Refine part implementation

conflict_resolving.py

@@ -265,6 +265,48 @@ def remove_node(self, node: str) -> None:
         for node, node in to_remove:
             self.bad_graph.pop(node, None)
 
+    def get_some_cycle_from_scc(self, scc: List[str]) -> List[str]:
+        """
+        Find and return some cycle from a strongly connected component (SCC).
+
+        Args:
+            scc (List[str]): A list of assertion IDs forming a strongly connected component.
+
+        Returns:
+            List[str]: A list of assertion IDs forming a cycle within the SCC in order.
+                       Returns an empty list if no cycle is found (e.g., single-node SCC with no self-loop).
+
+        Description:
+            - Performs a depth-first search (DFS) starting from the first node in `scc`.
+            - Maintains a `path` stack to keep track of the current DFS path.
+            - Uses `nonlocal cycle` to store the first detected cycle:
+                * As soon as a neighbor node is encountered that is already in `path`,
+                  the slice of `path` from that node to the current node is assigned to `cycle`.
+                * Further DFS recursion stops once a cycle is found.
+            - Only explores neighbors that are part of the SCC.
+            - Useful for detecting a cycle that can then be resolved (e.g., by node removal).
+        """
+        path = []
+        cycle = []
+
+        def dfs(curr):
+            nonlocal cycle
+            path.append(curr)
+            for neighbour in self.good_graph_1[curr]:
+                if cycle:
+                    continue
+                if neighbour not in scc:
+                    continue
+                if neighbour in path:
+                    # cycle found, return slice of path
+                    cycle = path[path.index(neighbour):]
+                    continue
+                dfs(neighbour)
+            path.pop()
+            return None
+
+        dfs(scc[0])
+        return cycle
 
     def pick_worst_node(self, nodes_part_of_scc) -> str:
         """
@@ -364,7 +406,7 @@ def resolve_cycles_and_conflicts(self, automatic) -> bool:
                 if automatic:
                     remove_nodes = [random.choice(nodes_part_of_scc)]
                 else:
-                    if node := self.resolve_cycle_by_user(list_of_scc[0]):
+                    if node := self.resolve_cycle_by_user(self.get_some_cycle_from_scc(list_of_scc[0])):
                         remove_nodes = [node]
             else:
                 if automatic:
@@ -377,7 +419,7 @@ def resolve_cycles_and_conflicts(self, automatic) -> bool:
                         if des := self.resolve_contradiction_by_user(min_node, list(self.bad_graph[min_node])) is not None:
                             remove_nodes = [min_node] if des else list(self.bad_graph[min_node])
                     else:
-                        if node := self.resolve_cycle_by_user(list_of_scc[0]):
+                        if node := self.resolve_cycle_by_user(self.get_some_cycle_from_scc(list_of_scc[0])):
                             remove_nodes = [node]
 
             if remove_nodes != []:

streamlit_app.py

@@ -20,6 +20,8 @@
 from models import Assertion, Relationship
 from app import ClarusApp, create_clarus_app
 from structure import evaluate_relationship_quality
+from langchain_openai import ChatOpenAI
+from langchain_core.messages import HumanMessage
 
 # Page configuration
 st.set_page_config(
@@ -1003,9 +1005,89 @@ def prose_tab():
 
     st.markdown('<div class="prose-mode">', unsafe_allow_html=True)
     st.header("📖 Prose Mode")
-    st.markdown("Transform your refined assertions into fluent, reader-friendly text.")
-
-    st.info("🚧 Prose mode is coming soon! This will transform your assertions into well-structured prose.")
+    st.markdown("Transform your refined assertions into fluent, reader-friendly academic prose.")
+
+    # Ensure plan exists
+    if "result_text" not in st.session_state:
+        st.session_state.result_text = ""
+
+    # Editable plan text area
+    plan_text = st.text_area(
+        "Plan from Review (edit if needed)",
+        value=st.session_state.get("result_text", ""),
+        key="prose_plan_text",
+        height=220,
+        help="This plan will be expanded into a full academic text."
+    )
+
+    # Sync back to result_text to keep it current
+    st.session_state.result_text = plan_text
+
+    # Model and options
+    col1, col2, col3 = st.columns([1, 1.5, 1.5])
+    with col1:
+        temperature = st.slider("Creativity", 0.0, 1.0, 0.3, 0.1, help="Lower = more formal/precise, Higher = more creative")
+    with col2:
+        style = st.selectbox("Style", ["Academic", "Technical"], index=0, help="Choose the writing style for the generated text")
+    with col3:
+        add_headings = st.checkbox("Include suggested headings", value=False)
+
+    # Generation button
+    disabled = len(plan_text.strip()) == 0
+    if disabled:
+        st.warning("The plan is empty. Please provide or generate a plan in Review mode.")
+
+    button_label = f"Generate {style} Text"
+    spinner_label = f"Generating {style.lower()} text..."
+
+    if st.button(button_label, disabled=disabled, type="primary"):
+        try:
+            with st.spinner(spinner_label):
+                llm = ChatOpenAI(model="gpt-4o-mini", temperature=temperature)
+                if style == "Academic":
+                    instruction = (
+                        "You are an expert academic writer. Expand the provided plan into a coherent, formal academic text suitable for a short paper or report section. "
+                        "Requirements: 1) Maintain logical flow and argumentative coherence; 2) Use precise, objective, and formal tone; 3) Add connective tissue (transitions, definitions, brief context) as needed; "
+                        "4) Avoid bullet points; write continuous prose; 5) Do not invent unsupported claims, but you may elaborate reasoning from the plan; "
+                        "6) Where appropriate, structure content with an introduction, logically ordered paragraphs, and a concise concluding synthesis; "
+                        "7) Remove meta-commentary about the task itself."
+                    )
+                    if add_headings:
+                        instruction += " 8) Include concise section headings appropriate for academic writing."
+                else:
+                    instruction = (
+                        "You are a senior technical writer. Expand the provided plan into clear, precise, implementation-oriented technical text suitable for engineering documentation or a design note. "
+                        "Requirements: 1) Prioritize clarity, unambiguous terminology, and actionability; 2) Use concise sentences and active voice; 3) Where helpful, include numbered steps, bullet lists, or short code-style blocks; "
+                        "4) Provide definitions and assumptions upfront; 5) Include explicit inputs, outputs, constraints, and edge cases when implied by the plan; 6) Avoid marketing language and rhetorical flourish; "
+                        "7) Keep sections well-scoped and skimmable."
+                    )
+                    if add_headings:
+                        instruction += " 8) Include concise section headings appropriate for technical documentation."
+
+                prompt = (
+                    f"{instruction}\n\n"
+                    "PLAN TO EXPAND:\n"
+                    "----------------\n"
+                    f"{plan_text}\n"
+                )
+                message = HumanMessage(content=prompt)
+                response = llm.invoke([message])
+                generated_text = getattr(response, "content", str(response))
+                st.session_state.academic_text = generated_text
+        except Exception as e:
+            st.error(f"Failed to generate {style.lower()} text: {e}")
+
+    # Show output if available
+    if st.session_state.get("academic_text"):
+        st.subheader(f"Generated {style} Text")
+        st.text_area(f"{style} Text", value=st.session_state.academic_text, height=420, key="academic_text_display")
+        st.download_button(
+            label=f"Download {style} Text (.txt)",
+            data=st.session_state.academic_text,
+            file_name=f"{style.lower()}_text.txt",
+            mime="text/plain"
+        )
+
     st.markdown('</div>', unsafe_allow_html=True)
 
 def main():