diff --git a/server/src/analyze/backend.rs b/server/src/analyze/backend.rs
index 077f6ba..f9077b5 100644
--- a/server/src/analyze/backend.rs
+++ b/server/src/analyze/backend.rs
@@ -42,26 +42,27 @@ const LLAMA_FORMAT_INSTRUCTION: &str = "AUSGABEFORMAT: Antworte ausschließlich
 /// Default system prompt for the consolidation LLM. Used by all backends
 /// unless overridden via `with_system_prompt`.
 ///
-/// Sandbox-validated against case `c414cf52` (3 runs each, fair pipeline
-/// with pre- and post-LLM gazetteer pass against the current vocab):
-/// - eliminates two hallucination classes the previous prompt produced —
-///   unmarked "Hypotonie" when the dictation said "Hypertonie" (0/3 vs 2/3),
-///   and inventing a unit ("35 ng/l") for a dictated value without unit
-///   (0/3 vs 2/3)
-/// - keeps Latin terms intact ("Punctum Maximum", "Apex Cordis", "Axilla"
-///   in 3/3 vs 2/3)
-/// - relies on the gazetteer for drug-name correction (Enoxaparin etc.) —
-///   the prompt no longer needs a separate "AKTIVE KORREKTUR" hammer-block
-///   because the gazetteer's pre-LLM pass repairs typical typos before the
-///   LLM ever sees them
-/// - block layout: AUFGABE / QUELLE / KORREKTUR-POLITIK / TREUE / CHRONOLOGIE
-///   / DOSIERUNGSSCHEMA / MARKIERUNGEN — each rule appears exactly once
+/// This is the "flag, don't guess" prompt: the model keeps the dictated
+/// (ASR) wording verbatim and only wraps uncertain stretches in `==...==`
+/// for the doctor to check. It must never substitute a term with a guess,
+/// drop a negation / finding / value, invert meaning, or invent facts —
+/// term repair is delegated to the gazetteer's pre-LLM pass, so the prompt
+/// stays purely about clinical fidelity.
+///
+/// It replaced an earlier permissive prompt that let the model rewrite
+/// terms and merely bracket the original — which, when the guess was wrong,
+/// inverted meanings, deleted negations, or invented drugs/doses. A/B'd in
+/// the experiments sandbox against 16 real alpha-test cases (full pipeline,
+/// three runs each, independently judged): it cut the per-run dangerous-error
+/// rate several-fold and eliminated the occasional empty-output run
+/// (reasoning length-exhaustion) the old prompt produced. Content-identical
+/// to [`LLAMA_SYSTEM_PROMPT`].
 const DEFAULT_SYSTEM_PROMPT: &str = include_str!("../../prompts/default_system_prompt.md");
 
-/// Llama-specific system prompt. Currently identical to the default; kept
-/// as a separate file so adjustments for Llama-3.1's behaviour (e.g.
-/// stricter "do not hallucinate" wording) can be tuned independently of
-/// gpt-oss without forking the default for everyone.
+/// Llama system prompt. Content-identical to [`DEFAULT_SYSTEM_PROMPT`] — the
+/// default was replaced by this same "flag, don't guess" prompt. Kept as a
+/// separate file so Llama-3.1-specific wording can diverge later without
+/// forking the default for everyone.
 const LLAMA_SYSTEM_PROMPT: &str = include_str!("../../prompts/llama_system_prompt.md");
 
 #[derive(Debug)]