fix: enforce deterministic JSON depth limits across runtimes

Author: KIM3310

Makefile

@@ -1,3 +1,4 @@
+.SHELLFLAGS := -eu -c
 PYTHON ?= python3
 VENV ?= .venv
 BIN := $(VENV)/bin

README.md

@@ -1,5 +1,7 @@
 # MemoryFlow Lab
 
+[Engineering notes](docs/engineering-notes.md) · [Core model](src/memoryflow/simulator.py) · [Reproduction protocol](#reproduce)
+
 [![CI](https://github.com/KIM3310/memoryflow-lab/actions/workflows/ci.yml/badge.svg)](https://github.com/KIM3310/memoryflow-lab/actions/workflows/ci.yml)
 [![Live results](https://img.shields.io/badge/live-results-0d6447)](https://kim3310.github.io/memoryflow-lab/)
 [![Python 3.11+](https://img.shields.io/badge/python-3.11%2B-3776ab)](https://www.python.org/)

docs/engineering-notes.md

@@ -0,0 +1,9 @@
+# A stable JSON input boundary across Python versions
+
+The input loader relied on the JSON decoder raising `RecursionError` for excessively nested documents. On this machine's Python 3.12 runtime, the 2,000-level regression input was accepted by the decoder, making both CLI and API behavior depend on the interpreter.
+
+The loader now performs a bounded, non-recursive nesting scan before decoding, with an explicit maximum of 64 object/array levels. Quoted braces and escaped quotes do not count as containers. Existing byte-size, duplicate-key, finite-number, Unicode, and scenario-schema checks remain in place.
+
+`tests/test_json_depth_contract.py` checks the exact boundary, quoted syntax, and rejection before invoking the decoder. `make verify` also regenerates and compares simulation evidence, checks measurement artifacts, and builds and validates the source distribution and wheel. The Makefile now fails immediately when either deterministic comparison fails.
+
+Committed simulation evidence was regenerated from the changed source. The pre-existing MPS measurement files were not remeasured or relabeled; they remain aggregate summaries of the original limited experiments.

evidence/benchmark-summary.md

@@ -1,7 +1,7 @@
 # Reproducible Benchmark Summary
 
 Scenario-set SHA-256: `6c494ca01417e6b51ee6aec7359b084d57e17c1cca8e8b9b48347cded529e302`
-Model-source SHA-256: `b2ff2bfb644c3aca0a1307731b046add648dd46a6b8c5f3b1c44c06bae429a7c`
+Model-source SHA-256: `eff89ab1a969bb96a3739ed950b36d55213884d224344b75d46da29b845ff6a4`
 Generator SHA-256: `dd9d905137eba4496e3865caa8dbc167703c286f165fd95f2a4a38aafb6cbfc8`
 
 All policy results below use bundled synthetic hardware knobs. They are deterministic first-order estimates, not measurements or product claims.

site/results.json

@@ -1394,15 +1394,15 @@
     "command": "python -m scripts.build_evidence",
     "generator_path": "scripts/build_evidence.py",
     "generator_sha256": "dd9d905137eba4496e3865caa8dbc167703c286f165fd95f2a4a38aafb6cbfc8",
-    "model_source_sha256": "b2ff2bfb644c3aca0a1307731b046add648dd46a6b8c5f3b1c44c06bae429a7c",
+    "model_source_sha256": "eff89ab1a969bb96a3739ed950b36d55213884d224344b75d46da29b845ff6a4",
     "model_sources": [
       {
         "path": "src/memoryflow/domain.py",
         "sha256": "2eaecabd38fe7a7c16c9d4eb97eecfd7f9d2930418ed4ebb4594a6bbc7e7b72f"
       },
       {
         "path": "src/memoryflow/io.py",
-        "sha256": "5ea2f52a00f46bdbb2a20c09ff1001d8a7f53ad597743ef66450d8d755ab0f55"
+        "sha256": "5bb80100a2daa56ea17bab78300829e5ef2825324203d21bd42e1e6aace36571"
       },
       {
         "path": "src/memoryflow/simulator.py",

src/memoryflow/io.py

@@ -16,6 +16,7 @@
 
 SCENARIO_SCHEMA_VERSION = "2.0"
 MAX_SCENARIO_BYTES = 1_048_576
+MAX_JSON_NESTING = 64
 T = TypeVar("T", Workload, MemorySystem, PlacementPolicy, ScenarioProvenance)
 
 
@@ -101,6 +102,29 @@ def _validate_parsed_scalars(payload: object) -> None:
             pending.extend(value)
 
 
+def _validate_json_nesting(text: str) -> None:
+    """Bound container nesting before the interpreter-dependent JSON decoder."""
+    depth = 0
+    in_string = False
+    escaped = False
+    for char in text:
+        if in_string:
+            if escaped:
+                escaped = False
+            elif char == "\\":
+                escaped = True
+            elif char == '"':
+                in_string = False
+        elif char == '"':
+            in_string = True
+        elif char in "[{":
+            depth += 1
+            if depth > MAX_JSON_NESTING:
+                raise ValueError("JSON nesting exceeds the parser limit")
+        elif char in "]}":
+            depth -= 1
+
+
 def json_object_from_bytes(
     content: bytes, *, maximum_bytes: int = MAX_SCENARIO_BYTES
 ) -> dict[str, Any]:
@@ -110,6 +134,7 @@ def json_object_from_bytes(
         text = content.decode("utf-8")
     except UnicodeDecodeError as exc:
         raise ValueError("JSON input must use UTF-8 encoding") from exc
+    _validate_json_nesting(text)
     try:
         payload = json.loads(
             text,

tests/test_json_depth_contract.py

@@ -0,0 +1,30 @@
+import json
+from unittest.mock import patch
+
+import pytest
+
+from memoryflow.io import MAX_JSON_NESTING, json_object_from_bytes
+
+
+def nested_arrays(depth: int) -> bytes:
+    return b'{"x":' + b"[" * (depth - 1) + b"0" + b"]" * (depth - 1) + b"}"
+
+
+def test_exact_container_depth_boundary() -> None:
+    assert "x" in json_object_from_bytes(nested_arrays(MAX_JSON_NESTING))
+    with pytest.raises(ValueError, match="nesting exceeds"):
+        json_object_from_bytes(nested_arrays(MAX_JSON_NESTING + 1))
+
+
+def test_quoted_braces_and_escaped_quotes_do_not_count_as_containers() -> None:
+    payload = {"text": '[{\\"' * 1000 + "}]" * 1000}
+    assert json_object_from_bytes(json.dumps(payload).encode()) == payload
+
+
+def test_excessive_depth_is_rejected_before_json_decode() -> None:
+    with (
+        patch("memoryflow.io.json.loads") as decoder,
+        pytest.raises(ValueError, match="nesting exceeds"),
+    ):
+        json_object_from_bytes(nested_arrays(2000))
+    decoder.assert_not_called()