool: Option<&sqlx::PgPool>,
-    user_id: Option<uuid::Uuid>,
-    job_id: Option<uuid::Uuid>,
-) -> Result<Vec<String>, AppError> {
-    let base_url = Url::parse(source_url)
-        .map_err(|e| AppError::BadRequest(format!("Invalid source URL: {}", e)))?;
-
-    // SSRF check before fetching
-    if let Err(e) = crate::services::scraper::check_ssrf(&base_url).await {
-        tracing::warn!(url = source_url, error = %e, "Source URL failed SSRF check");
-        return Ok(Vec::new());
-    }
-
-    let base_domain = base_url.host_str().unwrap_or("").to_lowercase();
-
-    let response = http_client.get(source_url).send().await.map_err(|e| {
-        tracing::warn!(url = source_url, error = %e, "Failed to fetch source page");
-        AppError::Internal(anyhow::anyhow!("Failed to fetch source page"))
-    })?;
-
-    if !response.status().is_success() {
-        tracing::warn!(url = source_url, status = %response.status(), "Source page returned non-200");
-        return Ok(Vec::new());
-    }
-
-    let html_text = response.text().await.map_err(|e| {
-        AppError::Internal(anyhow::anyhow!("Failed to read source page body: {}", e))
-    })?;
-
-    let pairs = extract_links_as_pairs(&html_text, &base_url);
-    let links_text = format_links_for_llm(&pairs);
-    let (system, user) = build_link_extraction_prompt(&links_text);
-    let schema = build_link_extraction_schema();
-
-    let llm_start = std::time::Instant::now();
-    let llm_result = provider.call_llm(model, &system, &user, &schema).await;
-    let llm_duration = llm_start.elapsed().as_millis() as u64;
-
-    // Log the LLM call if pool/user_id/job_id are provided
-    if let (Some(pool), Some(uid), Some(jid)) = (pool, user_id, job_id) {
-        let response_str = match &llm_result {
-            Ok(resp) => serde_json::to_string_pretty(resp).unwrap_or_default(),
-            Err(e) => format!("Error: {}", e),
-        };
-        crate::db::llm_call_log::insert(
-            pool, uid, jid, "link_extraction", model,
-            &system, &user, &response_str, llm_duration as i32,
-            Some(source_url),
-        ).await.ok();
-    }
-
-    match llm_result {
-        Ok(llm_response) => {
-            let urls: Vec<String> = llm_response
-                .get("urls")
-                .and_then(|u| u.as_array())
-                .map(|arr| {
-                    arr.iter()
-                        .filter_map(|v| v.as_str())
-                        .filter_map(|href| {
-                            let resolved = base_url.join(href).ok()?;
-                            if resolved.scheme() != "http" && resolved.scheme() != "https" {
-                                return None;
-                            }
-                            if resolved.host_str()?.to_lowercase() != base_domain {
-                                return None;
-                            }
-                            Some(resolved.to_string())
-                        })
-                        .collect()
-                })
-                .unwrap_or_default();
-
-            if urls.is_empty() {
-                tracing::warn!(url = source_url, "LLM returned no links, falling back to heuristic");
-                let fallback = extract_links_from_html(&html_text, &base_url, &base_domain);
-                Ok(fallback.into_iter().take(max_links).collect())
-            } else {
-                let mut seen = std::collections::HashSet::new();
-                let deduped: Vec<String> = urls.into_iter().filter(|u| seen.insert(u.clone())).collect();
-                Ok(deduped.into_iter().take(max_links).collect())
-            }
-        }
-        Err(e) => {
-            tracing::warn!(url = source_url, error = %e, "LLM link extraction failed, falling back to heuristic");
-            let fallback = extract_links_from_html(&html_text, &base_url, &base_domain);
-            Ok(fallback.into_iter().take(max_links).collect())
-        }
-    }
-}
-
 #[cfg(test)]
 mod tests {
     use super::*;
@@ -376,67 +212,4 @@ mod tests {
         let links = extract_links_from_html("", &base_url("https://example.com"), "example.com");
         assert!(links.is_empty());
     }
-
-    #[test]
-    fn extract_pairs_returns_href_and_text() {
-        let html = r#"
-        <html><body>
-            <a href="/blog/article-1">Breaking AI News</a>
-            <a href="/blog/article-2">GPT-6 Released</a>
-        </body></html>"#;
-        let base = base_url("https://example.com/blog");
-        let pairs = extract_links_as_pairs(html, &base);
-        assert_eq!(pairs.len(), 2);
-        assert!(pairs[0].0.contains("/blog/article-1"));
-        assert_eq!(pairs[0].1, "Breaking AI News");
-        assert!(pairs[1].0.contains("/blog/article-2"));
-        assert_eq!(pairs[1].1, "GPT-6 Released");
-    }
-
-    #[test]
-    fn extract_pairs_filters_external_links() {
-        let html = r#"<a href="https://other.com/article">External</a>"#;
-        let base = base_url("https://example.com");
-        let pairs = extract_links_as_pairs(html, &base);
-        assert!(pairs.is_empty());
-    }
-
-    #[test]
-    fn extract_pairs_filters_root_path() {
-        let html = r#"<a href="/">Home</a>"#;
-        let base = base_url("https://example.com");
-        let pairs = extract_links_as_pairs(html, &base);
-        assert!(pairs.is_empty());
-    }
-
-    #[test]
-    fn extract_pairs_handles_empty_anchor_text() {
-        let html = r#"<a href="/article"><img src="pic.jpg"/></a>"#;
-        let base = base_url("https://example.com");
-        let pairs = extract_links_as_pairs(html, &base);
-        assert_eq!(pairs.len(), 1);
-        assert_eq!(pairs[0].1, "");
-    }
-
-    #[test]
-    fn format_links_for_llm_formats_correctly() {
-        let pairs = vec![
-            ("https://example.com/a".to_string(), "Article One".to_string()),
-            ("https://example.com/b".to_string(), "".to_string()),
-        ];
-        let result = format_links_for_llm(&pairs);
-        assert!(result.contains("- https://example.com/a | \"Article One\""));
-        assert!(result.contains("- https://example.com/b"));
-        assert!(!result.contains("| \"\""));
-    }
-
-    #[test]
-    fn format_links_for_llm_caps_at_200() {
-        let pairs: Vec<(String, String)> = (0..300)
-            .map(|i| (format!("https://example.com/{}", i), format!("Link {}", i)))
-            .collect();
-        let result = format_links_for_llm(&pairs);
-        let line_count = result.lines().count();
-        assert_eq!(line_count, 200);
-    }
 }

backend/src/services/synthesis.rs

@@ -314,24 +314,9 @@ pub async fn run_generation_inner(
                     let client = state.http_client.clone();
                     let source_url = source.url.clone();
                     let source_title = source.title.clone();
-                    let use_llm = settings.use_llm_for_source_links;
-                    let provider_clone = std::sync::Arc::clone(&provider);
-                    let model = Arc::clone(&model_research);
                     let max_l = max_links;
-                    let pool = state.pool.clone();
-                    let uid = user_id;
-                    let jid = job_id;
                     join_set.spawn(async move {
-                        let links = if use_llm {
-                            source_scraper::extract_article_links_with_llm(
-                                &client, &source_url, max_l, &provider_clone, &model,
-                                Some(&pool), Some(uid), Some(jid),
-                            ).await
-                        } else {
-                            source_scraper::extract_article_links(
-                                &client, &source_url, max_l,
-                            ).await
-                        };
+                        let links = source_scraper::extract_article_links(&client, &source_url, max_l).await;
                         (source_url, source_title, links)
                     });
                 }
@@ -503,6 +488,19 @@ pub async fn run_generation_inner(
                                 }
                             };
 
+                            // Check if LLM considers this a real article
+                            let is_article = class_response.get("is_article").and_then(|v| v.as_bool()).unwrap_or(true);
+                            if !is_article {
+                                tracing::info!(url = %final_url, "Article filtered by LLM: not a real article");
+                                pending_traces.push(build_trace_entry(user_id, job_id, &ArticleTrace {
+                                    url: &final_url, title: &page_title, source_type: "personalized_source",
+                                    source_url: Some(&source_url), category: None, synthesis_id: None,
+                                    status: "filtered_not_article", scraped_ok: true,
+                                    published_date: None,
+                                }));
+                                continue;
+                            }
+
                             // Check LLM-extracted date as fallback for articles without a scraper date
                             if let Some(date_str) = class_response.get("date").and_then(|d| d.as_str()) {
                                 if !date_str.is_empty() {
@@ -715,6 +713,19 @@ pub async fn run_generation_inner(
                                 }
                             };
 
+                            // Check if LLM considers this a real article
+                            let is_article = class_response.get("is_article").and_then(|v| v.as_bool()).unwrap_or(true);
+                            if !is_article {
+                                tracing::info!(url = %final_url, "Article filtered by LLM: not a real article");
+                                pending_traces.push(build_trace_entry(user_id, job_id, &ArticleTrace {
+                                    url: &final_url, title: &page_title, source_type: "brave_search",
+                                    source_url: None, category: None, synthesis_id: None,
+                                    status: "filtered_not_article", scraped_ok: true,
+                                    published_date: None,
+                                }));
+                                continue;
+                            }
+
                             // Check LLM-extracted date as fallback
                             if let Some(date_str) = class_response.get("date").and_then(|d| d.as_str()) {
                                 if !date_str.is_empty() {

backend/tests/api_settings_test.rs

@@ -47,7 +47,7 @@ async fn put_settings_without_auth_returns_401() {
         "categories": ["Cat"],
         "max_items_per_category": 4,
         "max_articles_per_source": 3,
-        "use_llm_for_source_links": false,
+
         "use_brave_search": false,
         "article_history_days": 90,
         "batch_size": 5,
@@ -134,7 +134,7 @@ async fn put_settings_with_valid_data_returns_200() {
         "categories": ["Vulnerabilites", "Patch Tuesday", "Threat Intel"],
         "max_items_per_category": 6,
         "max_articles_per_source": 3,
-        "use_llm_for_source_links": false,
+
         "use_brave_search": false,
         "article_history_days": 90,
         "batch_size": 5,
@@ -192,7 +192,7 @@ async fn put_then_get_returns_updated_data() {
         "categories": ["Macro", "Finance"],
         "max_items_per_category": 10,
         "max_articles_per_source": 3,
-        "use_llm_for_source_links": false,
+
         "use_brave_search": false,
         "article_history_days": 90,
         "batch_size": 5,
@@ -244,7 +244,7 @@ async fn put_settings_empty_theme_returns_422() {
         "categories": ["Cat"],
         "max_items_per_category": 4,
         "max_articles_per_source": 3,
-        "use_llm_for_source_links": false,
+
         "use_brave_search": false,
         "article_history_days": 90,
         "batch_size": 5,
@@ -288,7 +288,7 @@ async fn put_settings_too_many_categories_returns_422() {
         "categories": categories,
         "max_items_per_category": 4,
         "max_articles_per_source": 3,
-        "use_llm_for_source_links": false,
+
         "use_brave_search": false,
         "article_history_days": 90,
         "batch_size": 5,
@@ -331,7 +331,7 @@ async fn put_settings_empty_categories_returns_422() {
         "categories": [],
         "max_items_per_category": 4,
         "max_articles_per_source": 3,
-        "use_llm_for_source_links": false,
+
         "use_brave_search": false,
         "article_history_days": 90,
         "batch_size": 5,
@@ -375,7 +375,7 @@ async fn put_settings_max_age_days_out_of_range_returns_422() {
         "categories": ["Cat"],
         "max_items_per_category": 4,
         "max_articles_per_source": 3,
-        "use_llm_for_source_links": false,
+
         "use_brave_search": false,
         "article_history_days": 90,
         "batch_size": 5,
@@ -404,7 +404,7 @@ async fn put_settings_max_age_days_out_of_range_returns_422() {
         "categories": ["Cat"],
         "max_items_per_category": 4,
         "max_articles_per_source": 3,
-        "use_llm_for_source_links": false,
+
         "use_brave_search": false,
         "article_history_days": 90,
         "batch_size": 5,
@@ -445,7 +445,7 @@ async fn put_settings_max_items_out_of_range_returns_422() {
         "categories": ["Cat"],
         "max_items_per_category": 51,
         "max_articles_per_source": 3,
-        "use_llm_for_source_links": false,
+
         "use_brave_search": false,
         "article_history_days": 90,
         "batch_size": 5,
@@ -494,7 +494,7 @@ async fn settings_are_per_user_isolated() {
         "categories": ["A-Category"],
         "max_items_per_category": 2,
         "max_articles_per_source": 3,
-        "use_llm_for_source_links": false,
+
         "use_brave_search": false,
         "article_history_days": 90,
         "batch_size": 5,
@@ -519,7 +519,7 @@ async fn settings_are_per_user_isolated() {
         "categories": ["B-Category-1", "B-Category-2"],
         "max_items_per_category": 8,
         "max_articles_per_source": 3,
-        "use_llm_for_source_links": false,
+
         "use_brave_search": false,
         "article_history_days": 90,
         "batch_size": 5,
@@ -575,7 +575,7 @@ async fn put_settings_boundary_values_succeed() {
         "categories": ["C"],
         "max_items_per_category": 1,
         "max_articles_per_source": 3,
-        "use_llm_for_source_links": false,
+
         "use_brave_search": false,
         "article_history_days": 90,
         "batch_size": 5,
@@ -601,7 +601,7 @@ async fn put_settings_boundary_values_succeed() {
         "categories": categories_max,
         "max_items_per_category": 50,
         "max_articles_per_source": 3,
-        "use_llm_for_source_links": false,
+
         "use_brave_search": false,
         "article_history_days": 90,
         "batch_size": 5,

backend/tests/api_syntheses_test.rs

@@ -631,7 +631,7 @@ async fn generate_pipeline_resolves_model_from_admin_config() {
         "categories": ["Test Category"],
         "max_items_per_category": 4,
         "max_articles_per_source": 3,
-        "use_llm_for_source_links": false,
+
         "use_brave_search": false,
         "article_history_days": 90,
         "batch_size": 5,

backend/tests/pipeline_test.rs

@@ -45,7 +45,6 @@ async fn setup_user_with_settings(
     app: &common::TestApp,
     categories: Vec<&str>,
     max_items: i32,
-    use_llm_for_links: bool,
 ) -> (uuid::Uuid, String) {
     let email = format!("pipeline-{}@test.com", uuid::Uuid::new_v4());
     let (user_id, session) = app.create_authenticated_user(&email).await;
@@ -57,7 +56,6 @@ async fn setup_user_with_settings(
         "categories": categories_json,
         "max_items_per_category": max_items,
         "max_articles_per_source": 10,
-        "use_llm_for_source_links": use_llm_for_links,
         "use_brave_search": false,
         "article_history_days": 90,
         "batch_size": 5,
@@ -86,12 +84,11 @@ fn make_progress_channel() -> (Arc<watch::Sender<synthesis::ProgressEvent>>, wat
 }
 
 #[tokio::test]
-async fn phase1_with_llm_link_extraction_classifies_articles() {
+async fn phase1_heuristic_extraction_classifies_articles() {
     let app = common::TestApp::new().await;
     let mock_server = setup_mock_server().await;
 
-    // Use LLM link extraction to bypass SSRF on source page
-    let (user_id, session) = setup_user_with_settings(&app, vec!["AI News"], 4, true).await;
+    let (user_id, session) = setup_user_with_settings(&app, vec!["AI News"], 4).await;
 
     // Add a source pointing to wiremock (same host as article URLs)
     let source_url = format!("{}/blog", mock_server.uri());
@@ -99,14 +96,8 @@ async fn phase1_with_llm_link_extraction_classifies_articles() {
     let (status, _) = app.post_with_session("/api/v1/sources", &source, &session).await;
     assert!(status.is_success());
 
-    // Mock provider: LLM link extraction returns wiremock article URLs (same domain)
-    let article_urls: Vec<String> = (1..=3)
-        .map(|i| format!("{}/article-{}", mock_server.uri(), i))
-        .collect();
-
     let mock_provider = MockLlmProvider::new()
         .with_default_category("AI News")
-        .with_link_urls(article_urls)
         .into_arc();
 
     let job_id = uuid::Uuid::new_v4();
@@ -159,7 +150,7 @@ async fn phase2_search_fills_gaps_when_no_sources() {
     let mock_server = setup_mock_server().await;
 
     // No sources — Phase 1 produces nothing
-    let (user_id, _session) = setup_user_with_settings(&app, vec!["AI News"], 2, false).await;
+    let (user_id, _session) = setup_user_with_settings(&app, vec!["AI News"], 2).await;
 
     let mock_provider = MockLlmProvider::new()
         .with_default_category("AI News")
@@ -202,19 +193,14 @@ async fn category_overflow_spills_to_autre() {
     let mock_server = setup_mock_server().await;
 
     // max_items_per_category=1, but LLM classifies all articles to "AI News"
-    let (user_id, session) = setup_user_with_settings(&app, vec!["AI News"], 1, true).await;
+    let (user_id, session) = setup_user_with_settings(&app, vec!["AI News"], 1).await;
 
     let source_url = format!("{}/blog", mock_server.uri());
     let source = serde_json::json!({"title": "Test Source", "url": source_url});
     app.post_with_session("/api/v1/sources", &source, &session).await;
 
-    let article_urls: Vec<String> = (1..=3)
-        .map(|i| format!("{}/article-{}", mock_server.uri(), i))
-        .collect();
-
     let mock_provider = MockLlmProvider::new()
         .with_default_category("AI News")
-        .with_link_urls(article_urls)
         .into_arc();
 
     let job_id = uuid::Uuid::new_v4();

e2e/tests/generation-live.spec.ts

@@ -141,7 +141,6 @@ test.describe('Live generation with OpenAI', () => {
       ai_provider: 'openai',
       ai_model: 'gpt-4o-mini',
       ai_model_websearch: 'gpt-4o-mini',
-      use_llm_for_source_links: false,
       use_brave_search: false,
       article_history_days: 90,
       batch_size: 5,

frontend/src/components/settings/SettingsAdvanced.tsx

@@ -14,7 +14,6 @@ interface SettingsAdvancedProps {
  * Groups fields that control extraction and pipeline behaviour:
  * - `article_history_days` — deduplication window
  * - `batch_size` — number of sources processed per LLM batch
- * - `use_llm_for_source_links` — whether to use LLM to extract links
  * - `search_agent_behavior` — free-text prompt injection for the search agent
  */
 const SettingsAdvanced: Component<SettingsAdvancedProps> = (props) => {
@@ -106,32 +105,6 @@ const SettingsAdvanced: Component<SettingsAdvancedProps> = (props) => {
         </div>
       </div>
 
-      {/* Advanced extraction */}
-      <div class="mt-6">
-        <h3 class="text-lg font-medium text-gray-900 mb-4">
-          {t('settings.advancedExtraction')}
-        </h3>
-        <div class="space-y-4">
-          <div class="flex items-center">
-            <input
-              type="checkbox"
-              id="useLlmSourceLinks"
-              checked={props.settings().use_llm_for_source_links}
-              onChange={(e) =>
-                props.setSettings((prev) => ({
-                  ...prev,
-                  use_llm_for_source_links: e.currentTarget.checked,
-                }))
-              }
-              class="h-4 w-4 text-indigo-600 focus:ring-indigo-500 border-gray-300 rounded"
-            />
-            <label for="useLlmSourceLinks" class="ml-2 block text-sm text-gray-700">
-              {t('settings.useLlmForSourceLinks')}
-            </label>
-          </div>
-        </div>
-      </div>
-
       {/* Search agent behavior */}
       <div>
         <label

frontend/src/i18n/fr.ts

@@ -150,8 +150,6 @@ const fr = {
   'settings.rateLimitHelp': "Configurez le nombre maximum de requetes autorisees pendant la fenetre de temps specifiee. Laissez vide pour utiliser les valeurs par defaut de l'administrateur.",
   'settings.rateLimitEffective': '{max} requetes / {window} secondes',
   'settings.rateLimitReset': 'Reinitialiser',
-  'settings.advancedExtraction': 'Extraction avancee',
-  'settings.useLlmForSourceLinks': "Utiliser l'IA pour extraire les liens",
   'settings.useBraveSearch': 'Utiliser Brave Search pour la recherche web',
   'settings.useBraveSearchHelp': "Remplace la recherche web par IA par l'API Brave Search pour des resultats plus precis.",
   'settings.braveSearch': 'Brave Search',

frontend/src/types.ts

@@ -44,7 +44,6 @@ export interface UserSettings {
   max_age_days: number;
   max_items_per_category: number;
   max_articles_per_source: number;
-  use_llm_for_source_links: boolean;
   use_brave_search: boolean;
   article_history_days: number;
   batch_size: number;
@@ -64,7 +63,6 @@ export const DEFAULT_SETTINGS: UserSettings = {
   max_age_days: 7,
   max_items_per_category: 4,
   max_articles_per_source: 3,
-  use_llm_for_source_links: false,
   use_brave_search: false,
   article_history_days: 90,
   batch_size: 5,