diff --git a/src/deriver/deriver.py b/src/deriver/deriver.py
index b4e746d3..c95105eb 100644
--- a/src/deriver/deriver.py
+++ b/src/deriver/deriver.py
@@ -331,10 +331,15 @@ class CertaintyReasoner:
             latest_message.created_at,
         )
 
-        # Step 2: Deductive reasoning (receives explicit observations)
+        # Step 2: Deductive reasoning (receives atomic propositions)
+        # Extract atomic propositions from explicit and implicit observations
+        atomic_propositions = [
+            obs.content for obs in explicit_observations.explicit
+        ] + [obs.content for obs in explicit_observations.implicit]
+
         deductive_response = await self.deductive_reasoner.reason(
             working_representation=working_representation,
-            explicit_observations=explicit_observations,
+            atomic_propositions=atomic_propositions,
             history=history,
             speaker_peer_card=speaker_peer_card,
         )
@@ -357,7 +362,7 @@ class CertaintyReasoner:
         analysis_duration_ms = (time.perf_counter() - analysis_start) * 1000
         accumulate_metric(
             f"deriver_{latest_message.id}_{self.observer}",
-            "critical_analysis_duration",
+            "reasoning_duration",
             analysis_duration_ms,
             "ms",
         )
diff --git a/src/deriver/prompts.py b/src/deriver/prompts.py
index ff25be10..2998c9e5 100644
--- a/src/deriver/prompts.py
+++ b/src/deriver/prompts.py
@@ -100,7 +100,7 @@ def estimate_base_prompt_tokens() -> int:
             peer_card=None,
             message_created_at=datetime.datetime.now(datetime.timezone.utc),
             working_representation=Representation(),
-            explicit_observations=Representation(),
+            atomic_propositions=[],
             history="",
             new_turns=[],
         )
@@ -153,7 +153,7 @@ def deductive_reasoning_prompt(
     peer_card: list[str] | None,
     message_created_at: datetime.datetime,
     working_representation: Representation,
-    explicit_observations: Representation,
+    atomic_propositions: list[str],
     history: str,
     new_turns: list[str],
 ) -> str:
@@ -165,23 +165,29 @@ def deductive_reasoning_prompt(
         peer_card (list[str] | None): The bio card of the user being analyzed.
         message_created_at (datetime.datetime): Timestamp of the message.
         working_representation (Representation): Current user understanding context.
-        explicit_observations (Representation): New explicit observations from current batch
-            (includes both explicit and implicit propositions).
+        atomic_propositions (list[str]): New atomic propositions from explicit reasoning
+            (includes both explicit and implicit observations as content strings).
         history (str): Recent conversation history.
         new_turns (list[str]): New conversation turns to analyze.
 
     Returns:
         Formatted prompt string for deductive reasoning
     """
-    # Format atomic propositions (includes both explicit and implicit from ExplicitReasoner)
-    # as numbered list - combine both lists
-    all_atomic_propositions = [
-        obs.content for obs in explicit_observations.explicit
-    ] + [obs.content for obs in explicit_observations.implicit]
+    # Format atomic propositions as numbered list
     atomic_propositions_section = "\n".join(
-        [f"{i}. {prop}" for i, prop in enumerate(all_atomic_propositions, 1)]
+        [f"{i}. {prop}" for i, prop in enumerate(atomic_propositions, 1)]
     )
 
+    # Format existing deductions from working representation
+    # Uses the same format as Representation.__str__() for DEDUCTIVE section
+    existing_deductions_section = ""
+    if working_representation.deductive:
+        deduction_strings = [
+            f"{i}. {deduction}"
+            for i, deduction in enumerate(working_representation.deductive, 1)
+        ]
+        existing_deductions_section = "\n".join(deduction_strings)
+
     return render_template(
         settings.DERIVER.DEDUCTIVE_REASONING_TEMPLATE,
         {
@@ -190,9 +196,8 @@ def deductive_reasoning_prompt(
             "message_created_at": message_created_at,
             "working_representation": str(working_representation),
             "has_working_representation": not working_representation.is_empty(),
-            "explicit_observations": str(explicit_observations),
-            "has_explicit_observations": not explicit_observations.is_empty(),
             "atomic_propositions_section": atomic_propositions_section,
+            "existing_deductions_section": existing_deductions_section,
             "history": history,
             "new_turns": new_turns,
         },
diff --git a/src/deriver/reasoners/deductive.py b/src/deriver/reasoners/deductive.py
index cbe71d93..d5632615 100644
--- a/src/deriver/reasoners/deductive.py
+++ b/src/deriver/reasoners/deductive.py
@@ -55,7 +55,7 @@ class DeductiveReasoner(BaseReasoner):
     async def reason(
         self,
         working_representation: Representation,
-        explicit_observations: Representation,
+        atomic_propositions: list[str],
         history: str,
         speaker_peer_card: list[str] | None,
     ) -> DeductiveResponse:
@@ -63,7 +63,8 @@ class DeductiveReasoner(BaseReasoner):
 
         Args:
             working_representation: Current representation context
-            explicit_observations: New explicit observations from current batch
+            atomic_propositions: New atomic propositions from explicit reasoning
+                (both explicit and implicit observations as content strings)
             history: Recent conversation history
             speaker_peer_card: Peer card for the observed peer
 
@@ -81,7 +82,7 @@ class DeductiveReasoner(BaseReasoner):
             peer_card=speaker_peer_card,
             message_created_at=latest_message.created_at,
             working_representation=working_representation,
-            explicit_observations=explicit_observations,
+            atomic_propositions=atomic_propositions,
             history=history,
             new_turns=new_turns,
         )
diff --git a/src/templates/deriver/deductive.jinja b/src/templates/deriver/deductive.jinja
index 666f68c2..319933c7 100644
--- a/src/templates/deriver/deductive.jinja
+++ b/src/templates/deriver/deductive.jinja
@@ -65,7 +65,7 @@ Therefore, you must ONLY generate deductions that:
   * "Carlos has a daughter starting kindergarten" + "Kindergarten typically starts at age 5" → "Carlos's daughter is approximately 5 years old" (domain-specific knowledge application)
   * "Elena graduated with a PhD in neuroscience" + "PhDs require bachelor's degrees" → "Elena completed a bachelor's degree in a relevant field" (non-obvious educational prerequisite)
 
-**KEY PRINCIPLE:** Only generate deductions that add substantive information that is semantically differentiated from the atomic propositions. Ask yourself: 
+**KEY PRINCIPLE:** Only generate deductions that add substantive information that is semantically differentiated from the atomic propositions. Ask yourself:
 1. "Would the explicit/implicit extraction have already captured this?" If yes, don't generate it.
 2. "Does this conclusion connect or build on multiple propositions in a non-obvious way?" If no, don't generate it.
 3. "Does this add meaningful context about who {{ peer_id }} is that isn't already present?" If no, don't generate it.
@@ -118,7 +118,7 @@ Common valid deductive patterns include:
 **EXAMPLES OF VALID DEDUCTIONS:**
 
 Example 1 - Categorical Syllogism:
-- PREMISES: 
+- PREMISES:
   * "Maria attended college" (atomic proposition)
   * All people who attended college completed high school or equivalent (general knowledge)
 - CONCLUSION: "Maria completed high school or equivalent education"
@@ -148,7 +148,7 @@ Example 5 - Multi-step Scaffolding:
   * "Elena graduated with a PhD in neuroscience" (atomic proposition)
   * A PhD requires completing a bachelor's degree (general knowledge)
   * A bachelor's degree requires completing high school (general knowledge)
-- CONCLUSIONS: 
+- CONCLUSIONS:
   * "Elena completed a bachelor's degree"
   * "Elena completed high school or equivalent education"
 
@@ -177,7 +177,7 @@ Each deduction must be self-contained and include sufficient context:
 - Disambiguating details that make the conclusion independently meaningful
 - All necessary qualifiers to ensure accuracy
 
-{{ peer_card }}
+{{ peer_card|join('\n') }}
 
 {{ existing_deductions_section }}
 
@@ -228,4 +228,4 @@ Generate ALL valid deductions that can be derived from the available premises. S
         },
     ]
 }
-```
\ No newline at end of file
+```
diff --git a/src/templates/deriver/explicit.jinja b/src/templates/deriver/explicit.jinja
index 809e68be..53fcf76f 100644
--- a/src/templates/deriver/explicit.jinja
+++ b/src/templates/deriver/explicit.jinja
@@ -6,7 +6,7 @@ Extract atomic propositions from the peer's message. An atomic proposition is:
 **The Critical Balance:**
 Each proposition must be atomic (indivisible) yet contain enough semantic context to be interpretable without reference to other propositions.
 
-- ❌ TOO ATOMIC (lacks context): 
+- ❌ TOO ATOMIC (lacks context):
   * "Maria is happy" → Happy about what?
   * "James said hi" → Said hi to whom? In what context?
   * "Sarah went there" → Went where?
@@ -27,7 +27,7 @@ Each proposition must be atomic (indivisible) yet contain enough semantic contex
 1. **EXPLICIT EXTRACTION** - Directly stated facts:
    - Extract propositions directly asserted in the message
    - Each claim becomes a separate atomic proposition
-   
+
 2. **IMPLICIT EXTRACTION** - Clearly implied facts:
    - Extract propositions that are obviously implied by the message
    - Only include implications that are certain, not speculative
@@ -96,7 +96,7 @@ Example 4 - Implicit Extraction:
 
 {{ peer_id }}'s known biographical information:
 <peer_card>
-{{ peer_card }}
+{{ peer_card|join('\n') }}
 </peer_card>
 
 Current understanding of {{ peer_id }}:
@@ -111,23 +111,23 @@ Recent conversation history for context:
 
 New conversation turns to analyze:
 <new_turns>
-{{ new_turns }}
+{{ new_turns|join('\n') }}
 </new_turns>
 
 Extract ALL atomic propositions (both explicit and clearly implied) from the latest peer message. Output your response in JSON structured format:
 ```json
 {
-    "explicit":[
-        "explicit proposition 1",
-        "explicit proposition 2",
+    "explicit": [
+        {"content": "explicit proposition 1"},
+        {"content": "explicit proposition 2"},
         ...
-        "explicit proposition n"
+        {"content": "explicit proposition n"}
     ],
-    "implicit":[
-        "implicit proposition 1",
-        "implicit proposition 2",
+    "implicit": [
+        {"content": "implicit proposition 1"},
+        {"content": "implicit proposition 2"},
         ...
-        "implicit proposition n"
+        {"content": "implicit proposition n"}
     ]
 }
-```
\ No newline at end of file
+```