test(core): pin inference level in semantic preservation fixtures

Author: forhappy

benchmarks/agent_query/core_inference_fixture_review.json

@@ -0,0 +1,80 @@
+{
+  "afterFixtureSha256": "c34fa0bbaa7fc5a64d53fd142c34462c5843c0fe5e9f6fb1825248fdc3668ed9",
+  "afterSourceManifestSha256": "0c1cc9821267491f263ea22a84a92730e5b8658d96bcb05116cc978fa815c417",
+  "beforeAfterBaselineComparisonSha256": "20b9f219444095ceac3075716cf849ca8250632158cf68fd7935f213a27c3b6f",
+  "beforeFixtureSha256": "7fae4ec71c4fa1fbf06b77ee8a780f46efaf2c8d9e63c9be9ec5b34d4cd4cdcf",
+  "changedRustAndManifestInputs": [
+    "crates/compass-core/tests/loading_coverage.rs"
+  ],
+  "checks": [
+    {
+      "exitCode": 0,
+      "name": "fmt",
+      "stderrSha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
+      "stdoutSha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
+      "wallMs": 4406
+    },
+    {
+      "exitCode": 0,
+      "name": "focused-clippy",
+      "stderrSha256": "223b1728b76c2132cd2612388bdf9a7487f5f2945b9b5aaa0cc3531560bbc991",
+      "stdoutSha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
+      "wallMs": 93393
+    },
+    {
+      "exitCode": 0,
+      "name": "workspace-clippy",
+      "stderrSha256": "9564857d44c7db3de9b21adfd6f085f6306eead154431e5aab786e7e459eeb22",
+      "stdoutSha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
+      "wallMs": 1849
+    },
+    {
+      "exitCode": 0,
+      "name": "workspace-tests",
+      "stderrSha256": "ec94f0c63096545b2c68db32e63b494b7de4134f1fa41e75cd0c7e927f4f460d",
+      "stdoutSha256": "b3063489c879d36b558f1acefb9b9654590ccb5128edfa81ff759aa9734f1ea0",
+      "wallMs": 61130
+    },
+    {
+      "exitCode": 0,
+      "name": "core-integration",
+      "stderrSha256": "a1ba9c61977fda16c3fa5e0a52d349d9d2dbab0b4cc0cefc78c0e83d146f1ad2",
+      "stdoutSha256": "30b5fe8459b07bbfe6b410549164633eef275604ec4034547f4af4ab2ab0d3aa",
+      "wallMs": 14202
+    }
+  ],
+  "fixturePatchSha256": "917897cde35b4fbae5b7a3e8400395401a4fdb02ba22c58012472a4feaa3337b",
+  "limitations": [
+    "This fixture correction does not demonstrate a product behavior improvement or change the original baseline failures.",
+    "The preserved cache query and Graphify comparison results are unaffected; actual god-object diagnosis remains unproven."
+  ],
+  "newPassedTests": [
+    "loaded_graph_learning_overlay_marks_current_missing_and_unfingerprinted_sources",
+    "export_inputs_fall_back_to_node_communities_and_tolerate_partial_sidecars",
+    "build_pipeline_reports_missing_and_empty_roots_and_accepts_file_only_sources",
+    "semantic_build_normalizes_typed_origins_paths_and_incremental_updates",
+    "incremental_mixed_origin_edges_use_fresh_ast_relationship_sites",
+    "incremental_mixed_occurrence_cardinality_matches_exact_sites_and_preserves_residue",
+    "incremental_update_preserves_then_replaces_owned_semantic_facts",
+    "incremental_deleted_mixed_occurrence_keeps_only_revalidated_semantic_evidence",
+    "incremental_mixed_origin_alias_edges_use_canonical_fresh_relationship_sites",
+    "incremental_deleted_remapped_mixed_occurrence_rebinds_trusted_semantic_residue",
+    "incremental_mixed_origin_nodes_use_fresh_ast_typed_data",
+    "refreshed_mixed_edge_drops_stale_incremental_endpoint_remap_evidence",
+    "incremental_ast_endpoint_remap_retains_exact_typed_rewrite_evidence"
+  ],
+  "previousBaselineFailedTests": [
+    "incremental_ast_endpoint_remap_retains_exact_typed_rewrite_evidence",
+    "incremental_deleted_mixed_occurrence_keeps_only_revalidated_semantic_evidence",
+    "incremental_deleted_remapped_mixed_occurrence_rebinds_trusted_semantic_residue",
+    "incremental_mixed_occurrence_cardinality_matches_exact_sites_and_preserves_residue",
+    "incremental_mixed_origin_alias_edges_use_canonical_fresh_relationship_sites",
+    "incremental_mixed_origin_edges_use_fresh_ast_relationship_sites",
+    "refreshed_mixed_edge_drops_stale_incremental_endpoint_remap_evidence"
+  ],
+  "priorProductionQualificationSha256": "14250b55c069e69bd107bf0a1f6aac61fe95498c0ea7ecfa761cfd0592b20d43",
+  "qualificationReportSha256": "d7fcd7ac606a27c06f3406334fe9a69a2887b45e6dad05b7fe69f1a3a0c40e2c",
+  "schema": "compass.core-inference-fixture-review/1",
+  "scope": "A test-only correction: seven semantic-preservation fixtures explicitly use Max inference and keep their assertions. The public default remains Low; production behavior and frozen cache-evaluation binary are unchanged.",
+  "unchangedProductionBinarySha256": "03403b96658900adad039db08b5814a5b5d090075fe2a31dc8bebbe7c012aa1b"
+}

crates/compass-core/tests/loading_coverage.rs

@@ -4,8 +4,8 @@ use std::fs;
 use std::path::Path;
 
 use compass_core::{
-    BuildOptions, BuildPurpose, CoreError, ExportInputs, LoadedGraph, SemanticLayer,
-    build_graph_with_layers,
+    BuildOptions, BuildPurpose, CoreError, ExportInputs, InferenceLevel, LoadedGraph,
+    SemanticLayer, build_graph_with_layers,
 };
 use compass_files::BuildGuard;
 use compass_languages::{Engine, Extraction, RawEdgeRecord, RawNodeRecord};
@@ -476,6 +476,7 @@ fn incremental_mixed_origin_edges_use_fresh_ast_relationship_sites() -> Result<(
     )?;
 
     let mut options = BuildOptions::new(root.to_path_buf());
+    options.inference_level = InferenceLevel::Max;
     options.no_cluster = true;
     options.no_viz = true;
     let initial = build_graph_with_layers(&options, None, &[])?;
@@ -545,6 +546,7 @@ fn incremental_deleted_mixed_occurrence_keeps_only_revalidated_semantic_evidence
         semantic_owned_edge_extraction(Path::new("main.rs"), initial_source, "calls")?;
 
     let mut options = BuildOptions::new(root.to_path_buf());
+    options.inference_level = InferenceLevel::Max;
     options.no_cluster = true;
     options.no_viz = true;
     let initial_layer = semantic_layer(initial_semantic)?;
@@ -617,6 +619,7 @@ fn incremental_deleted_mixed_occurrence_keeps_only_revalidated_semantic_evidence
         evidence.origin == compass_model::provenance::EvidenceOrigin::Heuristic
     }));
     let mut clean_options = BuildOptions::new(root.to_path_buf());
+    clean_options.inference_level = InferenceLevel::Max;
     clean_options.output_root = Some(root.join("clean-deletion-out"));
     clean_options.no_cluster = true;
     clean_options.no_viz = true;
@@ -645,6 +648,7 @@ fn incremental_deleted_remapped_mixed_occurrence_rebinds_trusted_semantic_residu
         semantic_owned_edge_extraction(Path::new("main.rs"), initial_source, "calls")?;
 
     let mut options = BuildOptions::new(root.to_path_buf());
+    options.inference_level = InferenceLevel::Max;
     options.no_cluster = true;
     options.no_viz = true;
     let initial_layer = semantic_layer(initial_semantic)?;
@@ -756,6 +760,7 @@ fn incremental_deleted_remapped_mixed_occurrence_rebinds_trusted_semantic_residu
     assert_eq!(trusted.target, canonical_target);
 
     let mut clean_options = BuildOptions::new(root.to_path_buf());
+    clean_options.inference_level = InferenceLevel::Max;
     clean_options.output_root = Some(root.join("clean-deletion-remap-out"));
     clean_options.no_cluster = true;
     clean_options.no_viz = true;
@@ -789,6 +794,7 @@ fn incremental_mixed_occurrence_cardinality_matches_exact_sites_and_preserves_re
     assert_eq!(initial_semantic.edges.len(), 3);
 
     let mut options = BuildOptions::new(root.to_path_buf());
+    options.inference_level = InferenceLevel::Max;
     options.no_cluster = true;
     options.no_viz = true;
     let initial_layer = semantic_layer(initial_semantic)?;
@@ -959,6 +965,7 @@ fn incremental_mixed_occurrence_cardinality_matches_exact_sites_and_preserves_re
     );
 
     let mut clean_options = BuildOptions::new(root.to_path_buf());
+    clean_options.inference_level = InferenceLevel::Max;
     clean_options.output_root = Some(root.join("clean-cardinality-out"));
     clean_options.no_cluster = true;
     clean_options.no_viz = true;
@@ -987,6 +994,7 @@ fn incremental_mixed_origin_alias_edges_use_canonical_fresh_relationship_sites()
     assert_eq!(initial_semantic.edges.len(), 1);
 
     let mut options = BuildOptions::new(root.to_path_buf());
+    options.inference_level = InferenceLevel::Max;
     options.no_cluster = true;
     options.no_viz = true;
     let initial_layer = semantic_layer(initial_semantic)?;
@@ -1033,6 +1041,7 @@ fn incremental_mixed_origin_alias_edges_use_canonical_fresh_relationship_sites()
     let final_semantic = semantic_method_alias_extraction(Path::new("main.rs"), final_source)?;
     assert_eq!(final_semantic.edges.len(), 1);
     let mut clean_options = BuildOptions::new(root.to_path_buf());
+    clean_options.inference_level = InferenceLevel::Max;
     clean_options.output_root = Some(root.join("clean-out"));
     clean_options.no_cluster = true;
     clean_options.no_viz = true;
@@ -1058,6 +1067,7 @@ fn refreshed_mixed_edge_drops_stale_incremental_endpoint_remap_evidence()
     assert_eq!(initial_semantic.edges.len(), 1);
 
     let mut options = BuildOptions::new(root.to_path_buf());
+    options.inference_level = InferenceLevel::Max;
     options.no_cluster = true;
     options.no_viz = true;
     let initial_layer = semantic_layer(initial_semantic)?;
@@ -1127,6 +1137,7 @@ fn refreshed_mixed_edge_drops_stale_incremental_endpoint_remap_evidence()
 
     let final_semantic = semantic_edge_extraction(Path::new("main.rs"), final_source, "calls")?;
     let mut clean_options = BuildOptions::new(root.to_path_buf());
+    clean_options.inference_level = InferenceLevel::Max;
     clean_options.output_root = Some(root.join("clean-remap-out"));
     clean_options.no_cluster = true;
     clean_options.no_viz = true;
@@ -1220,6 +1231,7 @@ fn incremental_ast_endpoint_remap_retains_exact_typed_rewrite_evidence()
         ..Extraction::default()
     };
     let mut options = BuildOptions::new(root.to_path_buf());
+    options.inference_level = InferenceLevel::Max;
     options.no_cluster = true;
     options.no_viz = true;
 

docs/implementation/code-graph-intelligence-audit-2026-09-26.md

@@ -4238,3 +4238,22 @@ The correction improves a verified freshness failure and preserves all
 available fixed query outputs, while adding memory and lightweight-query cost.
 It does not establish better authored explanations, functional communities,
 actual god-object classification, or overall superiority to Graphify.
+
+### Core integration fixture inference policy
+
+The seven core integration failures in the cache review reproduce on the
+unchanged committed baseline. Their fixtures expect inferred semantic residue,
+but create `BuildOptions` without an inference level. The public default changed
+from `Max` to `Low` earlier; `Low` intentionally excludes inferred edges, so the
+fixtures were not exercising their stated semantic-preservation scenario. The
+test-only follow-up selects `Max` explicitly in seven tests and all 12 of their
+build configurations. It leaves every assertion unchanged and does not change
+the public `Low` default or production binary.
+
+All 13 tests in `loading_coverage` now pass under the pinned offline parser
+setup. Formatting, focused core integration Clippy, workspace library/binary
+Clippy, and 1,112 workspace library/binary tests (two existing ignored) pass.
+The [fixture review](../../benchmarks/agent_query/core_inference_fixture_review.json)
+binds the original seven failures, the sole changed Rust test file, the frozen
+production binary, and the before/after native logs. This restores the intended
+regression coverage; it does not prove any new graph query or god-object quality.