diff --git a/docs/development/testing-and-quality.md b/docs/development/testing-and-quality.md index 911c65c432..5559630fbd 100644 --- a/docs/development/testing-and-quality.md +++ b/docs/development/testing-and-quality.md @@ -1267,6 +1267,26 @@ TS RFC 管语义 owner 与跨语言成本，shared-authority RFC 管后端容量 ### Budget Failure Decisions +The actual-default model portfolio retains the complete capability candidate +context required by the shared diagnostic projection. On the unchanged three +noisy fixtures, `74b3d526ab1c1fd3d5f96308bef2482fcad7e636` emitted 4,944 / +4,858 / 3,089 pretty-JSON UTF-8 bytes; `53dcb03082985bf894b1fca8813ab9f6e92503c5` +emits 86,172 / 86,086 / 84,276. The sources remain 120,963 / 121,008 / 118,935 +bytes with the same decision semantics. The new projection preserves all 24 +runnable, 16 blocked and 16 resolution-binding candidates, including their full +text; the old projection removed those bodies. Calibrate only this synthetic +fixture allowance from 40,000 to 90,000 bytes, preserving the over-budget source, +selected-Todo/human-gate invariance, cold diagnostic checks, 22 scenarios, two +repeats and six contrasts. Candidate removal or text loss now fails preflight +before any model call. The original 40,000-byte failures remain failures; this +does not change production transport limits, execution/tool budgets or frozen +performance criteria, and does not qualify a model by itself. + +真实模型组合门的同一组三个 noisy fixture 现保留完整的 56 个 capability 候选及其 +正文。仅按上述实测把该合成 fixture 的 pretty-JSON UTF-8 字节预算从 40,000 调整到 +90,000；原超限失败保留。新增候选丢失/正文缩短的零模型调用负例，原选择、人工门禁、 +结算、重复次数、对照及生产硬边界不变；本次预算修正不等于真实模型通过。 + Complete registry routes exposed an existing brief heartbeat regression on the unchanged 128-character-root fixture. Commit `fb20c153ba2c2a7376efe68091a76ca470effbde` added those routes: its immediate parent diff --git a/loopx/control_plane/testing/actual_default_model_behavior_portfolio.py b/loopx/control_plane/testing/actual_default_model_behavior_portfolio.py index fd5a6b0d14..20593fd21b 100644 --- a/loopx/control_plane/testing/actual_default_model_behavior_portfolio.py +++ b/loopx/control_plane/testing/actual_default_model_behavior_portfolio.py @@ -64,7 +64,8 @@ "actual_default_model_behavior_scenario_catalog_v0" ) ACTUAL_DEFAULT_MODEL_BEHAVIOR_REPEAT_ATTEMPTS = 2 -ACTUAL_DEFAULT_MODEL_BEHAVIOR_HOT_PATH_JSON_BUDGET = 40_000 +# Synthetic pretty-JSON regression allowance, not a production transport limit. +ACTUAL_DEFAULT_MODEL_BEHAVIOR_HOT_PATH_JSON_BUDGET = 90_000 ACTUAL_DEFAULT_MODEL_BEHAVIOR_CONTRAST_SCHEMA_VERSION = ( "actual_default_model_behavior_contrast_v0" ) @@ -840,6 +841,8 @@ def _validate_quota_hot_path_compaction_regression( raise ValueError("compaction-regression source must exceed the hot-path budget") if _pretty_json_size(packet) > ACTUAL_DEFAULT_MODEL_BEHAVIOR_HOT_PATH_JSON_BUDGET: raise ValueError("compaction-regression packet exceeds the hot-path budget") + if packet.get("capability_gate") != source.get("capability_gate"): + raise ValueError("compaction regression must preserve complete capability candidate context") if contract.get("selected_todo_id") != expected_selected_todo_id: raise ValueError("compaction regression must preserve the selected todo") diff --git a/tests/control_plane/test_actual_default_model_behavior_portfolio.py b/tests/control_plane/test_actual_default_model_behavior_portfolio.py index d54e44fe7c..16a03ba944 100644 --- a/tests/control_plane/test_actual_default_model_behavior_portfolio.py +++ b/tests/control_plane/test_actual_default_model_behavior_portfolio.py @@ -927,9 +927,7 @@ def test_live_packet_builder_preserves_gate_and_vision_contracts(tmp_path: Path) assert regression_signature["action"]["selected_todo"]["todo_id"] == ( "todo_c0ffee123456" ) - assert regression["capability_gate"]["payload_compaction"][ - "omitted_candidate_counts" - ] == {"runnable": 24, "blocked": 16, "resolution_bindings": 16} + assert regression["capability_gate"] == regression_source["capability_gate"] assert set(regression["next_action_projection_warning"]["projection_refs"]) == { "active_state_next_action", "latest_run_recommended_action", @@ -957,6 +955,7 @@ def test_live_packet_builder_preserves_gate_and_vision_contracts(tmp_path: Path) "external_write_requested", ) for clean_id, noisy_id in invariant_pairs: + assert packets[noisy_id]["capability_gate"] == sources[noisy_id]["capability_gate"] assert ( len( json.dumps(sources[noisy_id], ensure_ascii=False, indent=2).encode( @@ -1198,6 +1197,33 @@ def turn_actor(request: Mapping[str, Any]) -> Mapping[str, Any]: assert calls == 0 +@pytest.mark.parametrize("lane", ["runnable_candidates", "blocked_candidates", "resolution_bindings"]) +@pytest.mark.parametrize("mutation", ["drop", "text"]) +def test_portfolio_rejects_candidate_context_loss_before_actor_call( + tmp_path: Path, lane: str, mutation: str, +) -> None: + sources, packets = _scenario_inputs(tmp_path) + candidates = packets["turn_quota_hot_path_compaction_regression"]["capability_gate"][lane] + if mutation == "drop": + candidates.pop() + else: + candidates[0]["text"] = "Shortened candidate context." + + def unexpected_actor(*_args: Any, **_kwargs: Any) -> Mapping[str, Any]: + pytest.fail("candidate context must be validated before any actor call") + + with pytest.raises(ValueError, match="must preserve complete capability candidate context"): + run_actual_default_model_behavior_portfolio( + packets, + scenario_sources=sources, + qualification_id="actual-default-candidate-context-loss", + turn_actor=unexpected_actor, + onboarding_actor=unexpected_actor, + selected_todo_actor=_selected_todo_actor, + replan_semantic_action_actor=_replan_semantic_action_actor, + ) + + def test_portfolio_source_oracle_rejects_mutated_compact_user_action( tmp_path: Path, ) -> None: @@ -1728,7 +1754,7 @@ def collapsed_actor(request: Mapping[str, Any]) -> dict[str, Any]: capability_gate = packet.get("capability_gate") or {} if ( capability_gate.get("action") == "operator_gate" - and "payload_compaction" in capability_gate + and capability_gate.get("runnable_candidates") ): result["decision"] = { **result["decision"],