From 8462afa56fe52377e96a497d596072425d0d696d Mon Sep 17 00:00:00 2001
From: Gahow Wang <gahow.wang@gmail.com>
Date: Fri, 24 Jul 2026 00:40:15 +0800
Subject: [PATCH] Encode zero cacheable blocks explicitly

---
 runs/frontier-s3-real-v0/remap_hash_blocks.py |  9 +++-
 .../test_remap_hash_blocks.py                 | 43 +++++++++++++++++++
 2 files changed, 50 insertions(+), 2 deletions(-)

diff --git a/runs/frontier-s3-real-v0/remap_hash_blocks.py b/runs/frontier-s3-real-v0/remap_hash_blocks.py
index a52973b..0628cc8 100644
--- a/runs/frontier-s3-real-v0/remap_hash_blocks.py
+++ b/runs/frontier-s3-real-v0/remap_hash_blocks.py
@@ -325,8 +325,13 @@ def materialize(args: argparse.Namespace, *, tokenizer: Any | None = None) -> di
                     "num_prefill_tokens": isl,
                     "num_decode_tokens": osl,
                     "session_id": sid,
-                    "block_hash_ids": "|".join(
-                        str(value) for value in cacheable_block_ids
+                    # "[]" is the explicit no-cacheable-block encoding accepted
+                    # by Frontier. An empty CSV field is treated as missing
+                    # metadata and rejected before scheduling.
+                    "block_hash_ids": (
+                        "|".join(str(value) for value in cacheable_block_ids)
+                        if cacheable_block_ids
+                        else "[]"
                     ),
                 }
             )
diff --git a/runs/frontier-s3-real-v0/test_remap_hash_blocks.py b/runs/frontier-s3-real-v0/test_remap_hash_blocks.py
index 55e8f50..510c14f 100644
--- a/runs/frontier-s3-real-v0/test_remap_hash_blocks.py
+++ b/runs/frontier-s3-real-v0/test_remap_hash_blocks.py
@@ -141,6 +141,49 @@ class StrictRemapTest(unittest.TestCase):
             len(frontier_row["block_hash_ids"].split("|")), math.floor(513 / 16)
         )
 
+    def test_sub_block_prompt_uses_explicit_empty_cache_sequence(self) -> None:
+        with tempfile.TemporaryDirectory() as tmp:
+            root = Path(tmp)
+            trace = root / "short.jsonl"
+            trace.write_text(
+                json.dumps(
+                    {
+                        "chat_id": "short",
+                        "parent_chat_id": -1,
+                        "timestamp": 1.0,
+                        "input_length": 7,
+                        "output_length": 1,
+                        "hash_ids": [101],
+                        "sampling_u": 0.1,
+                    }
+                )
+                + "\n"
+            )
+            output = root / "mapped"
+            args = SimpleNamespace(
+                input=trace,
+                prompt=None,
+                tokenizer=None,
+                input_is_remapped=False,
+                output_root=output,
+                rho=1.0,
+                start_timestamp=None,
+                duration_s=None,
+                vocab_size=151936,
+                token_offset=1024,
+                served_model="test-model",
+                source_block_size=512,
+                workload_mode="prefill_decode",
+                validate_parents=False,
+                max_total_tokens=None,
+                frontier_only=True,
+            )
+            materialize(args)
+            with (output / "frontier.csv").open(newline="") as stream:
+                frontier_row = next(csv.DictReader(stream))
+
+        self.assertEqual(frontier_row["block_hash_ids"], "[]")
+
     def test_real_prompt_tokens_preserve_parent_and_four_to_one_contract(self) -> None:
         tokenizer = CharacterTokenizer()
         parent_prompt = "A" * 64