diff --git a/Makefile b/Makefile
index 47e4168..2d47049 100644
--- a/Makefile
+++ b/Makefile
@@ -211,9 +211,18 @@ query: bootstrap ## ask the corpus a question [JSON=1 BURN=1 LLM=qwen|hermes REP
 # tree node with a Bates id + a session-wide subtree_hash. Forking is
 # implicit: /cd to a parent node and ask again → sibling under that
 # parent. See arborist/qa/session.py + docs/sessions.md.
-session: bootstrap ## interactive multi-turn Q&A REPL [SID=... LLM=qwen|hermes ANSWER_MODE=... BURN=1]
-	$(if $(LLM_ENDPOINT),@echo "# llm:    $(LLM_ENDPOINT) / $(LLM_MODEL)" >&2,)
-	$(ARBORIST) --shards-dir $(SHARDS_DIR) session $(if $(SID),--sid $(SID),) --top-k $(QUERY_TOP_K) $(if $(LLM_ENDPOINT),--endpoint $(LLM_ENDPOINT) --model $(LLM_MODEL),) $(if $(ANSWER_MODE),--answer-mode $(ANSWER_MODE),) $(if $(BURN),--burn,)
+#
+# Session defaults to Qwen3.6-27B (better answers for interactive
+# multi-turn use). Override with LLM=hermes for the 8B fast path.
+SESSION_LLM_ENDPOINT ?= https://qwen.ai.unturf.com/v1
+SESSION_LLM_MODEL    ?= Qwen3.6-27B-UD-Q4_K_XL.gguf
+ifeq ($(LLM),hermes)
+SESSION_LLM_ENDPOINT := https://hermes.ai.unturf.com/v1
+SESSION_LLM_MODEL    := adamo1139/Hermes-3-Llama-3.1-8B-FP8-Dynamic
+endif
+session: bootstrap ## interactive multi-turn Q&A REPL [SID=... LLM=qwen|hermes ANSWER_MODE=... BURN=1]; defaults LLM=qwen
+	@echo "# llm:    $(SESSION_LLM_ENDPOINT) / $(SESSION_LLM_MODEL)" >&2
+	$(ARBORIST) --shards-dir $(SHARDS_DIR) session $(if $(SID),--sid $(SID),) --top-k $(QUERY_TOP_K) --endpoint $(SESSION_LLM_ENDPOINT) --model $(SESSION_LLM_MODEL) $(if $(ANSWER_MODE),--answer-mode $(ANSWER_MODE),) $(if $(BURN),--burn,)
 
 session-list: bootstrap ## list all session dbs in ~/.arborist/sessions/
 	$(ARBORIST) session --list $(if $(JSON),--json,)
diff --git a/arborist/cli.py b/arborist/cli.py
index afe5860..85a6110 100644
--- a/arborist/cli.py
+++ b/arborist/cli.py
@@ -7502,6 +7502,15 @@ def build_parser() -> argparse.ArgumentParser:
     return p
 
 
+def _ignore(fn, *a, **kw):
+    """Run fn, swallow any exception. Used for atexit hooks where
+    failure (e.g. write to a closed file) shouldn't crash shutdown."""
+    try:
+        fn(*a, **kw)
+    except Exception:
+        pass
+
+
 def _cmd_session(args: argparse.Namespace) -> int:
     """Merkle-rooted multi-turn Q&A REPL with forks.
 
@@ -7661,6 +7670,28 @@ def _session_repl(args: argparse.Namespace, sess) -> int:
     )
     from arborist.store import connect as _connect
 
+    # readline gives up-arrow line history + emacs/vi line editing
+    # in the input() call below. The module is in the stdlib but
+    # importing it is what wires it into input(). Persist history to
+    # ~/.arborist/sessions/.history so commands survive across runs.
+    try:
+        import readline
+        _hist = (
+            Path.home() / ".arborist" / "sessions" / ".history"
+        )
+        _hist.parent.mkdir(parents=True, exist_ok=True)
+        try:
+            readline.read_history_file(str(_hist))
+        except (FileNotFoundError, OSError):
+            pass
+        readline.set_history_length(2000)
+        import atexit
+        atexit.register(
+            lambda: _ignore(readline.write_history_file, str(_hist))
+        )
+    except ImportError:
+        pass
+
     shards_dir = getattr(args, "global_shards_dir", None) or getattr(
         args, "shards_dir", None
     )
diff --git a/arborist/qa/providence_query.py b/arborist/qa/providence_query.py
index 4def8ef..2384673 100644
--- a/arborist/qa/providence_query.py
+++ b/arborist/qa/providence_query.py
@@ -308,6 +308,59 @@ def providence_query(
             def emit(self, *a, **kw): pass
         progress = _NoProgress()
 
+    # 0. π* canonical projection preflight (math / algebra / logic).
+    # For questions that are pure arithmetic ("33/66", "0.1+0.2"),
+    # pure algebra ("x^2 - 4 = 0"), or pure propositional logic
+    # ("A AND NOT B"), short-circuit retrieval + LLM entirely and
+    # return the deterministic canonical bytes. Without this,
+    # "33/66" was tokenized by FTS5 as ["33", "66"] which matched
+    # the year-article titles "33" and "66" — LLM then "interpreted"
+    # the bare fraction as years 33 AD / 66 AD. The legacy query()
+    # pipeline has the same preflight; providence_query was missing
+    # it (#000015 / #000030).
+    # Off-by-default for tests + bench paths that need deterministic
+    # retrieval shapes; ON for interactive (set by callers).
+    if policy.get("canonical_projection_preflight", True):
+        try:
+            from arborist.qa.query import _canonical_projection_preflight
+            preflight = _canonical_projection_preflight(question)
+        except Exception:
+            preflight = None
+        if preflight is not None:
+            pi_star_ref, canonical_bytes = preflight
+            answer_text = canonical_bytes.decode("utf-8", errors="replace")
+            progress.emit("canonical_projection.hit", pi_star_ref=pi_star_ref)
+            return {
+                "status": "canonical_projection",
+                "answer_text": answer_text,
+                "raw_answer": answer_text,
+                "audit_mode": "CANONICAL_PROJECTION",
+                "verifier_method": "canonical_projection",
+                "n_quotes": 0, "n_verified": 0, "violations": [],
+                "sources": [],
+                "pi_star_ref": pi_star_ref,
+                "capacity": {},
+                "timings": {"total": _time.time() - t_total},
+                "model": model_id,
+                "cache_key": "",
+                "lookup_path": "preflight_canonical",
+                "burned_existing": 0,
+                "audit_event_hash": None,
+                "run_dag_root": None,
+                "context_root": "",
+                "prompt_chars": {
+                    "messages_total": 0, "system_prompt": 0,
+                    "evidence_or_context": 0, "user_question": len(question),
+                    "grounding_reminder": 0,
+                },
+                "answer_chars": len(answer_text),
+                "elapsed_s": round(_time.time() - t_total, 3),
+                "unverified_quotes": [],
+                "partially_verified_quotes": [],
+                "warrant_proven_claim_idxs": [],
+                "format_collapsed": None,
+            }
+
     # 1. RETRIEVAL ONLY — fast phase. Body + title FTS5 in parallel,
     # merge by best bm25 per doc_root. Body alone misses canonical
     # primary articles when high-DF tokens dilute the match (e.g.