feat: Handle analyze retry with failure marker

When a previous analysis attempt fails and leaves a failure marker, the
retry mechanism now correctly identifies this state. It removes the
stale
input file, allowing the retry to proceed without a conflict error. This
ensures users can recover from analysis failures more gracefully.

server/src/analyze/backend.rs

@@ -119,6 +119,11 @@ impl LlmBackend {
         self.format_instruction = Some(instruction.into());
         self
     }
+
+    fn with_max_completion_tokens(mut self, tokens: u32) -> Self {
+        self.max_completion_tokens = tokens;
+        self
+    }
 }
 
 /// Ionos provider: shared endpoint, `IONOS_API_KEY` env var as bearer token.
@@ -149,15 +154,17 @@ pub fn backends() -> &'static [LlmBackend] {
     static B: OnceLock<Vec<LlmBackend>> = OnceLock::new();
     B.get_or_init(|| {
         vec![
-            // gpt-oss-120b: schema OFF on purpose. With `reasoning_effort:
-            // medium`, reasoning tokens count against `max_completion_tokens`
-            // (4096); on large bodies (e.g. 6 recordings, ~3.8k chars) the
-            // reasoning eats the budget and the model returns
-            // `choices[0].message.content == null` — the LlmError::Parse path.
-            // Pre-multi-backend (commit bf6464d) gpt-oss ran without schema
-            // and worked fine; this restores that body byte-for-byte.
+            // gpt-oss-120b: schema OFF on purpose. Reasoning tokens count
+            // against `max_completion_tokens`; with `high` reasoning the
+            // CoT alone can eat several thousand tokens, so the budget is
+            // doubled to 8192 to leave room for the actual answer. Without
+            // headroom the model returns `choices[0].message.content == null`
+            // — the LlmError::Parse path. Pre-multi-backend (commit bf6464d)
+            // gpt-oss ran without schema and worked fine; this restores
+            // that body byte-for-byte except for the larger budget.
             ionos_backend("gpt_oss_120b", "GPT OSS 120b", "openai/gpt-oss-120b")
-                .with_reasoning_effort("medium")
+                .with_reasoning_effort("high")
+                .with_max_completion_tokens(8192)
                 .with_use_json_schema(false),
             ionos_backend(
                 "llama_3_1_405b",
@@ -234,7 +241,7 @@ mod tests {
     fn reasoning_effort_is_set_per_model_family() {
         assert_eq!(
             find_backend("gpt_oss_120b").unwrap().reasoning_effort,
-            Some("medium".to_string())
+            Some("high".to_string())
         );
         assert!(
             find_backend("llama_3_1_405b")
@@ -304,11 +311,11 @@ mod tests {
     }
 
     /// gpt-oss-120b runs in plain-text mode (no `response_format`, no
-    /// `top_p`, no second system message). Reason: with
-    /// `reasoning_effort: medium`, schema-constrained decoding plus reasoning
-    /// can exhaust `max_completion_tokens` on large bodies and return
-    /// `content: null`. This pin guarantees the body sent to Ionos stays
-    /// byte-identical to the pre-multi-backend path (commit bf6464d).
+    /// `top_p`, no second system message). Reason: schema-constrained
+    /// decoding plus any non-zero `reasoning_effort` can exhaust
+    /// `max_completion_tokens` on large bodies and return `content: null`.
+    /// This pin guarantees the body sent to Ionos stays byte-identical to
+    /// the pre-multi-backend path (commit bf6464d).
     #[test]
     fn gpt_oss_runs_in_plain_text_mode_no_schema_no_top_p_no_format_hint() {
         let b = find_backend("gpt_oss_120b").unwrap();

server/src/routes/case_actions.rs

@@ -14,6 +14,7 @@ use tracing::{info, warn};
 use doctate_common::timestamp::{filename_stem_to_recorded_at, now_rfc3339};
 use doctate_common::{RECORDING_META_SUFFIX, TranscriptState};
 
+use crate::analyze::auto_trigger::FAILURE_MARKER_FILE;
 use crate::analyze::backend::{any_backend_available, find_backend};
 use crate::analyze::{
     ANALYSIS_INPUT_FILE, AnalysisInput, AnalyzeJob, AnalyzeSender, DOCUMENT_FILE, RecordingInput,
@@ -107,6 +108,17 @@ pub async fn handle_analyze_case(
 
     let input = build_analysis_input(&case_dir, &backend_id).await?;
     let input_path = case_dir.join(ANALYSIS_INPUT_FILE);
+
+    // Manual retry path: a failure marker means the worker has explicitly
+    // given up on the prior input. Drop the stale sentinel so create_new
+    // does not reject the legitimate retry as "Analyse läuft bereits".
+    // The marker itself is left in place — the worker overwrites it on
+    // re-failure and removes it on success, so the UI's "analyzing" flag
+    // (driven by input file presence) takes over from the next render.
+    if case_dir.join(FAILURE_MARKER_FILE).exists() {
+        remove_if_exists(&input_path).await?;
+    }
+
     write_input_create_new(&input_path, &input).await?;
 
     // Unbounded send: only fails if the worker has gone away (programmer error