diff --git a/chat/runtime.py b/chat/runtime.py
index ffc3e981..1841f5f7 100644
--- a/chat/runtime.py
+++ b/chat/runtime.py
@@ -177,6 +177,13 @@ def _build_vault_probe(vault, vocab):
     return _probe
 
 
+def _vault_probe_for_context(context: SessionContext | None) -> Any | None:
+    """Return a _VaultProbe callable or None for the given session context."""
+    if context is None:
+        return None
+    return _build_vault_probe(context.vault, context.vocab)
+
+
 def _energy_scalar(energy_obj) -> float:
     if energy_obj is None:
         return 1.0
@@ -646,7 +653,15 @@ class ChatRuntime:
         if store is None:
             return
         store.save_recognizers(self._recognizer_registry.all())
-        store.save_discovery_candidates(self._pending_candidates)
+        candidates_to_save = self._pending_candidates
+        if self.config.auto_contemplate and candidates_to_save:
+            from teaching.contemplation import contemplate
+            vault_probe = _vault_probe_for_context(self._context) if self._context else None
+            candidates_to_save = [
+                contemplate(c, vault_probe=vault_probe)
+                for c in candidates_to_save
+            ]
+        store.save_discovery_candidates(candidates_to_save)
         store.save_manifest(self._turn_count)
 
     def _checkpointed_response(self, response: ChatResponse) -> ChatResponse:
diff --git a/core/config.py b/core/config.py
index 4084e10f..e00d92fd 100644
--- a/core/config.py
+++ b/core/config.py
@@ -269,6 +269,10 @@ class RuntimeConfig:
     # Unlocks W-007 (DerivedRecognizer derivation from promoted COHERENT entries).
     vault_promotion_enabled: bool = False
 
+    # ADR-0150 — run contemplation on pending discovery candidates at checkpoint.
+    # Activates ADR-0056 Phase C1. Null-drop when False.
+    auto_contemplate: bool = False
+
 
 DEFAULT_IDENTITY_PACK: str = "default_general_v1"
 DEFAULT_ETHICS_PACK: str = "default_general_ethics_v1"
diff --git a/docs/decisions/ADR-0150-autonomous-inter-session-contemplation.md b/docs/decisions/ADR-0150-autonomous-inter-session-contemplation.md
new file mode 100644
index 00000000..d4f4444c
--- /dev/null
+++ b/docs/decisions/ADR-0150-autonomous-inter-session-contemplation.md
@@ -0,0 +1,33 @@
+# ADR-0150 — Autonomous Inter-Session Contemplation
+
+Status: Accepted
+Date: 2026-05-25
+
+## Context
+ADR-0056 Phase C1 shipped `contemplate()` as a pure function that enriches
+DiscoveryCandidate with polarity, evidence, claim_domain, and sub_questions.
+It ran inline (opt-in via attach_contemplation) or via CLI batch. Neither path
+ran at session boundaries. Engine state (ADR-0146) persists discovery candidates
+to disk, but stored candidates were unenriched (raw Phase B output).
+
+## Decision
+Run `contemplate()` on pending session candidates at `checkpoint_engine_state()`
+before persisting to `engine_state/discovery_candidates.jsonl`. Enriched
+candidates (polarity/evidence/claim_domain populated) are stored instead of
+raw ones.
+
+Flag: `RuntimeConfig.auto_contemplate = False` (null-drop default).
+
+## Trust boundary
+`contemplate()` is read-only w.r.t. corpus, pack, and vault per ADR-0056.
+It enriches the in-memory candidate struct only. Nothing is written to any
+shared store during enrichment.
+
+## Why checkpoint, not inline
+Fresh candidates are produced during the turn and accumulated in
+`_pending_candidates`. Contemplation at checkpoint runs after the session
+completes, not on the hot turn path. This avoids blocking turn latency.
+
+## Unlocks
+W-017: auto-proposal pipeline can filter enriched candidates (polarity,
+evidence) to generate TeachingChainProposals.
diff --git a/tests/test_adr_0150_autonomous_contemplation.py b/tests/test_adr_0150_autonomous_contemplation.py
new file mode 100644
index 00000000..4af0ab25
--- /dev/null
+++ b/tests/test_adr_0150_autonomous_contemplation.py
@@ -0,0 +1,87 @@
+from __future__ import annotations
+
+import json
+from pathlib import Path
+from core.config import RuntimeConfig
+from chat.runtime import ChatRuntime
+from engine_state import EngineStateStore
+from teaching.discovery import DiscoveryCandidate, EvidencePointer, SubQuestion
+from chat.pack_grounding import _pack_index
+from chat.teaching_grounding import _corpus_index
+from teaching.contemplation import contemplate
+
+def _raw_candidate() -> DiscoveryCandidate:
+    return DiscoveryCandidate(
+        candidate_id="cand-1",
+        proposed_chain={"subject": "light", "intent": "verification", "connective": None, "object": None},
+        trigger="would_have_grounded",
+        source_turn_trace="trace-1",
+        pack_consistent=True,
+        boundary_clean=True,
+    )
+
+def test_auto_contemplate_off_stores_raw_candidates(tmp_path: Path) -> None:
+    config = RuntimeConfig(auto_contemplate=False)
+    runtime = ChatRuntime(config=config, engine_state_path=tmp_path)
+    
+    cand = _raw_candidate()
+    runtime._pending_candidates.append(cand)
+    runtime.checkpoint_engine_state()
+    
+    store = EngineStateStore(tmp_path)
+    loaded = store.load_discovery_candidates()
+    assert len(loaded) == 1
+    assert loaded[0].polarity == "undetermined"
+    assert len(loaded[0].evidence) == 0
+
+def test_auto_contemplate_enriches_at_checkpoint(tmp_path: Path) -> None:
+    config = RuntimeConfig(auto_contemplate=True)
+    runtime = ChatRuntime(config=config, engine_state_path=tmp_path)
+    
+    cand = _raw_candidate()
+    runtime._pending_candidates.append(cand)
+    runtime.checkpoint_engine_state()
+    
+    # Clear pending candidates and load from disk
+    runtime._pending_candidates = []
+    runtime._load_engine_state()
+    
+    assert len(runtime._pending_candidates) == 1
+    loaded_cand = runtime._pending_candidates[0]
+    
+    # Verify loaded candidate has claim_domain and evidence populated (not defaults)
+    assert loaded_cand.claim_domain == "factual"
+    assert len(loaded_cand.evidence) > 0
+    # The default polarity is undetermined, but evidence is populated
+    assert any(e.source == "pack" and e.ref == "light" for e in loaded_cand.evidence)
+
+def test_enriched_candidates_survive_jsonl_round_trip(tmp_path: Path) -> None:
+    cand = _raw_candidate()
+    enriched = contemplate(cand)
+    
+    store = EngineStateStore(tmp_path)
+    store.save_discovery_candidates([enriched])
+    
+    loaded = store.load_discovery_candidates()
+    assert len(loaded) == 1
+    loaded_cand = loaded[0]
+    
+    assert loaded_cand.polarity == enriched.polarity
+    assert loaded_cand.claim_domain == enriched.claim_domain
+    assert loaded_cand.evidence == enriched.evidence
+    assert loaded_cand.sub_questions == enriched.sub_questions
+    assert loaded_cand.contemplation_depth == enriched.contemplation_depth
+    assert loaded_cand.recursion_overflow == enriched.recursion_overflow
+
+def test_contemplate_does_not_write_corpus_or_pack() -> None:
+    corpus_before = dict(_corpus_index())
+    pack_before = dict(_pack_index())
+    
+    cand = _raw_candidate()
+    _ = contemplate(cand)
+    
+    corpus_after = dict(_corpus_index())
+    pack_after = dict(_pack_index())
+    
+    assert corpus_before == corpus_after
+    assert pack_before == pack_after