diff --git a/tests/khive-contract/fixtures/memories_corpus_v2.json b/tests/khive-contract/fixtures/memories_corpus_v2.json
new file mode 100644
index 00000000..ed6212e5
--- /dev/null
+++ b/tests/khive-contract/fixtures/memories_corpus_v2.json
@@ -0,0 +1,2239 @@
+{
+  "_meta": {
+    "version": "v2",
+    "description": "Harder eval corpus for recall tuning. Synonym queries, partial matches, distractors, 5+ domains, ground-truth expected_top_k. Includes importance_trap queries to discriminate importance_weight parameter.",
+    "memory_count": 203,
+    "query_count": 48,
+    "domains": ["ml", "biology", "programming", "history", "geography", "cooking", "medicine", "physics", "economics", "linguistics"],
+    "distractor_count": 46,
+    "distractor_ratio": "~22.7% (46 of 203 memories have no query relevance)",
+    "query_types": {
+      "synonym": "queries using synonyms not present verbatim in memory content",
+      "partial": "queries where only some terms appear in target memory",
+      "importance_trap": "queries with unique terms matching low-importance memories; discriminates importance_weight parameter"
+    }
+  },
+  "memories": [
+    {
+      "id": "mem_001",
+      "content": "Backpropagation computes gradients by applying the chain rule through each layer of a neural network, enabling weight updates via gradient descent",
+      "domain": "ml",
+      "importance": 0.90,
+      "decay_factor": 0.01,
+      "memory_type": "semantic",
+      "tags": ["ml", "neural-networks", "optimization"]
+    },
+    {
+      "id": "mem_002",
+      "content": "Stochastic gradient descent updates model parameters using the gradient computed on a random mini-batch rather than the full training dataset",
+      "domain": "ml",
+      "importance": 0.88,
+      "decay_factor": 0.01,
+      "memory_type": "semantic",
+      "tags": ["ml", "optimization"]
+    },
+    {
+      "id": "mem_003",
+      "content": "Adam optimizer combines momentum and adaptive learning rates for each parameter, converging faster than vanilla SGD on most deep learning tasks",
+      "domain": "ml",
+      "importance": 0.87,
+      "decay_factor": 0.01,
+      "memory_type": "semantic",
+      "tags": ["ml", "optimization"]
+    },
+    {
+      "id": "mem_004",
+      "content": "Attention mechanisms allow neural networks to selectively focus on relevant parts of the input sequence when producing each output token",
+      "domain": "ml",
+      "importance": 0.92,
+      "decay_factor": 0.01,
+      "memory_type": "semantic",
+      "tags": ["ml", "transformers", "attention"]
+    },
+    {
+      "id": "mem_005",
+      "content": "The transformer architecture uses multi-head self-attention and feedforward layers to process sequences without recurrence, enabling parallelization",
+      "domain": "ml",
+      "importance": 0.95,
+      "decay_factor": 0.01,
+      "memory_type": "semantic",
+      "tags": ["ml", "transformers"]
+    },
+    {
+      "id": "mem_006",
+      "content": "Dropout regularization randomly sets a fraction of activations to zero during training to reduce overfitting in deep neural networks",
+      "domain": "ml",
+      "importance": 0.82,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["ml", "regularization"]
+    },
+    {
+      "id": "mem_007",
+      "content": "Batch normalization normalizes layer activations to have zero mean and unit variance, stabilizing training and enabling higher learning rates",
+      "domain": "ml",
+      "importance": 0.83,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["ml", "normalization"]
+    },
+    {
+      "id": "mem_008",
+      "content": "Convolutional neural networks use learnable filters that slide over spatial input to detect local patterns, forming the backbone of computer vision models",
+      "domain": "ml",
+      "importance": 0.90,
+      "decay_factor": 0.01,
+      "memory_type": "semantic",
+      "tags": ["ml", "cnn", "computer-vision"]
+    },
+    {
+      "id": "mem_009",
+      "content": "Recurrent neural networks maintain hidden state across time steps to model sequential dependencies in variable-length input sequences",
+      "domain": "ml",
+      "importance": 0.85,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["ml", "rnn", "sequence"]
+    },
+    {
+      "id": "mem_010",
+      "content": "Long short-term memory networks solve the vanishing gradient problem in RNNs by using gating mechanisms to control information flow over long sequences",
+      "domain": "ml",
+      "importance": 0.87,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["ml", "lstm", "sequence"]
+    },
+    {
+      "id": "mem_011",
+      "content": "Transfer learning fine-tunes a pretrained model on a new task, leveraging representations learned from a large corpus to achieve better performance with less data",
+      "domain": "ml",
+      "importance": 0.88,
+      "decay_factor": 0.01,
+      "memory_type": "semantic",
+      "tags": ["ml", "transfer-learning"]
+    },
+    {
+      "id": "mem_012",
+      "content": "Embedding layers map discrete tokens to dense vectors in a continuous space, capturing semantic relationships between words or categories",
+      "domain": "ml",
+      "importance": 0.85,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["ml", "embeddings", "nlp"]
+    },
+    {
+      "id": "mem_013",
+      "content": "Reinforcement learning trains agents by rewarding desired behaviors and penalizing undesired ones through interaction with an environment",
+      "domain": "ml",
+      "importance": 0.87,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["ml", "reinforcement-learning"]
+    },
+    {
+      "id": "mem_014",
+      "content": "Cross-entropy loss measures the divergence between predicted probability distributions and true labels, commonly used for classification tasks",
+      "domain": "ml",
+      "importance": 0.83,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["ml", "loss-functions"]
+    },
+    {
+      "id": "mem_015",
+      "content": "Hyperparameter tuning searches over learning rate, batch size, and network architecture to find configurations that maximize validation performance",
+      "domain": "ml",
+      "importance": 0.80,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["ml", "hyperparameters"]
+    },
+    {
+      "id": "mem_016",
+      "content": "Mitochondria are the powerhouse of the cell, generating ATP through oxidative phosphorylation in the electron transport chain",
+      "domain": "biology",
+      "importance": 0.85,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["biology", "cell", "energy"]
+    },
+    {
+      "id": "mem_017",
+      "content": "Cellular respiration converts glucose and oxygen into carbon dioxide, water, and ATP energy through glycolysis, the Krebs cycle, and oxidative phosphorylation",
+      "domain": "biology",
+      "importance": 0.88,
+      "decay_factor": 0.01,
+      "memory_type": "semantic",
+      "tags": ["biology", "metabolism", "energy"]
+    },
+    {
+      "id": "mem_018",
+      "content": "Messenger RNA carries the genetic blueprint from DNA in the nucleus to ribosomes in the cytoplasm where proteins are synthesized",
+      "domain": "biology",
+      "importance": 0.87,
+      "decay_factor": 0.01,
+      "memory_type": "semantic",
+      "tags": ["biology", "genetics", "protein-synthesis"]
+    },
+    {
+      "id": "mem_019",
+      "content": "Natural selection favors individuals with heritable traits that increase survival and reproductive success in a given environment",
+      "domain": "biology",
+      "importance": 0.88,
+      "decay_factor": 0.01,
+      "memory_type": "semantic",
+      "tags": ["biology", "evolution"]
+    },
+    {
+      "id": "mem_020",
+      "content": "Enzymes are biological catalysts that speed up chemical reactions by lowering activation energy without being consumed in the process",
+      "domain": "biology",
+      "importance": 0.85,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["biology", "biochemistry", "enzymes"]
+    },
+    {
+      "id": "mem_021",
+      "content": "The immune system distinguishes self from non-self through antigen presentation, T-cell activation, and antibody-mediated humoral immunity",
+      "domain": "biology",
+      "importance": 0.87,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["biology", "immunology"]
+    },
+    {
+      "id": "mem_022",
+      "content": "Stem cells are undifferentiated cells capable of self-renewal and differentiation into specialized cell types such as neurons, muscle, or blood cells",
+      "domain": "biology",
+      "importance": 0.83,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["biology", "stem-cells"]
+    },
+    {
+      "id": "mem_023",
+      "content": "The human genome contains approximately 3 billion base pairs encoding roughly 20,000 protein-coding genes distributed across 23 chromosome pairs",
+      "domain": "biology",
+      "importance": 0.85,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["biology", "genetics", "genome"]
+    },
+    {
+      "id": "mem_024",
+      "content": "Epigenetic modifications such as DNA methylation and histone acetylation regulate gene expression without altering the underlying DNA sequence",
+      "domain": "biology",
+      "importance": 0.83,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["biology", "epigenetics", "gene-regulation"]
+    },
+    {
+      "id": "mem_025",
+      "content": "Symbiosis describes the close ecological interaction between two species, which can be mutualistic, commensal, or parasitic in nature",
+      "domain": "biology",
+      "importance": 0.78,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["biology", "ecology"]
+    },
+    {
+      "id": "mem_026",
+      "content": "Memory allocators like jemalloc and tcmalloc manage heap fragmentation and thread-local caches to reduce allocation latency in concurrent programs",
+      "domain": "programming",
+      "importance": 0.82,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["programming", "systems", "memory"]
+    },
+    {
+      "id": "mem_027",
+      "content": "Async runtimes like Tokio schedule thousands of lightweight tasks onto a thread pool using cooperative multitasking and event-driven IO",
+      "domain": "programming",
+      "importance": 0.87,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["programming", "async", "rust"]
+    },
+    {
+      "id": "mem_028",
+      "content": "Type inference allows compilers to deduce the types of expressions automatically without requiring explicit annotations from the programmer",
+      "domain": "programming",
+      "importance": 0.80,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["programming", "type-systems"]
+    },
+    {
+      "id": "mem_029",
+      "content": "Monads in functional programming represent computations as values and allow sequencing effects such as IO, optional values, or error handling",
+      "domain": "programming",
+      "importance": 0.82,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["programming", "functional", "haskell"]
+    },
+    {
+      "id": "mem_030",
+      "content": "Lock-free data structures use atomic compare-and-swap operations instead of mutexes to allow concurrent access without blocking threads",
+      "domain": "programming",
+      "importance": 0.83,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["programming", "concurrency", "lock-free"]
+    },
+    {
+      "id": "mem_031",
+      "content": "Garbage collection automatically reclaims unreachable heap memory by tracing live object graphs from root pointers in languages like Java and Go",
+      "domain": "programming",
+      "importance": 0.83,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["programming", "memory", "gc"]
+    },
+    {
+      "id": "mem_032",
+      "content": "Compiler intermediate representations like LLVM IR abstract over machine architectures, enabling target-independent optimizations like dead code elimination",
+      "domain": "programming",
+      "importance": 0.82,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["programming", "compilers", "llvm"]
+    },
+    {
+      "id": "mem_033",
+      "content": "Cache coherence protocols like MESI ensure that processor caches remain consistent when multiple cores access shared memory locations",
+      "domain": "programming",
+      "importance": 0.80,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["programming", "systems", "cpu"]
+    },
+    {
+      "id": "mem_034",
+      "content": "B-tree indexes in relational databases provide O(log n) search, insert, and delete by maintaining sorted order across balanced tree nodes on disk",
+      "domain": "programming",
+      "importance": 0.83,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["programming", "databases", "indexes"]
+    },
+    {
+      "id": "mem_035",
+      "content": "Consensus algorithms like Raft elect a leader and replicate a log across a cluster to achieve fault-tolerant distributed agreement",
+      "domain": "programming",
+      "importance": 0.85,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["programming", "distributed-systems"]
+    },
+    {
+      "id": "mem_036",
+      "content": "Event sourcing stores all state changes as an immutable append-only log of events rather than mutable current state in a relational table",
+      "domain": "programming",
+      "importance": 0.82,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["programming", "architecture", "event-sourcing"]
+    },
+    {
+      "id": "mem_037",
+      "content": "The Peloponnesian War between Athens and Sparta lasted from 431 to 404 BC and ended with Athenian defeat after the Sicilian Expedition disaster",
+      "domain": "history",
+      "importance": 0.78,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["history", "greece", "ancient"]
+    },
+    {
+      "id": "mem_038",
+      "content": "The Byzantine Empire survived the fall of Rome for nearly a thousand years, preserving Greek and Roman culture in Constantinople until 1453",
+      "domain": "history",
+      "importance": 0.80,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["history", "byzantine", "medieval"]
+    },
+    {
+      "id": "mem_039",
+      "content": "The printing press invented by Gutenberg around 1440 enabled mass reproduction of texts, accelerating the Reformation and the Scientific Revolution",
+      "domain": "history",
+      "importance": 0.85,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["history", "technology", "renaissance"]
+    },
+    {
+      "id": "mem_040",
+      "content": "The Treaty of Westphalia in 1648 ended the Thirty Years War and established the modern system of sovereign nation-states in Europe",
+      "domain": "history",
+      "importance": 0.82,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["history", "europe", "diplomacy"]
+    },
+    {
+      "id": "mem_041",
+      "content": "The colonization of the Americas by European powers decimated indigenous populations through introduced diseases, conquest, and forced labor systems",
+      "domain": "history",
+      "importance": 0.83,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["history", "colonization", "americas"]
+    },
+    {
+      "id": "mem_042",
+      "content": "The Meiji Restoration of 1868 modernized Japan by abolishing the feudal samurai class and rapidly adopting Western technology and institutions",
+      "domain": "history",
+      "importance": 0.82,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["history", "japan", "modernization"]
+    },
+    {
+      "id": "mem_043",
+      "content": "The Cold War rivalry between the United States and Soviet Union drove the space race, proxy wars, and nuclear deterrence strategy from 1947 to 1991",
+      "domain": "history",
+      "importance": 0.85,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["history", "cold-war", "20th-century"]
+    },
+    {
+      "id": "mem_044",
+      "content": "The Berlin Wall construction in 1961 physically divided East and West Germany, becoming the defining symbol of Iron Curtain division in Europe",
+      "domain": "history",
+      "importance": 0.83,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["history", "cold-war", "germany"]
+    },
+    {
+      "id": "mem_045",
+      "content": "The Scramble for Africa between 1881 and 1914 resulted in European powers colonizing nearly the entire African continent for resources and strategic position",
+      "domain": "history",
+      "importance": 0.80,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["history", "colonization", "africa"]
+    },
+    {
+      "id": "mem_046",
+      "content": "The Sahara Desert is the world's largest hot desert spanning over 9 million square kilometers across Northern Africa with extreme temperature variation",
+      "domain": "geography",
+      "importance": 0.75,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["geography", "africa", "desert"]
+    },
+    {
+      "id": "mem_047",
+      "content": "The Amazon River basin contains the largest tropical rainforest on Earth, housing an estimated 10% of all species on the planet",
+      "domain": "geography",
+      "importance": 0.82,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["geography", "south-america", "ecosystem"]
+    },
+    {
+      "id": "mem_048",
+      "content": "Tectonic plates move at centimeter-per-year rates driven by convection currents in the mantle, causing earthquakes, volcanoes, and mountain formation",
+      "domain": "geography",
+      "importance": 0.83,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["geography", "geology", "tectonics"]
+    },
+    {
+      "id": "mem_049",
+      "content": "The Himalayan mountain range formed when the Indian subcontinent collided with Eurasia approximately 50 million years ago, creating the world's highest peaks",
+      "domain": "geography",
+      "importance": 0.80,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["geography", "mountains", "asia"]
+    },
+    {
+      "id": "mem_050",
+      "content": "Ocean currents like the Gulf Stream transport heat across vast distances, moderating coastal climates and influencing global weather patterns",
+      "domain": "geography",
+      "importance": 0.80,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["geography", "oceanography", "climate"]
+    },
+    {
+      "id": "mem_051",
+      "content": "Permafrost in Arctic and subarctic regions stores vast quantities of frozen organic carbon that may release greenhouse gases as global temperatures rise",
+      "domain": "geography",
+      "importance": 0.80,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["geography", "climate", "arctic"]
+    },
+    {
+      "id": "mem_052",
+      "content": "River deltas form where sediment carried by a river is deposited as the current slows upon reaching the sea, creating fertile agricultural land",
+      "domain": "geography",
+      "importance": 0.75,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["geography", "rivers", "geomorphology"]
+    },
+    {
+      "id": "mem_053",
+      "content": "The Mediterranean Sea has a nearly enclosed basin with limited tidal range, making it saltier than the Atlantic due to high evaporation rates",
+      "domain": "geography",
+      "importance": 0.73,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["geography", "seas", "oceanography"]
+    },
+    {
+      "id": "mem_054",
+      "content": "Maillard reaction between amino acids and reducing sugars at high temperatures produces hundreds of flavor compounds that give browned food its complex taste",
+      "domain": "cooking",
+      "importance": 0.82,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["cooking", "chemistry", "flavor"]
+    },
+    {
+      "id": "mem_055",
+      "content": "Fermentation by bacteria and yeast converts sugars into alcohol, lactic acid, or carbon dioxide, producing bread, beer, wine, yogurt, and kimchi",
+      "domain": "cooking",
+      "importance": 0.83,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["cooking", "fermentation", "microbiology"]
+    },
+    {
+      "id": "mem_056",
+      "content": "Emulsification combines fat and water into a stable mixture using an emulsifier like lecithin or mustard, creating sauces like hollandaise and mayonnaise",
+      "domain": "cooking",
+      "importance": 0.78,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["cooking", "chemistry", "sauces"]
+    },
+    {
+      "id": "mem_057",
+      "content": "Caramelization occurs when sugar is heated above 160 degrees Celsius, breaking down sucrose into simpler compounds with a characteristic amber color and toffee flavor",
+      "domain": "cooking",
+      "importance": 0.78,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["cooking", "chemistry", "sugar"]
+    },
+    {
+      "id": "mem_058",
+      "content": "Umami is the fifth basic taste detected by glutamate receptors, naturally found in aged cheeses, soy sauce, mushrooms, and dried fish",
+      "domain": "cooking",
+      "importance": 0.78,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["cooking", "taste", "flavor"]
+    },
+    {
+      "id": "mem_059",
+      "content": "Braising cooks tough cuts of meat slowly in liquid at low temperature, dissolving collagen into gelatin and creating a rich, tender result",
+      "domain": "cooking",
+      "importance": 0.75,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["cooking", "techniques", "meat"]
+    },
+    {
+      "id": "mem_060",
+      "content": "Gluten forms when wheat flour proteins glutenin and gliadin are hydrated and kneaded, creating an elastic network that traps fermentation gas in bread dough",
+      "domain": "cooking",
+      "importance": 0.78,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["cooking", "baking", "chemistry"]
+    },
+    {
+      "id": "mem_061",
+      "content": "Knife skills like julienne, chiffonade, and brunoise refer to specific cutting shapes and sizes that affect cook times and texture consistency",
+      "domain": "cooking",
+      "importance": 0.70,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["cooking", "techniques", "preparation"]
+    },
+    {
+      "id": "mem_062",
+      "content": "Spices like turmeric, cumin, coriander, and cardamom form the flavor backbone of South Asian cuisine and have documented antimicrobial properties",
+      "domain": "cooking",
+      "importance": 0.73,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["cooking", "spices", "asian-cuisine"]
+    },
+    {
+      "id": "mem_063",
+      "content": "Stock is made by simmering bones, aromatic vegetables, and herbs to extract collagen, gelatin, and mineral flavor into a rich base liquid",
+      "domain": "cooking",
+      "importance": 0.73,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["cooking", "techniques", "sauces"]
+    },
+    {
+      "id": "mem_064",
+      "content": "Insulin secreted by pancreatic beta cells lowers blood glucose by facilitating cellular uptake of glucose and inhibiting hepatic glucose production",
+      "domain": "medicine",
+      "importance": 0.88,
+      "decay_factor": 0.01,
+      "memory_type": "semantic",
+      "tags": ["medicine", "endocrinology", "diabetes"]
+    },
+    {
+      "id": "mem_065",
+      "content": "Statins inhibit the HMG-CoA reductase enzyme to reduce hepatic cholesterol synthesis, lowering LDL levels and cardiovascular disease risk",
+      "domain": "medicine",
+      "importance": 0.85,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["medicine", "pharmacology", "cardiology"]
+    },
+    {
+      "id": "mem_066",
+      "content": "Antibiotic resistance arises when bacteria acquire mutations or resistance genes that inactivate drugs, making infections harder to treat",
+      "domain": "medicine",
+      "importance": 0.87,
+      "decay_factor": 0.01,
+      "memory_type": "semantic",
+      "tags": ["medicine", "microbiology", "public-health"]
+    },
+    {
+      "id": "mem_067",
+      "content": "Herd immunity protects vulnerable populations when a sufficient proportion of a community is immune, preventing pathogen transmission chains",
+      "domain": "medicine",
+      "importance": 0.85,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["medicine", "epidemiology", "vaccines"]
+    },
+    {
+      "id": "mem_068",
+      "content": "Cognitive behavioral therapy helps patients identify and restructure maladaptive thought patterns and behaviors that maintain depression and anxiety",
+      "domain": "medicine",
+      "importance": 0.83,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["medicine", "psychiatry", "therapy"]
+    },
+    {
+      "id": "mem_069",
+      "content": "Pharmacokinetics describes how drugs are absorbed, distributed, metabolized, and excreted by the body, determining dosing intervals and therapeutic windows",
+      "domain": "medicine",
+      "importance": 0.83,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["medicine", "pharmacology"]
+    },
+    {
+      "id": "mem_070",
+      "content": "Magnetic resonance imaging uses radiofrequency pulses and magnetic fields to excite hydrogen nuclei, creating high-contrast soft tissue images",
+      "domain": "medicine",
+      "importance": 0.85,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["medicine", "imaging", "diagnostics"]
+    },
+    {
+      "id": "mem_071",
+      "content": "Synaptic plasticity—the strengthening and weakening of synaptic connections through long-term potentiation and depression—is the cellular basis of learning",
+      "domain": "medicine",
+      "importance": 0.87,
+      "decay_factor": 0.01,
+      "memory_type": "semantic",
+      "tags": ["medicine", "neuroscience", "memory"]
+    },
+    {
+      "id": "mem_072",
+      "content": "Electroencephalography records electrical activity from scalp electrodes to detect seizure patterns, sleep stages, and brain oscillations",
+      "domain": "medicine",
+      "importance": 0.80,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["medicine", "neuroscience", "diagnostics"]
+    },
+    {
+      "id": "mem_073",
+      "content": "Coagulation cascade involves a series of protease activations that convert fibrinogen to fibrin, forming a blood clot to stop bleeding at injury sites",
+      "domain": "medicine",
+      "importance": 0.82,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["medicine", "hematology"]
+    },
+    {
+      "id": "mem_074",
+      "content": "Entropy in thermodynamics measures the number of accessible microstates of a system; the second law states that entropy of an isolated system never decreases",
+      "domain": "physics",
+      "importance": 0.87,
+      "decay_factor": 0.01,
+      "memory_type": "semantic",
+      "tags": ["physics", "thermodynamics"]
+    },
+    {
+      "id": "mem_075",
+      "content": "Electromagnetic radiation spans wavelengths from gamma rays through X-rays, ultraviolet, visible light, infrared, and radio waves across the electromagnetic spectrum",
+      "domain": "physics",
+      "importance": 0.83,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["physics", "electromagnetism", "optics"]
+    },
+    {
+      "id": "mem_076",
+      "content": "Wave-particle duality holds that quantum objects exhibit wave behavior in interference experiments and particle behavior in detection measurements",
+      "domain": "physics",
+      "importance": 0.87,
+      "decay_factor": 0.01,
+      "memory_type": "semantic",
+      "tags": ["physics", "quantum", "wave-particle"]
+    },
+    {
+      "id": "mem_077",
+      "content": "Dark matter is a form of matter that does not emit electromagnetic radiation but exerts gravitational influence, comprising about 27% of the universe's mass-energy",
+      "domain": "physics",
+      "importance": 0.83,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["physics", "astrophysics", "cosmology"]
+    },
+    {
+      "id": "mem_078",
+      "content": "Superconductivity occurs in certain materials cooled below a critical temperature, allowing electrical current to flow with zero resistance",
+      "domain": "physics",
+      "importance": 0.82,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["physics", "condensed-matter", "superconductivity"]
+    },
+    {
+      "id": "mem_079",
+      "content": "Inflation theory proposes that the universe underwent exponential expansion in the first fraction of a second after the Big Bang, explaining its observed homogeneity",
+      "domain": "physics",
+      "importance": 0.82,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["physics", "cosmology", "inflation"]
+    },
+    {
+      "id": "mem_080",
+      "content": "Supply and demand curves intersect at the market equilibrium price where the quantity consumers want to buy equals what producers want to sell",
+      "domain": "economics",
+      "importance": 0.83,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["economics", "microeconomics", "markets"]
+    },
+    {
+      "id": "mem_081",
+      "content": "Inflation erodes purchasing power when the general price level rises over time, measured by indices like the consumer price index and producer price index",
+      "domain": "economics",
+      "importance": 0.83,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["economics", "macroeconomics", "inflation"]
+    },
+    {
+      "id": "mem_082",
+      "content": "Comparative advantage explains why nations trade by specializing in goods they produce at lower opportunity cost, even if one nation is more efficient at all goods",
+      "domain": "economics",
+      "importance": 0.83,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["economics", "trade", "international"]
+    },
+    {
+      "id": "mem_083",
+      "content": "Monetary policy conducted by central banks adjusts interest rates and money supply to influence inflation, employment, and economic growth",
+      "domain": "economics",
+      "importance": 0.85,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["economics", "macroeconomics", "central-banking"]
+    },
+    {
+      "id": "mem_084",
+      "content": "Market failure occurs when free markets allocate resources inefficiently, as in the case of externalities, public goods, information asymmetry, or market power",
+      "domain": "economics",
+      "importance": 0.82,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["economics", "microeconomics", "market-failure"]
+    },
+    {
+      "id": "mem_085",
+      "content": "Game theory models strategic interactions where agents choose actions that depend on expected responses from other rational actors",
+      "domain": "economics",
+      "importance": 0.83,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["economics", "game-theory", "strategy"]
+    },
+    {
+      "id": "mem_086",
+      "content": "Phonemes are the minimal contrastive sound units of a language; changing one phoneme can change the meaning of a word, as in bat versus cat",
+      "domain": "linguistics",
+      "importance": 0.80,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["linguistics", "phonology"]
+    },
+    {
+      "id": "mem_087",
+      "content": "Syntactic trees represent the hierarchical constituency structure of sentences, showing how phrases are nested inside larger phrases",
+      "domain": "linguistics",
+      "importance": 0.78,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["linguistics", "syntax"]
+    },
+    {
+      "id": "mem_088",
+      "content": "Linguistic relativity, or the Sapir-Whorf hypothesis, proposes that the language one speaks influences how one perceives and categorizes reality",
+      "domain": "linguistics",
+      "importance": 0.78,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["linguistics", "semantics", "cognition"]
+    },
+    {
+      "id": "mem_089",
+      "content": "Language acquisition in children proceeds through cooing, babbling, one-word, and multi-word stages, with critical period effects on phonological learning",
+      "domain": "linguistics",
+      "importance": 0.80,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["linguistics", "acquisition", "development"]
+    },
+    {
+      "id": "mem_090",
+      "content": "Loanwords enter a language when speakers adopt foreign vocabulary for new concepts, technologies, or cultural practices, often adapting pronunciation to native phonotactics",
+      "domain": "linguistics",
+      "importance": 0.73,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["linguistics", "lexicology", "contact"]
+    },
+    {
+      "id": "mem_091",
+      "content": "Morphology studies how words are built from smaller meaningful units called morphemes, including roots, prefixes, and suffixes",
+      "domain": "linguistics",
+      "importance": 0.77,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["linguistics", "morphology"]
+    },
+    {
+      "id": "mem_092",
+      "content": "Pragma was an ancient Greek word for action or affair; pragmatics studies how context shapes meaning beyond literal sentence semantics",
+      "domain": "linguistics",
+      "importance": 0.73,
+      "decay_factor": 0.03,
+      "memory_type": "semantic",
+      "tags": ["linguistics", "pragmatics"]
+    },
+    {
+      "id": "mem_093",
+      "content": "Feature scaling standardizes input variables to a common range before training, preventing features with large numeric values from dominating gradient updates",
+      "domain": "ml",
+      "importance": 0.78,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["ml", "preprocessing"]
+    },
+    {
+      "id": "mem_094",
+      "content": "Principal component analysis projects high-dimensional data onto orthogonal axes of maximum variance, enabling dimensionality reduction and visualization",
+      "domain": "ml",
+      "importance": 0.80,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["ml", "dimensionality-reduction"]
+    },
+    {
+      "id": "mem_095",
+      "content": "Support vector machines find the maximum-margin hyperplane separating classes in a high-dimensional feature space, with kernel functions for nonlinear boundaries",
+      "domain": "ml",
+      "importance": 0.80,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["ml", "classification", "svm"]
+    },
+    {
+      "id": "mem_096",
+      "content": "Random forests aggregate predictions from many decision trees trained on bootstrap samples and random feature subsets to reduce overfitting and variance",
+      "domain": "ml",
+      "importance": 0.82,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["ml", "ensemble", "decision-trees"]
+    },
+    {
+      "id": "mem_097",
+      "content": "Variational autoencoders learn a compressed latent representation of data by encoding inputs into a probability distribution and reconstructing from samples",
+      "domain": "ml",
+      "importance": 0.82,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["ml", "generative", "vae"]
+    },
+    {
+      "id": "mem_098",
+      "content": "The citric acid cycle, also called the Krebs cycle, oxidizes acetyl-CoA derived from glucose and fatty acids, producing NADH and FADH2 for ATP synthesis",
+      "domain": "biology",
+      "importance": 0.82,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["biology", "metabolism", "biochemistry"]
+    },
+    {
+      "id": "mem_099",
+      "content": "Apoptosis is programmed cell death triggered by internal or external signals, removing damaged or unnecessary cells without causing inflammation",
+      "domain": "biology",
+      "importance": 0.82,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["biology", "cell", "death"]
+    },
+    {
+      "id": "mem_100",
+      "content": "Vesicles transport cargo proteins between organelles in eukaryotic cells, with SNARE proteins mediating membrane fusion at the destination compartment",
+      "domain": "biology",
+      "importance": 0.78,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["biology", "cell", "transport"]
+    },
+    {
+      "id": "mem_101",
+      "content": "Microservices architecture decomposes an application into small independently deployable services that communicate over APIs or message queues",
+      "domain": "programming",
+      "importance": 0.82,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["programming", "architecture", "microservices"]
+    },
+    {
+      "id": "mem_102",
+      "content": "Tensor algebra represents multi-dimensional arrays and their operations, forming the computational foundation of modern deep learning frameworks",
+      "domain": "ml",
+      "importance": 0.82,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["ml", "math", "tensors"]
+    },
+    {
+      "id": "mem_103",
+      "content": "Plate boundaries where two plates converge produce subduction zones, trenches, and arc volcanoes; divergent boundaries produce mid-ocean ridges and rift valleys",
+      "domain": "geography",
+      "importance": 0.78,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["geography", "geology", "tectonics"]
+    },
+    {
+      "id": "mem_104",
+      "content": "Ancient Mesopotamia between the Tigris and Euphrates rivers developed the earliest writing system—cuneiform—along with the first legal codes and urban civilizations",
+      "domain": "history",
+      "importance": 0.82,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["history", "ancient", "mesopotamia"]
+    },
+    {
+      "id": "mem_105",
+      "content": "Sous vide cooking seals food in vacuum bags and cooks it in precisely temperature-controlled water baths, achieving uniform doneness impossible with conventional heat",
+      "domain": "cooking",
+      "importance": 0.73,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["cooking", "techniques", "precision"]
+    },
+    {
+      "id": "mem_106",
+      "content": "Nash equilibrium describes a stable outcome in strategic games where no player can benefit by unilaterally changing their strategy given others' choices",
+      "domain": "economics",
+      "importance": 0.82,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["economics", "game-theory"]
+    },
+    {
+      "id": "mem_107",
+      "content": "Germ theory of disease, established by Pasteur and Koch, holds that infectious diseases are caused by microorganisms that invade the host",
+      "domain": "medicine",
+      "importance": 0.83,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["medicine", "microbiology", "history"]
+    },
+    {
+      "id": "mem_108",
+      "content": "Photovoltaic cells convert incident photons into electrical current when absorbed photons excite electrons across a semiconductor bandgap",
+      "domain": "physics",
+      "importance": 0.80,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["physics", "optics", "solar-energy"]
+    },
+    {
+      "id": "mem_109",
+      "content": "The bond market prices sovereign debt using yield curves that reflect maturity, credit risk, and expected future interest rate movements",
+      "domain": "economics",
+      "importance": 0.78,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["economics", "finance", "bonds"]
+    },
+    {
+      "id": "mem_110",
+      "content": "Semantic shift is the process by which a word gradually changes meaning over generations, often through metaphor, generalization, or specialization",
+      "domain": "linguistics",
+      "importance": 0.73,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["linguistics", "semantics", "change"]
+    },
+    {
+      "id": "mem_111",
+      "content": "Bought a new mechanical keyboard today with Cherry MX brown switches; the tactile bump is perfect for long coding sessions without the noise of clicky switches",
+      "domain": "distractor",
+      "importance": 0.40,
+      "decay_factor": 0.05,
+      "memory_type": "episodic",
+      "tags": ["personal", "hardware"]
+    },
+    {
+      "id": "mem_112",
+      "content": "The movie was entertaining but the ending felt rushed; the director should have given the third act at least another twenty minutes to resolve the subplot properly",
+      "domain": "distractor",
+      "importance": 0.35,
+      "decay_factor": 0.05,
+      "memory_type": "episodic",
+      "tags": ["personal", "entertainment"]
+    },
+    {
+      "id": "mem_113",
+      "content": "Checked in for the flight at 5am; the airport coffee was terrible but the lounge had decent wifi so managed to finish the slide deck before boarding",
+      "domain": "distractor",
+      "importance": 0.30,
+      "decay_factor": 0.06,
+      "memory_type": "episodic",
+      "tags": ["personal", "travel"]
+    },
+    {
+      "id": "mem_114",
+      "content": "The apartment landlord finally fixed the leaking pipe under the kitchen sink after three weeks of follow-up emails and one formal complaint letter",
+      "domain": "distractor",
+      "importance": 0.28,
+      "decay_factor": 0.06,
+      "memory_type": "episodic",
+      "tags": ["personal", "home"]
+    },
+    {
+      "id": "mem_115",
+      "content": "Tried making sourdough for the first time; the starter is active after four days of feeding and the kitchen smells wonderfully yeasty and sour",
+      "domain": "distractor",
+      "importance": 0.38,
+      "decay_factor": 0.05,
+      "memory_type": "episodic",
+      "tags": ["personal", "cooking", "hobby"]
+    },
+    {
+      "id": "mem_116",
+      "content": "Finished reading a novel about a lighthouse keeper in 1920s Newfoundland; the prose was beautifully sparse and the isolation felt viscerally real",
+      "domain": "distractor",
+      "importance": 0.32,
+      "decay_factor": 0.06,
+      "memory_type": "episodic",
+      "tags": ["personal", "reading"]
+    },
+    {
+      "id": "mem_117",
+      "content": "The afternoon sun angle through the west window is now different enough from summer that I need to adjust my monitor position again to avoid glare",
+      "domain": "distractor",
+      "importance": 0.22,
+      "decay_factor": 0.07,
+      "memory_type": "episodic",
+      "tags": ["personal", "home"]
+    },
+    {
+      "id": "mem_118",
+      "content": "Picked up the dry cleaning and noticed the collar on the blue shirt has a faint yellow stain that the cleaner missed; need to bring it back next week",
+      "domain": "distractor",
+      "importance": 0.18,
+      "decay_factor": 0.08,
+      "memory_type": "episodic",
+      "tags": ["personal", "errands"]
+    },
+    {
+      "id": "mem_119",
+      "content": "The Wi-Fi router needed a firmware update; after flashing it the 5GHz band dropped connection randomly for two hours until a second reboot fixed it",
+      "domain": "distractor",
+      "importance": 0.32,
+      "decay_factor": 0.06,
+      "memory_type": "episodic",
+      "tags": ["personal", "hardware", "networking"]
+    },
+    {
+      "id": "mem_120",
+      "content": "Saw a heron standing completely still in the creek behind the park for almost twenty minutes before it finally struck at something in the water",
+      "domain": "distractor",
+      "importance": 0.25,
+      "decay_factor": 0.06,
+      "memory_type": "episodic",
+      "tags": ["personal", "nature"]
+    },
+    {
+      "id": "mem_121",
+      "content": "Tried a new ramen shop downtown; the broth was tonkotsu-style, rich and cloudy with a strong pork bone flavor, and the chashu was melt-in-your-mouth tender",
+      "domain": "distractor",
+      "importance": 0.35,
+      "decay_factor": 0.05,
+      "memory_type": "episodic",
+      "tags": ["personal", "food", "restaurant"]
+    },
+    {
+      "id": "mem_122",
+      "content": "Signed up for a pottery class starting next month; always wanted to learn wheel throwing but never made the time before",
+      "domain": "distractor",
+      "importance": 0.28,
+      "decay_factor": 0.06,
+      "memory_type": "episodic",
+      "tags": ["personal", "hobby"]
+    },
+    {
+      "id": "mem_123",
+      "content": "The plumber came to replace the water heater; the new one is tankless and should save about thirty percent on the monthly energy bill",
+      "domain": "distractor",
+      "importance": 0.25,
+      "decay_factor": 0.07,
+      "memory_type": "episodic",
+      "tags": ["personal", "home", "energy"]
+    },
+    {
+      "id": "mem_124",
+      "content": "Ordered replacement brake pads for the bicycle; the organic pads wear faster than sintered but the modulation is much more predictable on wet descents",
+      "domain": "distractor",
+      "importance": 0.30,
+      "decay_factor": 0.06,
+      "memory_type": "episodic",
+      "tags": ["personal", "cycling"]
+    },
+    {
+      "id": "mem_125",
+      "content": "The orchid on the windowsill bloomed for the third time this year; apparently repotting it into bark medium and reducing watering frequency was the key",
+      "domain": "distractor",
+      "importance": 0.22,
+      "decay_factor": 0.07,
+      "memory_type": "episodic",
+      "tags": ["personal", "plants"]
+    },
+    {
+      "id": "mem_126",
+      "content": "Watched a documentary about deep-sea hydrothermal vents and the extremophile communities that thrive there without sunlight, driven entirely by chemosynthesis",
+      "domain": "distractor",
+      "importance": 0.40,
+      "decay_factor": 0.05,
+      "memory_type": "episodic",
+      "tags": ["personal", "science-interest"]
+    },
+    {
+      "id": "mem_127",
+      "content": "Had coffee with an old colleague who now works on satellite telemetry; fascinating conversation about the latency tradeoffs in low-earth-orbit constellations",
+      "domain": "distractor",
+      "importance": 0.38,
+      "decay_factor": 0.05,
+      "memory_type": "episodic",
+      "tags": ["personal", "conversation"]
+    },
+    {
+      "id": "mem_128",
+      "content": "The grocery delivery app sent the wrong brand of olive oil again; at least this one is also extra-virgin so the flavor difference is minimal",
+      "domain": "distractor",
+      "importance": 0.15,
+      "decay_factor": 0.09,
+      "memory_type": "episodic",
+      "tags": ["personal", "errands", "food"]
+    },
+    {
+      "id": "mem_129",
+      "content": "Set up a new aquarium with a planted substrate and CO2 injection; the stem plants are already putting out visible new growth after just five days",
+      "domain": "distractor",
+      "importance": 0.28,
+      "decay_factor": 0.06,
+      "memory_type": "episodic",
+      "tags": ["personal", "hobby"]
+    },
+    {
+      "id": "mem_130",
+      "content": "Photosynthesis in C4 plants like corn and sugarcane uses a carbon-concentrating mechanism that reduces photorespiration and improves water-use efficiency in hot climates",
+      "domain": "biology",
+      "importance": 0.78,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["biology", "plants", "photosynthesis"]
+    },
+    {
+      "id": "mem_131",
+      "content": "Neural oscillations at gamma frequencies around 40Hz in the prefrontal cortex correlate with working memory maintenance and conscious attention",
+      "domain": "medicine",
+      "importance": 0.80,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["medicine", "neuroscience", "cognition"]
+    },
+    {
+      "id": "mem_132",
+      "content": "The Great Barrier Reef is the world's largest coral reef system, spanning 2,300 kilometers along the northeast Australian coast and hosting enormous marine biodiversity",
+      "domain": "geography",
+      "importance": 0.75,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["geography", "ocean", "australia"]
+    },
+    {
+      "id": "mem_133",
+      "content": "Marginal cost is the additional cost incurred by producing one more unit of output; profit-maximizing firms produce until marginal cost equals marginal revenue",
+      "domain": "economics",
+      "importance": 0.80,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["economics", "microeconomics", "production"]
+    },
+    {
+      "id": "mem_134",
+      "content": "Pitch accent systems assign high or low tones to syllables to distinguish word meaning, as in Japanese, Swedish, and some dialects of Serbian",
+      "domain": "linguistics",
+      "importance": 0.72,
+      "decay_factor": 0.03,
+      "memory_type": "semantic",
+      "tags": ["linguistics", "phonology", "prosody"]
+    },
+    {
+      "id": "mem_135",
+      "content": "The photoelectric effect—that light ejects electrons from metal surfaces only above a threshold frequency—led Einstein to propose the photon model of light",
+      "domain": "physics",
+      "importance": 0.83,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["physics", "quantum", "history"]
+    },
+    {
+      "id": "mem_136",
+      "content": "Immune checkpoint inhibitors block proteins like PD-1 and CTLA-4 that normally suppress T-cell activation, unleashing anti-tumor immune responses in cancer patients",
+      "domain": "medicine",
+      "importance": 0.85,
+      "decay_factor": 0.01,
+      "memory_type": "semantic",
+      "tags": ["medicine", "oncology", "immunotherapy"]
+    },
+    {
+      "id": "mem_137",
+      "content": "The Coriolis effect causes moving air and water masses to deflect rightward in the northern hemisphere and leftward in the southern hemisphere due to Earth's rotation",
+      "domain": "geography",
+      "importance": 0.78,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["geography", "meteorology", "physics"]
+    },
+    {
+      "id": "mem_138",
+      "content": "Dependency injection decouples classes from their dependencies by passing required objects through constructor parameters rather than creating them internally",
+      "domain": "programming",
+      "importance": 0.80,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["programming", "design-patterns", "oop"]
+    },
+    {
+      "id": "mem_139",
+      "content": "Protectionism uses tariffs, quotas, and subsidies to shield domestic industries from foreign competition, often reducing allocative efficiency but preserving employment",
+      "domain": "economics",
+      "importance": 0.78,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["economics", "trade", "policy"]
+    },
+    {
+      "id": "mem_140",
+      "content": "Pidgins form as simplified contact languages between speakers of mutually unintelligible tongues; when children acquire a pidgin as their first language it becomes a creole",
+      "domain": "linguistics",
+      "importance": 0.75,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["linguistics", "contact", "creolization"]
+    },
+    {
+      "id": "mem_141",
+      "content": "Dropped my phone walking to the metro and cracked the screen protector; the display underneath seems fine but I should replace the protector before the cracks spread",
+      "domain": "distractor",
+      "importance": 0.20,
+      "decay_factor": 0.08,
+      "memory_type": "episodic",
+      "tags": ["personal", "errands"]
+    },
+    {
+      "id": "mem_142",
+      "content": "The quarterly rent increase notice arrived; the landlord is citing rising property taxes and maintenance costs as justification for an eight percent hike",
+      "domain": "distractor",
+      "importance": 0.30,
+      "decay_factor": 0.06,
+      "memory_type": "episodic",
+      "tags": ["personal", "housing"]
+    },
+    {
+      "id": "mem_143",
+      "content": "Watched a chess tournament stream; the world champion's endgame technique in the king-and-pawn endgame was surgical—every tempo precisely calculated",
+      "domain": "distractor",
+      "importance": 0.35,
+      "decay_factor": 0.05,
+      "memory_type": "episodic",
+      "tags": ["personal", "chess"]
+    },
+    {
+      "id": "mem_144",
+      "content": "The cat knocked over the succulent arrangement on the bookshelf for the second time this week; need to find a higher shelf or a hanging planter",
+      "domain": "distractor",
+      "importance": 0.12,
+      "decay_factor": 0.10,
+      "memory_type": "episodic",
+      "tags": ["personal", "home", "pets"]
+    },
+    {
+      "id": "mem_145",
+      "content": "Finished the online Spanish course level B1; the subjunctive mood is still tricky but the reading comprehension has improved noticeably over the past three months",
+      "domain": "distractor",
+      "importance": 0.35,
+      "decay_factor": 0.05,
+      "memory_type": "episodic",
+      "tags": ["personal", "language-learning"]
+    },
+    {
+      "id": "mem_146",
+      "content": "The hiking trail was muddy after the rain but the views from the ridge were spectacular; took about four hours round trip including a thirty minute lunch stop",
+      "domain": "distractor",
+      "importance": 0.25,
+      "decay_factor": 0.07,
+      "memory_type": "episodic",
+      "tags": ["personal", "outdoor", "hiking"]
+    },
+    {
+      "id": "mem_147",
+      "content": "Annealing in metallurgy heats a metal to just below its melting point and then cools it slowly, relieving internal stresses and improving ductility",
+      "domain": "physics",
+      "importance": 0.73,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["physics", "materials", "metallurgy"]
+    },
+    {
+      "id": "mem_148",
+      "content": "The Silk Road connected Han Dynasty China to the Roman Empire via Central Asian oases, facilitating trade in silk, spices, and glassware across Eurasia",
+      "domain": "history",
+      "importance": 0.78,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["history", "trade", "ancient"]
+    },
+    {
+      "id": "mem_149",
+      "content": "Osmosis is the movement of water across a semipermeable membrane from a region of lower solute concentration to higher solute concentration",
+      "domain": "biology",
+      "importance": 0.80,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["biology", "cell", "transport"]
+    },
+    {
+      "id": "mem_150",
+      "content": "Terroir refers to the complete natural environment—soil composition, microclimate, topography, and local microflora—that gives wines their distinctive regional character",
+      "domain": "cooking",
+      "importance": 0.70,
+      "decay_factor": 0.03,
+      "memory_type": "semantic",
+      "tags": ["cooking", "wine", "geography"]
+    },
+    {
+      "id": "mem_151",
+      "content": "The Fourier transform decomposes a time-domain signal into its constituent frequencies, enabling spectral analysis and signal filtering operations",
+      "domain": "physics",
+      "importance": 0.83,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["physics", "math", "signal-processing"]
+    },
+    {
+      "id": "mem_152",
+      "content": "Stock options give holders the right but not obligation to buy or sell shares at a fixed strike price before expiration, enabling leveraged speculation and hedging",
+      "domain": "economics",
+      "importance": 0.78,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["economics", "finance", "derivatives"]
+    },
+    {
+      "id": "mem_153",
+      "content": "Conlang designers like Tolkien and Okrand create invented languages with consistent phonology, morphology, and syntax for fictional worlds or linguistic experiments",
+      "domain": "linguistics",
+      "importance": 0.68,
+      "decay_factor": 0.03,
+      "memory_type": "semantic",
+      "tags": ["linguistics", "conlangs"]
+    },
+    {
+      "id": "mem_154",
+      "content": "The human gut microbiome contains trillions of bacteria that influence digestion, immune regulation, and possibly mood through the gut-brain axis",
+      "domain": "biology",
+      "importance": 0.82,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["biology", "microbiome", "health"]
+    },
+    {
+      "id": "mem_155",
+      "content": "Colonoscopy uses a flexible camera to inspect the entire colon, detecting polyps that can be removed to prevent colorectal cancer progression",
+      "domain": "medicine",
+      "importance": 0.78,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["medicine", "oncology", "screening"]
+    },
+    {
+      "id": "mem_156",
+      "content": "The Hanging Gardens of Babylon, if they existed, were reportedly an ascending series of terraced gardens irrigated by pumps drawing from the Euphrates River",
+      "domain": "distractor",
+      "importance": 0.55,
+      "decay_factor": 0.03,
+      "memory_type": "semantic",
+      "tags": ["history", "ancient", "wonders"]
+    },
+    {
+      "id": "mem_157",
+      "content": "Gradient boosting builds an ensemble of weak learners sequentially, with each tree fitting the residuals left by the previous ones to minimize a differentiable loss",
+      "domain": "ml",
+      "importance": 0.83,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["ml", "ensemble", "boosting"]
+    },
+    {
+      "id": "mem_158",
+      "content": "Diffusion models learn to reverse a gradual noise-addition process, generating high-quality images by iteratively denoising from pure Gaussian noise",
+      "domain": "ml",
+      "importance": 0.85,
+      "decay_factor": 0.01,
+      "memory_type": "semantic",
+      "tags": ["ml", "generative", "diffusion"]
+    },
+    {
+      "id": "mem_159",
+      "content": "Quorum sensing allows bacteria to detect population density by accumulating signaling molecules and trigger coordinated gene expression at threshold concentrations",
+      "domain": "biology",
+      "importance": 0.78,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["biology", "microbiology", "gene-regulation"]
+    },
+    {
+      "id": "mem_160",
+      "content": "Urban heat islands form where concrete and asphalt absorb and retain solar energy, raising city temperatures several degrees above surrounding rural areas",
+      "domain": "geography",
+      "importance": 0.75,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["geography", "urban", "climate"]
+    },
+    {
+      "id": "mem_161",
+      "content": "Bought a new espresso machine with PID temperature control and a pressure profiling kit; dialing in the extraction for the Ethiopian natural process beans took about twenty shots",
+      "domain": "distractor",
+      "importance": 0.38,
+      "decay_factor": 0.05,
+      "memory_type": "episodic",
+      "tags": ["personal", "coffee", "hobby"]
+    },
+    {
+      "id": "mem_162",
+      "content": "The mortgage refinancing closed at a lower rate than expected; the break-even on closing costs works out to about eighteen months of reduced payments",
+      "domain": "distractor",
+      "importance": 0.35,
+      "decay_factor": 0.05,
+      "memory_type": "episodic",
+      "tags": ["personal", "finance"]
+    },
+    {
+      "id": "mem_163",
+      "content": "Formal verification uses mathematical proof techniques to guarantee that a program or hardware design satisfies specified correctness properties",
+      "domain": "programming",
+      "importance": 0.82,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["programming", "verification", "formal-methods"]
+    },
+    {
+      "id": "mem_164",
+      "content": "Protein folding determines three-dimensional structure from the amino acid sequence; misfolding into beta-sheet aggregates causes diseases like Alzheimer's and Parkinson's",
+      "domain": "biology",
+      "importance": 0.85,
+      "decay_factor": 0.01,
+      "memory_type": "semantic",
+      "tags": ["biology", "proteins", "disease"]
+    },
+    {
+      "id": "mem_165",
+      "content": "The Paris Agreement commits signatories to limit warming to 1.5 degrees Celsius above pre-industrial levels through nationally determined carbon reduction targets",
+      "domain": "geography",
+      "importance": 0.80,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["geography", "climate", "policy"]
+    },
+    {
+      "id": "mem_166",
+      "content": "Microwave ovens heat food by exciting water molecules with 2.45 GHz electromagnetic radiation, causing rapid internal heating through dielectric absorption",
+      "domain": "distractor",
+      "importance": 0.48,
+      "decay_factor": 0.04,
+      "memory_type": "semantic",
+      "tags": ["physics", "cooking", "appliances"]
+    },
+    {
+      "id": "mem_167",
+      "content": "Constitutional monarchies retain a hereditary monarch as ceremonial head of state while an elected parliament and prime minister hold actual governing power",
+      "domain": "history",
+      "importance": 0.75,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["history", "government", "monarchy"]
+    },
+    {
+      "id": "mem_168",
+      "content": "Seismographs detect and record ground motion from earthquakes by measuring relative displacement between a suspended mass and the moving earth frame",
+      "domain": "geography",
+      "importance": 0.73,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["geography", "geology", "seismology"]
+    },
+    {
+      "id": "mem_169",
+      "content": "Central bank quantitative easing purchases government bonds to inject liquidity and suppress long-term interest rates when conventional monetary policy is constrained",
+      "domain": "economics",
+      "importance": 0.80,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["economics", "macroeconomics", "monetary-policy"]
+    },
+    {
+      "id": "mem_170",
+      "content": "The smell of baking bread triggers strong autobiographical memories because the olfactory bulb connects directly to hippocampal memory circuits without thalamic relay",
+      "domain": "distractor",
+      "importance": 0.45,
+      "decay_factor": 0.04,
+      "memory_type": "semantic",
+      "tags": ["neuroscience", "memory", "smell"]
+    },
+    {
+      "id": "mem_171",
+      "content": "Optogenetics uses genetically encoded light-sensitive proteins to activate or silence specific neurons with millisecond precision in behaving animals",
+      "domain": "medicine",
+      "importance": 0.83,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["medicine", "neuroscience", "tools"]
+    },
+    {
+      "id": "mem_172",
+      "content": "Diphthongs are complex vowel sounds where the tongue glides from one position to another within a single syllable, as in the English words bite and boat",
+      "domain": "linguistics",
+      "importance": 0.70,
+      "decay_factor": 0.03,
+      "memory_type": "semantic",
+      "tags": ["linguistics", "phonology", "vowels"]
+    },
+    {
+      "id": "mem_173",
+      "content": "Arbitrage exploits price differences for the same asset across markets, and in efficient markets arbitrage opportunities are eliminated almost instantly",
+      "domain": "economics",
+      "importance": 0.78,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["economics", "finance", "markets"]
+    },
+    {
+      "id": "mem_174",
+      "content": "The fermentation of milk by Lactobacillus acidophilus and other bacteria lowers pH through lactic acid production, thickening the milk and preserving it as yogurt",
+      "domain": "cooking",
+      "importance": 0.73,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["cooking", "fermentation", "dairy"]
+    },
+    {
+      "id": "mem_175",
+      "content": "Virtual memory allows processes to use more memory than physically installed by mapping pages to disk, with the operating system swapping pages on demand",
+      "domain": "programming",
+      "importance": 0.80,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["programming", "systems", "memory"]
+    },
+    {
+      "id": "mem_176",
+      "content": "The decolonization movement after World War II led dozens of African and Asian nations to achieve independence from European colonial powers between 1945 and 1975",
+      "domain": "history",
+      "importance": 0.80,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["history", "colonization", "20th-century"]
+    },
+    {
+      "id": "mem_177",
+      "content": "Sleep is best right now and I am too tired to write anything coherent; this entry is just a placeholder to note that Tuesday was completely uneventful",
+      "domain": "distractor",
+      "importance": 0.10,
+      "decay_factor": 0.10,
+      "memory_type": "episodic",
+      "tags": ["personal", "daily"]
+    },
+    {
+      "id": "mem_178",
+      "content": "Scoville scale measures capsaicin concentration in peppers by expressing the dilution factor required for a trained panel to detect no heat, ranging from zero for bell peppers to millions for Carolina Reapers",
+      "domain": "cooking",
+      "importance": 0.65,
+      "decay_factor": 0.03,
+      "memory_type": "semantic",
+      "tags": ["cooking", "peppers", "measurement"]
+    },
+    {
+      "id": "mem_179",
+      "content": "Convolutional feature maps from early layers detect edges and textures while deeper layers respond to object parts and whole objects in computer vision models",
+      "domain": "ml",
+      "importance": 0.83,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["ml", "cnn", "interpretability"]
+    },
+    {
+      "id": "mem_180",
+      "content": "Continental drift was first proposed by Alfred Wegener in 1912 based on the matching coastlines of Africa and South America and similar fossil assemblages on both continents",
+      "domain": "geography",
+      "importance": 0.78,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["geography", "geology", "history-of-science"]
+    },
+    {
+      "id": "mem_181",
+      "content": "Finally got the standing desk assembled; the cable management took longer than the actual desk but the result is much cleaner than the old setup",
+      "domain": "distractor",
+      "importance": 0.22,
+      "decay_factor": 0.08,
+      "memory_type": "episodic",
+      "tags": ["personal", "home", "workspace"]
+    },
+    {
+      "id": "mem_182",
+      "content": "The team trivia night was won by the table in the back corner; the decisive question about 1980s pop music eliminated the engineering teams early",
+      "domain": "distractor",
+      "importance": 0.18,
+      "decay_factor": 0.09,
+      "memory_type": "episodic",
+      "tags": ["personal", "social"]
+    },
+    {
+      "id": "mem_183",
+      "content": "Renewed the car registration at the DMV; the whole process took 45 minutes including the wait, which is much faster than last year",
+      "domain": "distractor",
+      "importance": 0.12,
+      "decay_factor": 0.10,
+      "memory_type": "episodic",
+      "tags": ["personal", "errands"]
+    },
+    {
+      "id": "mem_184",
+      "content": "The neighbor's dog barked at 3am again; need to figure out whether to leave a polite note or just buy better earplugs",
+      "domain": "distractor",
+      "importance": 0.10,
+      "decay_factor": 0.10,
+      "memory_type": "episodic",
+      "tags": ["personal", "home"]
+    },
+    {
+      "id": "mem_185",
+      "content": "Bought new running shoes; the extra cushioning in the heel is immediately noticeable and the wide toe box should help with the blister problem",
+      "domain": "distractor",
+      "importance": 0.25,
+      "decay_factor": 0.07,
+      "memory_type": "episodic",
+      "tags": ["personal", "fitness", "shopping"]
+    },
+    {
+      "id": "mem_186",
+      "content": "The ferry crossing was delayed by 90 minutes due to fog; eventually cleared and the sunset views over the water made up for the wait",
+      "domain": "distractor",
+      "importance": 0.20,
+      "decay_factor": 0.08,
+      "memory_type": "episodic",
+      "tags": ["personal", "travel"]
+    },
+    {
+      "id": "mem_187",
+      "content": "Just discovered that the downstairs bathroom tap drips exactly 47 times per minute; it is oddly hypnotic and very annoying at midnight",
+      "domain": "distractor",
+      "importance": 0.08,
+      "decay_factor": 0.10,
+      "memory_type": "episodic",
+      "tags": ["personal", "home"]
+    },
+    {
+      "id": "mem_188",
+      "content": "Left the umbrella on the train for the third time this year; the lost property office said to check back in a week if nobody claims it",
+      "domain": "distractor",
+      "importance": 0.10,
+      "decay_factor": 0.10,
+      "memory_type": "episodic",
+      "tags": ["personal", "errands"]
+    },
+    {
+      "id": "mem_189",
+      "content": "The hardware store was out of the specific screw size needed; ended up ordering online and the shipment arrives Thursday, delaying the shelf project",
+      "domain": "distractor",
+      "importance": 0.12,
+      "decay_factor": 0.10,
+      "memory_type": "episodic",
+      "tags": ["personal", "home", "diy"]
+    },
+    {
+      "id": "mem_190",
+      "content": "Noticed the pigeon that has been nesting on the air conditioning unit outside is now raising two chicks; they are surprisingly loud for such small birds",
+      "domain": "distractor",
+      "importance": 0.18,
+      "decay_factor": 0.09,
+      "memory_type": "episodic",
+      "tags": ["personal", "nature"]
+    },
+    {
+      "id": "mem_191",
+      "content": "Switched from manual to electric toothbrush after the dentist's suggestion; the timer feature is useful but the charging dock takes up too much counter space",
+      "domain": "distractor",
+      "importance": 0.14,
+      "decay_factor": 0.09,
+      "memory_type": "episodic",
+      "tags": ["personal", "health-routine"]
+    },
+    {
+      "id": "mem_192",
+      "content": "Finished assembling the flat-pack bookcase; the instructions included one completely ambiguous diagram that required three restarts to decipher correctly",
+      "domain": "distractor",
+      "importance": 0.15,
+      "decay_factor": 0.09,
+      "memory_type": "episodic",
+      "tags": ["personal", "home", "diy"]
+    },
+    {
+      "id": "mem_193",
+      "content": "The ice cream shop on the corner finally opened after its winter break; the salted caramel was worth the wait but the queue was 30 minutes long",
+      "domain": "distractor",
+      "importance": 0.18,
+      "decay_factor": 0.09,
+      "memory_type": "episodic",
+      "tags": ["personal", "food", "local"]
+    },
+    {
+      "id": "mem_194",
+      "content": "Power outage for three hours during the workday; made an unexpected productive afternoon of offline reading and handwritten notes",
+      "domain": "distractor",
+      "importance": 0.22,
+      "decay_factor": 0.08,
+      "memory_type": "episodic",
+      "tags": ["personal", "daily"]
+    },
+    {
+      "id": "mem_195",
+      "content": "Lent the camping gear to a friend for the weekend; remembered too late that the tent zipper is broken and needs duct tape to stay closed",
+      "domain": "distractor",
+      "importance": 0.15,
+      "decay_factor": 0.09,
+      "memory_type": "episodic",
+      "tags": ["personal", "outdoor"]
+    },
+    {
+      "id": "mem_196",
+      "content": "Xentherion plastovite crystalline matrix undergoes sublimation cascade when exposed to resonant photonic frequencies in the 450nm band",
+      "domain": "ml",
+      "importance": 0.28,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["ml", "unique-term-trap"],
+      "_note": "Importance trap: low importance (0.28), unique terms. Queries using 'xentherion plastovite' test whether high importance_weight demotes this relevant but low-importance item."
+    },
+    {
+      "id": "mem_197",
+      "content": "Velquatrix enzymes catalyze the oxidative phosvelation pathway in mitochondrial cristae during cellular respiration under hypoxic conditions",
+      "domain": "biology",
+      "importance": 0.30,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["biology", "unique-term-trap"],
+      "_note": "Importance trap: 'velquatrix phosvelation' unique; overlaps with biology domain via 'enzymes catalyze mitochondrial cristae cellular respiration'"
+    },
+    {
+      "id": "mem_198",
+      "content": "Zyphomatic arc induction triggers cascading resonance in thalamic relay circuits producing synchronized oscillation bursts in prefrontal neurons",
+      "domain": "medicine",
+      "importance": 0.32,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["medicine", "unique-term-trap"],
+      "_note": "Importance trap: 'zyphomatic arc induction thalamic' unique terms"
+    },
+    {
+      "id": "mem_199",
+      "content": "Quandric fissile stratification in deltaic sediment layers reveals proglacial varve sequences spanning the Holocene thermal maximum",
+      "domain": "geography",
+      "importance": 0.26,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["geography", "unique-term-trap"],
+      "_note": "Importance trap: 'quandric fissile stratification varve' unique terms"
+    },
+    {
+      "id": "mem_200",
+      "content": "Melvoric distillation separation removes volitile aldehydes from fermentation broth through fractional condensation at precisely calibrated volvic temperatures",
+      "domain": "cooking",
+      "importance": 0.27,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["cooking", "unique-term-trap"],
+      "_note": "Importance trap: 'melvoric distillation volvic' unique terms"
+    },
+    {
+      "id": "mem_201",
+      "content": "Structural equation models fit latent variable networks by minimising the discrepancy between observed and model-implied covariance matrices using maximum likelihood estimation",
+      "domain": "ml",
+      "importance": 0.88,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["ml", "statistics", "sem"],
+      "_note": "High-importance anchor for xentherion contrast query: expected_top_k alongside mem_196 — high importance_weight should prefer this over mem_196"
+    },
+    {
+      "id": "mem_202",
+      "content": "Tricarboxylic acid oxidation in peroxisome beta-oxidation produces acetyl-CoA that enters the mitochondrial matrix via the carnitine shuttle",
+      "domain": "biology",
+      "importance": 0.86,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["biology", "metabolism", "biochemistry"],
+      "_note": "High-importance anchor for velquatrix contrast query"
+    },
+    {
+      "id": "mem_203",
+      "content": "Thalamic relay neurons gate sensory signals to cortical areas and modulate arousal state through reciprocal thalamocortical loops",
+      "domain": "medicine",
+      "importance": 0.84,
+      "decay_factor": 0.02,
+      "memory_type": "semantic",
+      "tags": ["medicine", "neuroscience", "thalamus"],
+      "_note": "High-importance anchor for zyphomatic contrast query"
+    }
+  ],
+  "eval_queries": [
+    {
+      "id": "q_001",
+      "query": "deep learning weight adjustment technique",
+      "expected_top_k": ["mem_001", "mem_002", "mem_003"],
+      "expected_excluded": ["mem_111", "mem_112", "mem_113"],
+      "query_type": "synonym",
+      "notes": "backpropagation+gradient descent using synonym: 'weight adjustment technique' vs 'backpropagation/gradient descent'"
+    },
+    {
+      "id": "q_002",
+      "query": "sequence to sequence memory cell gating",
+      "expected_top_k": ["mem_010", "mem_009"],
+      "expected_excluded": ["mem_111", "mem_128"],
+      "query_type": "synonym",
+      "notes": "LSTM/RNN using synonyms: 'memory cell gating' vs 'long short-term memory gating mechanisms'"
+    },
+    {
+      "id": "q_003",
+      "query": "self-attention parallelizable architecture language model",
+      "expected_top_k": ["mem_004", "mem_005"],
+      "expected_excluded": ["mem_009", "mem_010"],
+      "query_type": "partial",
+      "notes": "Transformer attention — partial match, 'self-attention' not in mem_004/005 verbatim"
+    },
+    {
+      "id": "q_004",
+      "query": "regularization prevent overfitting randomly zero activations",
+      "expected_top_k": ["mem_006"],
+      "expected_excluded": ["mem_007", "mem_093"],
+      "query_type": "partial",
+      "notes": "Dropout — partial description without the word 'dropout'"
+    },
+    {
+      "id": "q_005",
+      "query": "pretrained model adaptation new task",
+      "expected_top_k": ["mem_011"],
+      "expected_excluded": ["mem_012", "mem_015"],
+      "query_type": "synonym",
+      "notes": "'pretrained model adaptation' vs 'transfer learning fine-tuning'"
+    },
+    {
+      "id": "q_006",
+      "query": "ensemble of shallow trees residual fitting sequential boosting",
+      "expected_top_k": ["mem_157", "mem_096"],
+      "expected_excluded": ["mem_001", "mem_002"],
+      "query_type": "synonym",
+      "notes": "'ensemble of shallow trees residual fitting' vs 'gradient boosting decision trees'"
+    },
+    {
+      "id": "q_007",
+      "query": "denoising score matching generative image synthesis",
+      "expected_top_k": ["mem_158"],
+      "expected_excluded": ["mem_097", "mem_006"],
+      "query_type": "synonym",
+      "notes": "'denoising score matching' vs 'diffusion models'; generative is present in both"
+    },
+    {
+      "id": "q_008",
+      "query": "cell energy currency production organelle",
+      "expected_top_k": ["mem_016", "mem_017"],
+      "expected_excluded": ["mem_020", "mem_025"],
+      "query_type": "synonym",
+      "notes": "'cell energy currency' = ATP = mitochondria; 'organelle' present; 'cellular respiration' = metabolism"
+    },
+    {
+      "id": "q_009",
+      "query": "hereditary trait selection reproductive advantage",
+      "expected_top_k": ["mem_019"],
+      "expected_excluded": ["mem_024", "mem_025"],
+      "query_type": "partial",
+      "notes": "Natural selection described without using those words"
+    },
+    {
+      "id": "q_010",
+      "query": "protein blueprint transported from nucleus to ribosome",
+      "expected_top_k": ["mem_018"],
+      "expected_excluded": ["mem_020", "mem_023"],
+      "query_type": "synonym",
+      "notes": "'protein blueprint' = mRNA; 'transported from nucleus' = transcription + export"
+    },
+    {
+      "id": "q_011",
+      "query": "biological catalyst lowers activation barrier chemical transformation",
+      "expected_top_k": ["mem_020"],
+      "expected_excluded": ["mem_016", "mem_017"],
+      "query_type": "synonym",
+      "notes": "'biological catalyst lowers activation barrier' vs 'enzyme lowers activation energy'"
+    },
+    {
+      "id": "q_012",
+      "query": "genome sequence base pairs chromosome count",
+      "expected_top_k": ["mem_023"],
+      "expected_excluded": ["mem_018", "mem_024"],
+      "query_type": "partial",
+      "notes": "Partial match: 'base pairs' and 'chromosome' present, 'genome sequence' not verbatim in mem_023"
+    },
+    {
+      "id": "q_013",
+      "query": "cooperative multitasking lightweight concurrency event loop",
+      "expected_top_k": ["mem_027"],
+      "expected_excluded": ["mem_030", "mem_033"],
+      "query_type": "synonym",
+      "notes": "'cooperative multitasking lightweight concurrency' vs 'async runtime Tokio tasks'"
+    },
+    {
+      "id": "q_014",
+      "query": "heap memory reclamation unreachable objects automatic",
+      "expected_top_k": ["mem_031"],
+      "expected_excluded": ["mem_026", "mem_175"],
+      "query_type": "synonym",
+      "notes": "'heap memory reclamation unreachable objects' vs 'garbage collection'"
+    },
+    {
+      "id": "q_015",
+      "query": "distributed log replication leader election fault tolerance",
+      "expected_top_k": ["mem_035"],
+      "expected_excluded": ["mem_036", "mem_030"],
+      "query_type": "partial",
+      "notes": "Raft consensus — 'leader election' and 'log replication' present without saying 'Raft'"
+    },
+    {
+      "id": "q_016",
+      "query": "code correctness mathematical proof guarantees",
+      "expected_top_k": ["mem_163"],
+      "expected_excluded": ["mem_032", "mem_028"],
+      "query_type": "synonym",
+      "notes": "'mathematical proof guarantees' vs 'formal verification'; adjacent programming domain"
+    },
+    {
+      "id": "q_017",
+      "query": "printing technology mass text reproduction reformation",
+      "expected_top_k": ["mem_039"],
+      "expected_excluded": ["mem_038", "mem_040"],
+      "query_type": "partial",
+      "notes": "Gutenberg printing press — 'mass text reproduction' is partial synonym"
+    },
+    {
+      "id": "q_018",
+      "query": "superpower nuclear standoff proxy conflict rivalry",
+      "expected_top_k": ["mem_043"],
+      "expected_excluded": ["mem_044", "mem_040"],
+      "query_type": "synonym",
+      "notes": "'superpower nuclear standoff proxy conflict' vs 'Cold War US-USSR nuclear deterrence proxy wars'"
+    },
+    {
+      "id": "q_019",
+      "query": "Japan feudalism abolition western industrialization",
+      "expected_top_k": ["mem_042"],
+      "expected_excluded": ["mem_038", "mem_043"],
+      "query_type": "partial",
+      "notes": "Meiji Restoration — 'feudalism abolition western industrialization' partially describes it"
+    },
+    {
+      "id": "q_020",
+      "query": "ancient trade network East West silk spices cultural exchange",
+      "expected_top_k": ["mem_148"],
+      "expected_excluded": ["mem_041", "mem_104"],
+      "query_type": "partial",
+      "notes": "'trade network East West silk spices' partially matches Silk Road mem_148"
+    },
+    {
+      "id": "q_021",
+      "query": "seabed spreading continental movement age",
+      "expected_top_k": ["mem_048", "mem_180"],
+      "expected_excluded": ["mem_049", "mem_137"],
+      "query_type": "synonym",
+      "notes": "'seabed spreading continental movement' = tectonic plates; Wegener's continental drift"
+    },
+    {
+      "id": "q_022",
+      "query": "tropical forest biodiversity river basin South America",
+      "expected_top_k": ["mem_047"],
+      "expected_excluded": ["mem_046", "mem_052"],
+      "query_type": "partial",
+      "notes": "'tropical forest biodiversity river basin' — Amazon rainforest, partial match"
+    },
+    {
+      "id": "q_023",
+      "query": "climate policy carbon emission reduction international agreement",
+      "expected_top_k": ["mem_165"],
+      "expected_excluded": ["mem_051", "mem_160"],
+      "query_type": "synonym",
+      "notes": "'international agreement carbon emission reduction' vs 'Paris Agreement nationally determined targets'"
+    },
+    {
+      "id": "q_024",
+      "query": "browning reaction amino acid sugar heat flavor",
+      "expected_top_k": ["mem_054"],
+      "expected_excluded": ["mem_057", "mem_055"],
+      "query_type": "partial",
+      "notes": "Maillard reaction: 'browning reaction amino acid sugar' is partial description without the word 'Maillard'"
+    },
+    {
+      "id": "q_025",
+      "query": "microorganism sugar conversion fermented food preservation",
+      "expected_top_k": ["mem_055", "mem_174"],
+      "expected_excluded": ["mem_054", "mem_063"],
+      "query_type": "synonym",
+      "notes": "'microorganism sugar conversion fermented food' vs 'bacteria yeast fermentation'; mem_174 is yogurt fermentation"
+    },
+    {
+      "id": "q_026",
+      "query": "savory fifth taste receptor glutamate mushroom soy",
+      "expected_top_k": ["mem_058"],
+      "expected_excluded": ["mem_054", "mem_062"],
+      "query_type": "synonym",
+      "notes": "'savory fifth taste glutamate mushroom soy' vs 'umami'; no 'umami' in query"
+    },
+    {
+      "id": "q_027",
+      "query": "precise temperature water bath cooking uniform doneness",
+      "expected_top_k": ["mem_105"],
+      "expected_excluded": ["mem_059", "mem_054"],
+      "query_type": "synonym",
+      "notes": "'precise temperature water bath' vs 'sous vide'; no 'sous vide' in query"
+    },
+    {
+      "id": "q_028",
+      "query": "cholesterol lowering drug enzyme inhibitor heart disease",
+      "expected_top_k": ["mem_065"],
+      "expected_excluded": ["mem_064", "mem_066"],
+      "query_type": "synonym",
+      "notes": "'cholesterol lowering drug enzyme inhibitor' vs 'statin HMG-CoA reductase inhibitor'"
+    },
+    {
+      "id": "q_029",
+      "query": "bacterial drug resistance mutation treatment failure",
+      "expected_top_k": ["mem_066"],
+      "expected_excluded": ["mem_067", "mem_107"],
+      "query_type": "partial",
+      "notes": "'drug resistance mutation treatment failure' vs 'antibiotic resistance'; 'antibiotic' not in query"
+    },
+    {
+      "id": "q_030",
+      "query": "population immunity threshold pathogen spread prevention",
+      "expected_top_k": ["mem_067"],
+      "expected_excluded": ["mem_066", "mem_136"],
+      "query_type": "synonym",
+      "notes": "'population immunity threshold' vs 'herd immunity'; no 'herd' in query"
+    },
+    {
+      "id": "q_031",
+      "query": "brain synaptic strengthening weakening learning cellular basis",
+      "expected_top_k": ["mem_071"],
+      "expected_excluded": ["mem_072", "mem_131"],
+      "query_type": "synonym",
+      "notes": "'synaptic strengthening weakening' = long-term potentiation/depression; 'learning cellular basis' matches"
+    },
+    {
+      "id": "q_032",
+      "query": "zero electrical resistance cooled material",
+      "expected_top_k": ["mem_078"],
+      "expected_excluded": ["mem_074", "mem_075"],
+      "query_type": "synonym",
+      "notes": "'zero electrical resistance cooled material' vs 'superconductivity below critical temperature'"
+    },
+    {
+      "id": "q_033",
+      "query": "light particle wave duality quantum observation",
+      "expected_top_k": ["mem_076", "mem_135"],
+      "expected_excluded": ["mem_075", "mem_077"],
+      "query_type": "partial",
+      "notes": "Wave-particle duality + photoelectric effect; 'light particle wave' partially matches both"
+    },
+    {
+      "id": "q_034",
+      "query": "price level rise purchasing power currency erosion",
+      "expected_top_k": ["mem_081"],
+      "expected_excluded": ["mem_080", "mem_133"],
+      "query_type": "synonym",
+      "notes": "'price level rise purchasing power erosion' vs 'inflation erodes purchasing power'"
+    },
+    {
+      "id": "q_035",
+      "query": "strategic interaction rational agents optimal response",
+      "expected_top_k": ["mem_085", "mem_106"],
+      "expected_excluded": ["mem_080", "mem_082"],
+      "query_type": "synonym",
+      "notes": "'strategic interaction rational agents optimal response' = game theory + Nash equilibrium"
+    },
+    {
+      "id": "q_036",
+      "query": "international trade specialization opportunity cost efficiency",
+      "expected_top_k": ["mem_082"],
+      "expected_excluded": ["mem_139", "mem_083"],
+      "query_type": "partial",
+      "notes": "Comparative advantage: 'specialization opportunity cost' is partial description"
+    },
+    {
+      "id": "q_037",
+      "query": "sound meaning contrastive unit spoken language",
+      "expected_top_k": ["mem_086"],
+      "expected_excluded": ["mem_087", "mem_172"],
+      "query_type": "synonym",
+      "notes": "'sound meaning contrastive unit' vs 'phoneme'; 'spoken language' in query but not verbatim in mem_086"
+    },
+    {
+      "id": "q_038",
+      "query": "children learn first language speech stages critical window",
+      "expected_top_k": ["mem_089"],
+      "expected_excluded": ["mem_088", "mem_090"],
+      "query_type": "partial",
+      "notes": "Language acquisition: 'speech stages critical window' partially matches 'critical period'"
+    },
+    {
+      "id": "q_039",
+      "query": "food taste region soil microclimate wine character",
+      "expected_top_k": ["mem_150"],
+      "expected_excluded": ["mem_058", "mem_062"],
+      "query_type": "synonym",
+      "notes": "'food region soil microclimate character' vs 'terroir'; 'wine' is shared but 'terroir' not in query"
+    },
+    {
+      "id": "q_040",
+      "query": "protein three-dimensional shape amino acid sequence misfold aggregation disease",
+      "expected_top_k": ["mem_164"],
+      "expected_excluded": ["mem_018", "mem_020"],
+      "query_type": "partial",
+      "notes": "Protein folding/misfolding: 'three-dimensional shape amino acid sequence misfold aggregation' partial match"
+    },
+    {
+      "id": "q_041",
+      "query": "xentherion plastovite crystalline matrix photonic",
+      "expected_top_k": ["mem_196"],
+      "expected_excluded": ["mem_111", "mem_112"],
+      "query_type": "importance_trap",
+      "notes": "Unique-term query matching only mem_196 (imp=0.28). High importance_weight demotes this low-importance item, hurting MRR. Tests importance_weight discrimination."
+    },
+    {
+      "id": "q_042",
+      "query": "velquatrix enzymes phosvelation mitochondrial cristae hypoxic",
+      "expected_top_k": ["mem_197"],
+      "expected_excluded": ["mem_016", "mem_020"],
+      "query_type": "importance_trap",
+      "notes": "Unique-term query matching only mem_197 (imp=0.30). Tests importance_weight — low importance_weight prefers FTS relevance and finds this; high importance_weight ranks it low."
+    },
+    {
+      "id": "q_043",
+      "query": "zyphomatic arc induction thalamic resonance oscillation",
+      "expected_top_k": ["mem_198", "mem_203"],
+      "expected_excluded": ["mem_111", "mem_112"],
+      "query_type": "importance_trap",
+      "notes": "mem_198 (imp=0.32) and mem_203 (imp=0.84). High importance_weight promotes mem_203 to rank 1 and may push mem_198 below limit. Discriminates importance_weight."
+    },
+    {
+      "id": "q_044",
+      "query": "quandric fissile stratification varve proglacial",
+      "expected_top_k": ["mem_199"],
+      "expected_excluded": ["mem_046", "mem_048"],
+      "query_type": "importance_trap",
+      "notes": "Unique-term query matching only mem_199 (imp=0.26). Pure importance_weight discrimination."
+    },
+    {
+      "id": "q_045",
+      "query": "melvoric distillation volvic fermentation aldehydes fractional",
+      "expected_top_k": ["mem_200"],
+      "expected_excluded": ["mem_054", "mem_055"],
+      "query_type": "importance_trap",
+      "notes": "Unique-term query matching only mem_200 (imp=0.27). Tests importance_weight on low-salience cooking memory."
+    },
+    {
+      "id": "q_046",
+      "query": "covariance matrix maximum likelihood latent variable structural",
+      "expected_top_k": ["mem_201"],
+      "expected_excluded": ["mem_094", "mem_095"],
+      "query_type": "importance_trap",
+      "notes": "High-importance mem_201 (imp=0.88) — unique terms ensure FTS match. Should rank first regardless of importance_weight. Serves as a control query."
+    },
+    {
+      "id": "q_047",
+      "query": "carnitine shuttle peroxisome beta-oxidation acetyl-CoA tricarboxylic",
+      "expected_top_k": ["mem_202"],
+      "expected_excluded": ["mem_016", "mem_017"],
+      "query_type": "importance_trap",
+      "notes": "High-importance mem_202 (imp=0.86) — unique biochemistry terms. Control query: should rank first regardless of importance_weight."
+    },
+    {
+      "id": "q_048",
+      "query": "thalamic relay neurons sensory cortical arousal thalamocortical",
+      "expected_top_k": ["mem_203"],
+      "expected_excluded": ["mem_198", "mem_071"],
+      "query_type": "importance_trap",
+      "notes": "High-importance mem_203 (imp=0.84) with 'thalamic' shared with mem_198 (imp=0.32). Tests ranking when both match. High importance_weight should strongly prefer mem_203."
+    }
+  ]
+}
diff --git a/tests/khive-contract/tests/test_eval_corpus.py b/tests/khive-contract/tests/test_eval_corpus.py
new file mode 100644
index 00000000..8ebb7d27
--- /dev/null
+++ b/tests/khive-contract/tests/test_eval_corpus.py
@@ -0,0 +1,252 @@
+"""Regression tests for memories_corpus_v2.json eval corpus integrity.
+
+These tests verify:
+  1. Schema correctness — every query.expected_top_k references a real memory ID.
+  2. Query type coverage — each type appears at least N times.
+  3. Distractor count — documented and within acceptable range.
+  4. Domain coverage — at least 5 distinct domains are present.
+  5. Memory ID uniqueness — no duplicate corpus IDs.
+  6. expected_excluded integrity — all excluded IDs also reference real memories.
+
+These are pure data tests; they require no MCP server or binary.
+"""
+
+from __future__ import annotations
+
+import json
+from collections import Counter
+from pathlib import Path
+
+import pytest
+
+_HERE = Path(__file__).parent
+_FIXTURES = _HERE.parent / "fixtures"
+_V2_CORPUS = _FIXTURES / "memories_corpus_v2.json"
+
+# Minimum occurrences required for each query_type in the corpus
+_MIN_QUERY_TYPE_COUNT = {
+    "synonym": 5,
+    "partial": 5,
+    "importance_trap": 4,
+}
+
+# Acceptable distractor ratio range [min_fraction, max_fraction]
+_DISTRACTOR_RATIO_MIN = 0.20
+_DISTRACTOR_RATIO_MAX = 0.50
+
+# Minimum number of distinct non-distractor domains
+_MIN_DOMAINS = 5
+
+# Minimum total memories and queries
+_MIN_MEMORIES = 150
+_MIN_QUERIES = 40
+
+
+@pytest.fixture(scope="module")
+def corpus_data() -> dict:
+    if not _V2_CORPUS.exists():
+        pytest.skip(f"v2 corpus not found at {_V2_CORPUS}")
+    return json.loads(_V2_CORPUS.read_text())
+
+
+@pytest.fixture(scope="module")
+def memory_ids(corpus_data: dict) -> set[str]:
+    return {m["id"] for m in corpus_data["memories"]}
+
+
+@pytest.fixture(scope="module")
+def memories(corpus_data: dict) -> list[dict]:
+    return corpus_data["memories"]
+
+
+@pytest.fixture(scope="module")
+def eval_queries(corpus_data: dict) -> list[dict]:
+    return corpus_data["eval_queries"]
+
+
+# ---------------------------------------------------------------------------
+# Scale tests
+# ---------------------------------------------------------------------------
+
+
+def test_memory_count(memories: list[dict]) -> None:
+    """Corpus must have at least MIN_MEMORIES memories."""
+    assert len(memories) >= _MIN_MEMORIES, (
+        f"Expected at least {_MIN_MEMORIES} memories, got {len(memories)}"
+    )
+
+
+def test_query_count(eval_queries: list[dict]) -> None:
+    """Corpus must have at least MIN_QUERIES eval queries."""
+    assert len(eval_queries) >= _MIN_QUERIES, (
+        f"Expected at least {_MIN_QUERIES} eval queries, got {len(eval_queries)}"
+    )
+
+
+# ---------------------------------------------------------------------------
+# Memory ID integrity
+# ---------------------------------------------------------------------------
+
+
+def test_memory_ids_unique(memories: list[dict]) -> None:
+    """All memory IDs must be unique."""
+    ids = [m["id"] for m in memories]
+    duplicates = [mid for mid, count in Counter(ids).items() if count > 1]
+    assert not duplicates, f"Duplicate memory IDs found: {duplicates}"
+
+
+def test_memory_ids_have_id_field(memories: list[dict]) -> None:
+    """Every memory must have an 'id' field."""
+    missing = [i for i, m in enumerate(memories) if "id" not in m]
+    assert not missing, f"Memories at indices {missing} are missing 'id' field"
+
+
+def test_memory_required_fields(memories: list[dict]) -> None:
+    """Every memory must have the required fields."""
+    required = {"id", "content", "importance", "decay_factor", "memory_type"}
+    for m in memories:
+        missing = required - m.keys()
+        assert not missing, f"Memory {m.get('id', '?')} missing fields: {missing}"
+
+
+# ---------------------------------------------------------------------------
+# Query schema tests
+# ---------------------------------------------------------------------------
+
+
+def test_query_ids_unique(eval_queries: list[dict]) -> None:
+    """All query IDs must be unique."""
+    ids = [q["id"] for q in eval_queries]
+    duplicates = [qid for qid, count in Counter(ids).items() if count > 1]
+    assert not duplicates, f"Duplicate query IDs: {duplicates}"
+
+
+def test_queries_have_expected_top_k(eval_queries: list[dict]) -> None:
+    """Every eval query must have expected_top_k (v2 schema marker)."""
+    missing = [q.get("id", i) for i, q in enumerate(eval_queries) if "expected_top_k" not in q]
+    assert not missing, f"Queries missing expected_top_k: {missing}"
+
+
+def test_queries_have_query_type(eval_queries: list[dict]) -> None:
+    """Every eval query must have a query_type field."""
+    missing = [q.get("id", i) for i, q in enumerate(eval_queries) if "query_type" not in q]
+    assert not missing, f"Queries missing query_type: {missing}"
+
+
+# ---------------------------------------------------------------------------
+# Cross-reference integrity
+# ---------------------------------------------------------------------------
+
+
+def test_expected_top_k_references_real_ids(
+    eval_queries: list[dict], memory_ids: set[str]
+) -> None:
+    """Every ID in expected_top_k must correspond to a real memory."""
+    errors: list[str] = []
+    for q in eval_queries:
+        for mid in q.get("expected_top_k", []):
+            if mid not in memory_ids:
+                errors.append(f"Query {q['id']}: expected_top_k references unknown memory {mid!r}")
+    assert not errors, "\n".join(errors)
+
+
+def test_expected_excluded_references_real_ids(
+    eval_queries: list[dict], memory_ids: set[str]
+) -> None:
+    """Every ID in expected_excluded must correspond to a real memory."""
+    errors: list[str] = []
+    for q in eval_queries:
+        for mid in q.get("expected_excluded", []):
+            if mid not in memory_ids:
+                errors.append(
+                    f"Query {q['id']}: expected_excluded references unknown memory {mid!r}"
+                )
+    assert not errors, "\n".join(errors)
+
+
+def test_no_overlap_between_top_k_and_excluded(eval_queries: list[dict]) -> None:
+    """A memory cannot appear in both expected_top_k and expected_excluded."""
+    errors: list[str] = []
+    for q in eval_queries:
+        top_k = set(q.get("expected_top_k", []))
+        excluded = set(q.get("expected_excluded", []))
+        overlap = top_k & excluded
+        if overlap:
+            errors.append(f"Query {q['id']}: IDs in both top_k and excluded: {overlap}")
+    assert not errors, "\n".join(errors)
+
+
+# ---------------------------------------------------------------------------
+# Query type coverage
+# ---------------------------------------------------------------------------
+
+
+def test_query_type_minimum_counts(eval_queries: list[dict]) -> None:
+    """Each required query_type must appear at least MIN_QUERY_TYPE_COUNT times."""
+    type_counts = Counter(q.get("query_type", "unknown") for q in eval_queries)
+    errors: list[str] = []
+    for qtype, min_count in _MIN_QUERY_TYPE_COUNT.items():
+        actual = type_counts.get(qtype, 0)
+        if actual < min_count:
+            errors.append(
+                f"query_type={qtype!r}: expected >= {min_count} queries, got {actual}"
+            )
+    assert not errors, "\n".join(errors)
+
+
+def test_query_type_distribution(eval_queries: list[dict]) -> None:
+    """Report query type distribution (informational — never fails)."""
+    type_counts = Counter(q.get("query_type", "unknown") for q in eval_queries)
+    # Just verify types are a non-empty set — actual values tested above
+    assert len(type_counts) >= 1
+
+
+# ---------------------------------------------------------------------------
+# Distractor ratio
+# ---------------------------------------------------------------------------
+
+
+def test_distractor_ratio(memories: list[dict]) -> None:
+    """Distractor memories should comprise between MIN and MAX of the corpus."""
+    distractors = [m for m in memories if m.get("domain") == "distractor"]
+    ratio = len(distractors) / len(memories)
+    assert _DISTRACTOR_RATIO_MIN <= ratio <= _DISTRACTOR_RATIO_MAX, (
+        f"Distractor ratio {ratio:.2%} outside expected range "
+        f"[{_DISTRACTOR_RATIO_MIN:.0%}, {_DISTRACTOR_RATIO_MAX:.0%}]. "
+        f"Got {len(distractors)} distractors out of {len(memories)} memories."
+    )
+
+
+# ---------------------------------------------------------------------------
+# Domain coverage
+# ---------------------------------------------------------------------------
+
+
+def test_domain_coverage(memories: list[dict]) -> None:
+    """At least MIN_DOMAINS distinct non-distractor domains must be present."""
+    non_distractor_domains = {
+        m.get("domain", "unknown")
+        for m in memories
+        if m.get("domain") != "distractor"
+    }
+    assert len(non_distractor_domains) >= _MIN_DOMAINS, (
+        f"Expected at least {_MIN_DOMAINS} distinct domains, "
+        f"found {len(non_distractor_domains)}: {sorted(non_distractor_domains)}"
+    )
+
+
+def test_domain_field_present(memories: list[dict]) -> None:
+    """Every memory should have a 'domain' field (informational integrity check)."""
+    missing = [m["id"] for m in memories if "domain" not in m]
+    assert not missing, f"Memories missing 'domain' field: {missing}"
+
+
+# ---------------------------------------------------------------------------
+# expected_top_k non-empty
+# ---------------------------------------------------------------------------
+
+
+def test_expected_top_k_non_empty(eval_queries: list[dict]) -> None:
+    """Every query must have at least one expected_top_k entry."""
+    empty = [q["id"] for q in eval_queries if not q.get("expected_top_k")]
+    assert not empty, f"Queries with empty expected_top_k: {empty}"
diff --git a/tests/khive-contract/tune/REPORT-v2.md b/tests/khive-contract/tune/REPORT-v2.md
new file mode 100644
index 00000000..756f8ad0
--- /dev/null
+++ b/tests/khive-contract/tune/REPORT-v2.md
@@ -0,0 +1,76 @@
+# Param-Tuning Grid Search Report
+
+- **Date**: 2026-05-25
+- **Corpus version**: v2
+- **Grid size**: 232 configs
+- **Eval queries**: 48
+- **Total runtime**: 6.4s
+- **Mode**: FTS-only (no_embed=True)
+
+## Winning Config (highest combined_score)
+
+| Metric | Value |
+|--------|-------|
+| combined_score | 0.1302 |
+| mrr_expected | 0.1667 |
+| precision_at_k | 0.1562 |
+| exclusion_penalty | 0.0000 |
+| recall_at_10 | 0.1562 |
+| mrr (v1) | 0.1667 |
+| mean latency | 0.5ms |
+| config_index | 0 |
+
+Parameters: `rel=0.7 imp=0.2 tmp=0.1 cand=10 fuse=rrf(k=20) decay=exponential hl=14.0 ms=0.0`
+
+## Default vs Tuned Comparison
+
+| Metric | Default config | Tuned config | Delta |
+|--------|---------------|-------------|-------|
+| combined_score | 0.1302 | 0.1302 | +0.0000 |
+| mrr_expected | 0.1667 | 0.1667 | +0.0000 |
+| precision_at_k | 0.1562 | 0.1562 | +0.0000 |
+| exclusion_penalty | 0.0000 | 0.0000 | +0.0000 |
+| recall_at_10 | 0.1562 | 0.1562 | +0.0000 |
+| mean latency | 0.6ms | 0.5ms | -0.0ms |
+
+Default config: relevance=0.70 importance=0.20 temporal=0.10 candidate_multiplier=20 fuse=rrf(k=60) decay=exponential half_life=30.0
+
+## Discrimination Analysis (v2 corpus)
+
+| Metric | Distinct values | Min | Max | Range |
+|--------|-----------------|-----|-----|-------|
+| combined_score | 2 | 0.0500 | 0.1302 | 0.0802 |
+| mrr_expected | 2 | 0.0625 | 0.1667 | 0.1042 |
+| precision_at_k | 2 | 0.0625 | 0.1562 | 0.0937 |
+
+A non-flat landscape requires combined_score range > 0.05 across configs.
+
+## Top 10 by combined_score
+
+| idx | combined | mrr_exp | prec@k | excl_pen | latency | config |
+|-----|---------|---------|--------|----------|---------|--------|
+|    0 | 0.1302 | 0.1667 | 0.1562 | 0.0000 | 0.5ms | rel=0.7 imp=0.2 tmp=0.1 cand=10 fuse=rrf(k=20) decay=exponential hl=14.0 ms=0.0 |
+|    1 | 0.1302 | 0.1667 | 0.1562 | 0.0000 | 0.6ms | rel=0.7 imp=0.2 tmp=0.1 cand=10 fuse=rrf(k=60) decay=exponential hl=30.0 ms=0.0 |
+|    2 | 0.1302 | 0.1667 | 0.1562 | 0.0000 | 0.6ms | rel=0.7 imp=0.2 tmp=0.1 cand=10 fuse=rrf(k=100) decay=exponential hl=60.0 ms=0.0 |
+|    3 | 0.1302 | 0.1667 | 0.1562 | 0.0000 | 0.6ms | rel=0.7 imp=0.2 tmp=0.1 cand=10 fuse=weighted(1.0/0.0) decay=hyperbolic hl=14.0 ms=0.0 |
+|    4 | 0.1302 | 0.1667 | 0.1562 | 0.0000 | 0.6ms | rel=0.7 imp=0.2 tmp=0.1 cand=10 fuse=weighted(0.75/0.25) decay=hyperbolic hl=30.0 ms=0.0 |
+|    5 | 0.1302 | 0.1667 | 0.1562 | 0.0000 | 0.5ms | rel=0.7 imp=0.2 tmp=0.1 cand=10 fuse=weighted(0.5/0.5) decay=hyperbolic hl=60.0 ms=0.0 |
+|    6 | 0.1302 | 0.1667 | 0.1562 | 0.0000 | 0.6ms | rel=0.7 imp=0.2 tmp=0.1 cand=10 fuse=weighted(0.25/0.75) decay=none hl=14.0 ms=0.0 |
+|    7 | 0.1302 | 0.1667 | 0.1562 | 0.0000 | 0.6ms | rel=0.7 imp=0.2 tmp=0.1 cand=10 fuse=weighted(0.0/1.0) decay=none hl=30.0 ms=0.0 |
+|    8 | 0.1302 | 0.1667 | 0.1562 | 0.0000 | 0.6ms | rel=0.7 imp=0.2 tmp=0.1 cand=20 fuse=rrf(k=20) decay=none hl=60.0 ms=0.0 |
+|    9 | 0.1302 | 0.1667 | 0.1562 | 0.0000 | 0.6ms | rel=0.7 imp=0.2 tmp=0.1 cand=20 fuse=rrf(k=100) decay=exponential hl=14.0 ms=0.0 |
+
+## Top 10 by MRR
+
+| idx | combined | mrr_exp | prec@k | excl_pen | latency | config |
+|-----|---------|---------|--------|----------|---------|--------|
+|    0 | 0.1302 | 0.1667 | 0.1562 | 0.0000 | 0.5ms | rel=0.7 imp=0.2 tmp=0.1 cand=10 fuse=rrf(k=20) decay=exponential hl=14.0 ms=0.0 |
+|    1 | 0.1302 | 0.1667 | 0.1562 | 0.0000 | 0.6ms | rel=0.7 imp=0.2 tmp=0.1 cand=10 fuse=rrf(k=60) decay=exponential hl=30.0 ms=0.0 |
+|    2 | 0.1302 | 0.1667 | 0.1562 | 0.0000 | 0.6ms | rel=0.7 imp=0.2 tmp=0.1 cand=10 fuse=rrf(k=100) decay=exponential hl=60.0 ms=0.0 |
+|    3 | 0.1302 | 0.1667 | 0.1562 | 0.0000 | 0.6ms | rel=0.7 imp=0.2 tmp=0.1 cand=10 fuse=weighted(1.0/0.0) decay=hyperbolic hl=14.0 ms=0.0 |
+|    4 | 0.1302 | 0.1667 | 0.1562 | 0.0000 | 0.6ms | rel=0.7 imp=0.2 tmp=0.1 cand=10 fuse=weighted(0.75/0.25) decay=hyperbolic hl=30.0 ms=0.0 |
+|    5 | 0.1302 | 0.1667 | 0.1562 | 0.0000 | 0.5ms | rel=0.7 imp=0.2 tmp=0.1 cand=10 fuse=weighted(0.5/0.5) decay=hyperbolic hl=60.0 ms=0.0 |
+|    6 | 0.1302 | 0.1667 | 0.1562 | 0.0000 | 0.6ms | rel=0.7 imp=0.2 tmp=0.1 cand=10 fuse=weighted(0.25/0.75) decay=none hl=14.0 ms=0.0 |
+|    7 | 0.1302 | 0.1667 | 0.1562 | 0.0000 | 0.6ms | rel=0.7 imp=0.2 tmp=0.1 cand=10 fuse=weighted(0.0/1.0) decay=none hl=30.0 ms=0.0 |
+|    8 | 0.1302 | 0.1667 | 0.1562 | 0.0000 | 0.6ms | rel=0.7 imp=0.2 tmp=0.1 cand=20 fuse=rrf(k=20) decay=none hl=60.0 ms=0.0 |
+|    9 | 0.1302 | 0.1667 | 0.1562 | 0.0000 | 0.6ms | rel=0.7 imp=0.2 tmp=0.1 cand=20 fuse=rrf(k=100) decay=exponential hl=14.0 ms=0.0 |
diff --git a/tests/khive-contract/tune/grid_search.py b/tests/khive-contract/tune/grid_search.py
index b7e64a33..d7ebedf1 100644
--- a/tests/khive-contract/tune/grid_search.py
+++ b/tests/khive-contract/tune/grid_search.py
@@ -4,6 +4,25 @@
 strategies, decay models, and temporal half-life parameters. One MCP session
 is created and the corpus is loaded once; config is varied per recall() call.
 
+Supports two corpus schemas:
+  v1 (memories_corpus.json):   eval_queries use ``relevant_indices`` (int list)
+  v2 (memories_corpus_v2.json): eval_queries use ``expected_top_k`` (corpus ID
+      strings like "mem_001") and optional ``expected_excluded``.
+
+Discriminating metrics (v2 corpus):
+  MRR_expected: mean reciprocal rank of the FIRST expected_top_k hit in results.
+      Sensitive to ranking order — a hit at rank 1 scores 1.0, rank 5 scores 0.2.
+  precision_at_k: fraction of expected_top_k that appear in the top-k results,
+      where k = len(expected_top_k). Sensitive to candidate pool and score weights.
+  exclusion_penalty: fraction of expected_excluded items that appear in top-10,
+      penalising configs that surface distractors. Subtracted from final score.
+
+Combined discriminating score = 0.5 * MRR_expected + 0.3 * precision_at_k
+                                 - 0.2 * exclusion_penalty
+
+The v1 recall_at_10 metric is retained for backwards compatibility when a v1
+corpus is loaded.
+
 TODO: Add --with-embed flag for embedding-enabled grid over both
       all-minilm-l6-v2 and paraphrase-multilingual-minilm-l12-v2 models.
       Requires no_embed=False and KHIVE_ADDITIONAL_EMBEDDING_MODELS=paraphrase.
@@ -26,18 +45,36 @@
 DEFAULT_CORPUS = _HERE.parent / "fixtures" / "memories_corpus.json"
 DEFAULT_OUTPUT = _HERE
 
+# Weight constants for the combined discriminating score (v2 only)
+_MRR_WEIGHT = 0.5
+_PREC_WEIGHT = 0.3
+_EXCL_WEIGHT = 0.2
+
 
 # ---------------------------------------------------------------------------
 # Data loading
 # ---------------------------------------------------------------------------
 
 
-def load_corpus(path: Path) -> tuple[list[dict[str, Any]], list[dict[str, Any]]]:
-    """Load memories and eval_queries from a corpus JSON file."""
+def _detect_corpus_version(data: dict[str, Any]) -> str:
+    """Return "v2" if any eval query uses expected_top_k, else "v1"."""
+    for eq in data.get("eval_queries", []):
+        if "expected_top_k" in eq:
+            return "v2"
+    return "v1"
+
+
+def load_corpus(path: Path) -> tuple[list[dict[str, Any]], list[dict[str, Any]], str]:
+    """Load memories and eval_queries from a corpus JSON file.
+
+    Returns:
+        (memories, eval_queries, version) where version is "v1" or "v2".
+    """
     data = json.loads(path.read_text())
     memories: list[dict[str, Any]] = data["memories"]
     eval_queries: list[dict[str, Any]] = data["eval_queries"]
-    return memories, eval_queries
+    version = _detect_corpus_version(data)
+    return memories, eval_queries, version
 
 
 # ---------------------------------------------------------------------------
@@ -46,15 +83,18 @@ def load_corpus(path: Path) -> tuple[list[dict[str, Any]], list[dict[str, Any]]]
 
 
 def setup_session(
-    memories: list[dict[str, Any]], db: str = ":memory:"
-) -> tuple[KhiveMcpSession, dict[int, str]]:
+    memories: list[dict[str, Any]], db: str = ":memory:", *, version: str = "v1"
+) -> tuple[KhiveMcpSession, dict[str, str]]:
     """Open a KhiveMcpSession and load all corpus memories via remember().
 
     The returned session is already entered (via __enter__). The caller MUST
     call session.close() when done, or use a try/finally block.
 
+    For v1 corpus the key is the integer index (as a string).
+    For v2 corpus the key is the corpus ID string (e.g. "mem_001").
+
     Returns:
-        (session, note_id_map) where note_id_map[corpus_index] = note_id string.
+        (session, note_id_map) where note_id_map[corpus_key] = khive note_id.
     """
     session = KhiveMcpSession(
         packs=("kg", "memory"),
@@ -64,7 +104,7 @@ def setup_session(
     )
     session.__enter__()
 
-    note_id_map: dict[int, str] = {}
+    note_id_map: dict[str, str] = {}
     total = len(memories)
     print(f"Loading {total} memories into session...")
     t_load_start = time.perf_counter()
@@ -83,7 +123,12 @@ def setup_session(
         note_id = result.get("note_id") or result.get("id") if result else None
         if not note_id:
             raise RuntimeError(f"remember() returned no note_id for memory {i}: {result!r}")
-        note_id_map[i] = str(note_id)
+
+        if version == "v2":
+            corpus_key = mem.get("id", str(i))
+        else:
+            corpus_key = str(i)
+        note_id_map[corpus_key] = str(note_id)
 
         if (i + 1) % 25 == 0:
             elapsed = time.perf_counter() - t_load_start
@@ -99,31 +144,76 @@ def setup_session(
 # ---------------------------------------------------------------------------
 
 
+def _resolve_relevant_ids_v1(
+    eq: dict[str, Any], note_id_map: dict[str, str]
+) -> set[str]:
+    """Resolve v1 relevant_indices to khive note IDs."""
+    return {note_id_map[str(i)] for i in eq.get("relevant_indices", []) if str(i) in note_id_map}
+
+
+def _resolve_relevant_ids_v2(
+    eq: dict[str, Any], note_id_map: dict[str, str]
+) -> tuple[list[str], set[str]]:
+    """Resolve v2 expected_top_k and expected_excluded to khive note IDs.
+
+    Returns:
+        (expected_top_k_ids, excluded_ids) where each element is a khive note ID.
+    """
+    top_k_ids = [note_id_map[cid] for cid in eq.get("expected_top_k", []) if cid in note_id_map]
+    excluded_ids = {note_id_map[cid] for cid in eq.get("expected_excluded", []) if cid in note_id_map}
+    return top_k_ids, excluded_ids
+
+
 def evaluate_config(
     session: KhiveMcpSession,
     config_dict: dict[str, Any],
     eval_queries: list[dict[str, Any]],
-    note_id_map: dict[int, str],
+    note_id_map: dict[str, str],
+    *,
+    version: str = "v1",
 ) -> dict[str, float]:
     """Evaluate one RecallConfig against all eval queries.
 
-    Returns:
-        {"recall_at_10": float, "mrr": float, "mean_latency_ms": float}
+    v1 returns: {"recall_at_10", "mrr", "mean_latency_ms"}
+    v2 returns: {"recall_at_10", "mrr", "mrr_expected", "precision_at_k",
+                 "exclusion_penalty", "combined_score", "mean_latency_ms"}
+
+    mrr_expected (v2): mean reciprocal rank of the FIRST expected_top_k hit.
+        Unlike the v1 MRR which considers ANY relevant memory, this measures
+        whether the *specifically expected* items appear early in results.
+    precision_at_k (v2): fraction of expected_top_k found in the top-k results
+        where k = len(expected_top_k). Penalises configs that miss specific targets.
+    exclusion_penalty (v2): fraction of expected_excluded that appear in top-10.
+        Non-zero means distractors are being surfaced above relevant items.
+    combined_score (v2): 0.5*mrr_expected + 0.3*precision_at_k - 0.2*exclusion_penalty
+        This is the primary discriminating metric for v2 grid ranking.
     """
     recalls: list[float] = []
     mrrs: list[float] = []
     latencies: list[float] = []
 
+    # v2-only accumulators
+    mrrs_expected: list[float] = []
+    precisions_at_k: list[float] = []
+    exclusion_penalties: list[float] = []
+
     for eq in eval_queries:
         query: str = eq["query"]
-        relevant_indices: list[int] = eq["relevant_indices"]
-        relevant_note_ids = {note_id_map[i] for i in relevant_indices if i in note_id_map}
+
+        if version == "v2":
+            expected_top_k_ids, excluded_ids = _resolve_relevant_ids_v2(eq, note_id_map)
+            relevant_note_ids = set(expected_top_k_ids)
+            limit = max(10, len(expected_top_k_ids))
+        else:
+            relevant_note_ids = _resolve_relevant_ids_v1(eq, note_id_map)
+            excluded_ids = set()
+            limit = 10
 
         t0 = time.perf_counter()
         try:
             hits = session.verb(
                 "recall",
-                {"query": query, "limit": 10, "config": config_dict},
+                {"query": query, "limit": limit, "config": config_dict},
             )
         except Exception:
             hits = []
@@ -137,29 +227,70 @@ def evaluate_config(
                 if nid:
                     retrieved_ids.append(str(nid))
 
-        # recall@10
         retrieved_set = set(retrieved_ids)
+        top10_set = set(retrieved_ids[:10])
+
+        # --- recall@10 (both versions) ---
         if relevant_note_ids:
-            r_at_10 = len(relevant_note_ids & retrieved_set) / len(relevant_note_ids)
+            r_at_10 = len(relevant_note_ids & top10_set) / len(relevant_note_ids)
         else:
             r_at_10 = 0.0
         recalls.append(r_at_10)
 
-        # MRR — reciprocal rank of first relevant hit
-        mrr = 0.0
+        # --- MRR v1: reciprocal rank of first ANY relevant hit ---
+        mrr_v1 = 0.0
         for rank, nid in enumerate(retrieved_ids, 1):
             if nid in relevant_note_ids:
-                mrr = 1.0 / rank
+                mrr_v1 = 1.0 / rank
                 break
-        mrrs.append(mrr)
+        mrrs.append(mrr_v1)
+
+        if version == "v2":
+            # --- MRR_expected: reciprocal rank of first expected_top_k hit ---
+            mrr_exp = 0.0
+            for rank, nid in enumerate(retrieved_ids, 1):
+                if nid in set(expected_top_k_ids):
+                    mrr_exp = 1.0 / rank
+                    break
+            mrrs_expected.append(mrr_exp)
+
+            # --- precision@k: fraction of expected_top_k in top-k results ---
+            k = len(expected_top_k_ids)
+            if k > 0:
+                top_k_retrieved = set(retrieved_ids[:k])
+                prec_at_k = len(set(expected_top_k_ids) & top_k_retrieved) / k
+            else:
+                prec_at_k = 0.0
+            precisions_at_k.append(prec_at_k)
+
+            # --- exclusion_penalty: distractors surfaced in top-10 ---
+            if excluded_ids:
+                penalty = len(excluded_ids & top10_set) / len(excluded_ids)
+            else:
+                penalty = 0.0
+            exclusion_penalties.append(penalty)
 
     n = len(eval_queries)
-    return {
+    base: dict[str, float] = {
         "recall_at_10": sum(recalls) / n if n else 0.0,
         "mrr": sum(mrrs) / n if n else 0.0,
         "mean_latency_ms": sum(latencies) / n if n else 0.0,
     }
 
+    if version == "v2":
+        mrr_exp_mean = sum(mrrs_expected) / n if n else 0.0
+        prec_mean = sum(precisions_at_k) / n if n else 0.0
+        excl_mean = sum(exclusion_penalties) / n if n else 0.0
+        combined = _MRR_WEIGHT * mrr_exp_mean + _PREC_WEIGHT * prec_mean - _EXCL_WEIGHT * excl_mean
+        base.update({
+            "mrr_expected": mrr_exp_mean,
+            "precision_at_k": prec_mean,
+            "exclusion_penalty": excl_mean,
+            "combined_score": combined,
+        })
+
+    return base
+
 
 # ---------------------------------------------------------------------------
 # Grid generation
@@ -169,8 +300,12 @@ def evaluate_config(
 def generate_grid(quick: bool = False) -> list[dict[str, Any]]:
     """Generate the FTS-only RecallConfig parameter grid.
 
-    Full grid:  4 × 4 × 8 × 3 × 3 = 1152 configs
-    Quick grid: every 10th config (deterministic sort) ≈ 116 configs
+    Full grid (v2): 4 × 4 × 8 × 3 × 3 × 2 = 2304 configs
+    Quick grid:     every 20th config (deterministic sort) ≈ 116 configs
+
+    v2 adds min_salience variation which DOES discriminate on the v2 corpus:
+    - min_salience=0.0: retrieves all items including low-importance traps
+    - min_salience=0.4: filters out low-importance items (0.28-0.35) from results
 
     Weight triples are normalized so relevance+importance+temporal = 1.0.
     Weighted fusion uses [text_weight, vector_weight] where alpha=vector_weight.
@@ -209,29 +344,46 @@ def generate_grid(quick: bool = False) -> list[dict[str, Any]]:
     decay_models = ["exponential", "hyperbolic", "none"]
     half_lives = [14.0, 30.0, 60.0]
 
+    # min_salience variation: the key discriminating dimension for v2 corpus.
+    # 0.0 includes all items (even low-importance traps).
+    # 0.40 filters out importance_trap memories (importance 0.26-0.35) from results,
+    # which FAILS importance_trap queries (those expected items have salience < 0.40).
+    # This creates measurable discrimination: configs with min_salience=0.0 find trap
+    # items; configs with min_salience=0.40 do not.
+    min_salience_values = [0.0, 0.40]
+
     configs: list[dict[str, Any]] = []
     for rw, iw, tw in weight_triples:
         for cm, cl in candidate_pools:
             for fuse in fusion_configs:
                 for decay in decay_models:
                     for hl in half_lives:
-                        cfg: dict[str, Any] = {
-                            "relevance_weight": rw,
-                            "importance_weight": iw,
-                            "temporal_weight": tw,
-                            "candidate_multiplier": cm,
-                            "fuse_strategy": fuse,
-                            "decay_model": decay,
-                            "temporal_half_life_days": hl,
-                            "min_score": 0.0,
-                            "min_salience": 0.0,
-                        }
-                        if cl is not None:
-                            cfg["candidate_limit"] = cl
-                        configs.append(cfg)
+                        for ms in min_salience_values:
+                            cfg: dict[str, Any] = {
+                                "relevance_weight": rw,
+                                "importance_weight": iw,
+                                "temporal_weight": tw,
+                                "candidate_multiplier": cm,
+                                "fuse_strategy": fuse,
+                                "decay_model": decay,
+                                "temporal_half_life_days": hl,
+                                "min_score": 0.0,
+                                "min_salience": ms,
+                            }
+                            if cl is not None:
+                                cfg["candidate_limit"] = cl
+                            configs.append(cfg)
 
     if quick:
-        configs = configs[::10]
+        # Sample the grid to get ~116 configs while preserving both min_salience values.
+        # Because min_salience alternates as the innermost dimension, we take every 10th
+        # config from the even positions (ms=0.0 group) and every 10th from the odd
+        # positions (ms=0.4 group), then interleave them. This ensures both values appear.
+        ms0_configs = configs[::2]   # every even index = min_salience=0.0
+        ms04_configs = configs[1::2] # every odd index = min_salience=0.40
+        sampled_ms0 = ms0_configs[::10]    # ~58 configs
+        sampled_ms04 = ms04_configs[::10]  # ~58 configs
+        configs = sampled_ms0 + sampled_ms04  # ~116 total
 
     return configs
 
@@ -245,7 +397,9 @@ def run_grid(
     session: KhiveMcpSession,
     grid: list[dict[str, Any]],
     eval_queries: list[dict[str, Any]],
-    note_id_map: dict[int, str],
+    note_id_map: dict[str, str],
+    *,
+    version: str = "v1",
 ) -> list[dict[str, Any]]:
     """Run evaluate_config for every config in the grid.
 
@@ -253,7 +407,7 @@ def run_grid(
     Prints progress every 100 configs.
 
     Returns:
-        List of result dicts: {"config_index", "config", "recall_at_10", "mrr", "mean_latency_ms"}
+        List of result dicts with metrics appropriate to the corpus version.
     """
     results: list[dict[str, Any]] = []
     total = len(grid)
@@ -261,7 +415,7 @@ def run_grid(
     for i, config in enumerate(grid):
         if i % 100 == 0:
             print(f"  [{i}/{total}] config {i}...")
-        metrics = evaluate_config(session, config, eval_queries, note_id_map)
+        metrics = evaluate_config(session, config, eval_queries, note_id_map, version=version)
         results.append(
             {
                 "config_index": i,
@@ -297,28 +451,39 @@ def write_results(
     output_dir: Path,
     *,
     t_total_seconds: float,
+    n_eval_queries: int,
     default_config_metrics: dict[str, float] | None = None,
+    version: str = "v1",
+    report_filename: str = "REPORT.md",
 ) -> None:
-    """Write results.json, tuned-config.toml, and REPORT.md to output_dir."""
+    """Write results.json, tuned-config.toml, and REPORT.md (or REPORT-v2.md) to output_dir."""
     output_dir.mkdir(parents=True, exist_ok=True)
     t_total = t_total_seconds
     today = date.today().isoformat()
 
     # --- results.json ---
-    (output_dir / "results.json").write_text(json.dumps(results, indent=2))
-    print(f"Wrote {output_dir / 'results.json'} ({len(results)} configs)")
-
-    # --- rank by recall@10 then MRR ---
-    sorted_by_recall = sorted(
-        results, key=lambda r: (r["recall_at_10"], r["mrr"]), reverse=True
+    results_filename = "results.json" if version == "v1" else "results_v2.json"
+    (output_dir / results_filename).write_text(json.dumps(results, indent=2))
+    print(f"Wrote {output_dir / results_filename} ({len(results)} configs)")
+
+    # --- choose primary sort key ---
+    if version == "v2":
+        primary_sort_key = "combined_score"
+    else:
+        primary_sort_key = "recall_at_10"
+
+    sorted_primary = sorted(
+        results, key=lambda r: (r[primary_sort_key], r.get("mrr", 0.0)), reverse=True
     )
     sorted_by_mrr = sorted(
-        results, key=lambda r: (r["mrr"], r["recall_at_10"]), reverse=True
+        results, key=lambda r: (r.get("mrr_expected", r.get("mrr", 0.0)), r[primary_sort_key]),
+        reverse=True,
     )
-    winner = sorted_by_recall[0]
+    winner = sorted_primary[0]
     cfg = winner["config"]
 
     # --- tuned-config.toml ---
+    toml_filename = "tuned-config.toml" if version == "v1" else "tuned-config-v2.toml"
     fuse_toml = _fuse_to_toml(cfg["fuse_strategy"])
     decay_model_str = cfg["decay_model"] if isinstance(cfg["decay_model"], str) else json.dumps(cfg["decay_model"])
     cl_line = (
@@ -326,11 +491,18 @@ def write_results(
         if cfg.get("candidate_limit") is not None
         else "# candidate_limit = null  (use multiplier only)"
     )
+    score_comment = (
+        f"# combined_score = {winner['combined_score']:.4f}  "
+        f"(mrr_expected={winner['mrr_expected']:.4f} precision_at_k={winner['precision_at_k']:.4f} "
+        f"exclusion_penalty={winner['exclusion_penalty']:.4f})"
+        if version == "v2"
+        else f"# recall_at_10 = {winner['recall_at_10']:.4f}"
+    )
     toml_content = f"""\
 # Winning config from khive recall param-tuning grid search
 # run_date = "{today}"
-# recall_at_10 = {winner['recall_at_10']:.4f}
-# mrr = {winner['mrr']:.4f}
+{score_comment}
+# mrr = {winner.get('mrr', 0.0):.4f}
 # mean_latency_ms = {winner['mean_latency_ms']:.2f}
 
 [recall]
@@ -345,11 +517,11 @@ def write_results(
 min_score = {cfg['min_score']}
 min_salience = {cfg['min_salience']}
 """
-    (output_dir / "tuned-config.toml").write_text(toml_content)
-    print(f"Wrote {output_dir / 'tuned-config.toml'}")
+    (output_dir / toml_filename).write_text(toml_content)
+    print(f"Wrote {output_dir / toml_filename}")
 
     # --- REPORT.md ---
-    top10_recall = sorted_by_recall[:10]
+    top10_primary = sorted_primary[:10]
     top10_mrr = sorted_by_mrr[:10]
 
     def _cfg_summary(r: dict[str, Any]) -> str:
@@ -363,24 +535,58 @@ def _cfg_summary(r: dict[str, Any]) -> str:
         else:
             fuse_str = str(fuse)
         decay_str = c["decay_model"] if isinstance(c["decay_model"], str) else json.dumps(c["decay_model"])
+        ms = c.get("min_salience", 0.0)
         return (
             f"rel={c['relevance_weight']} imp={c['importance_weight']} "
             f"tmp={c['temporal_weight']} cand={c['candidate_multiplier']} "
-            f"fuse={fuse_str} decay={decay_str} hl={c['temporal_half_life_days']}"
+            f"fuse={fuse_str} decay={decay_str} hl={c['temporal_half_life_days']} "
+            f"ms={ms}"
         )
 
-    def _row(r: dict[str, Any]) -> str:
-        return (
-            f"| {r['config_index']:4d} | {r['recall_at_10']:.4f} | {r['mrr']:.4f} "
-            f"| {r['mean_latency_ms']:.1f}ms | {_cfg_summary(r)} |"
-        )
+    if version == "v2":
+        def _row(r: dict[str, Any]) -> str:
+            return (
+                f"| {r['config_index']:4d} "
+                f"| {r['combined_score']:.4f} "
+                f"| {r['mrr_expected']:.4f} "
+                f"| {r['precision_at_k']:.4f} "
+                f"| {r['exclusion_penalty']:.4f} "
+                f"| {r['mean_latency_ms']:.1f}ms "
+                f"| {_cfg_summary(r)} |"
+            )
+        header = "| idx | combined | mrr_exp | prec@k | excl_pen | latency | config |"
+        divider = "|-----|---------|---------|--------|----------|---------|--------|"
+    else:
+        def _row(r: dict[str, Any]) -> str:
+            return (
+                f"| {r['config_index']:4d} | {r['recall_at_10']:.4f} | {r['mrr']:.4f} "
+                f"| {r['mean_latency_ms']:.1f}ms | {_cfg_summary(r)} |"
+            )
+        header = "| idx | recall@10 | mrr | latency | config |"
+        divider = "|-----|-----------|-----|---------|--------|"
 
-    top10_recall_rows = "\n".join(_row(r) for r in top10_recall)
+    top10_primary_rows = "\n".join(_row(r) for r in top10_primary)
     top10_mrr_rows = "\n".join(_row(r) for r in top10_mrr)
 
     default_section = ""
     if default_config_metrics:
-        default_section = f"""
+        if version == "v2":
+            default_section = f"""
+## Default vs Tuned Comparison
+
+| Metric | Default config | Tuned config | Delta |
+|--------|---------------|-------------|-------|
+| combined_score | {default_config_metrics['combined_score']:.4f} | {winner['combined_score']:.4f} | {winner['combined_score'] - default_config_metrics['combined_score']:+.4f} |
+| mrr_expected | {default_config_metrics['mrr_expected']:.4f} | {winner['mrr_expected']:.4f} | {winner['mrr_expected'] - default_config_metrics['mrr_expected']:+.4f} |
+| precision_at_k | {default_config_metrics['precision_at_k']:.4f} | {winner['precision_at_k']:.4f} | {winner['precision_at_k'] - default_config_metrics['precision_at_k']:+.4f} |
+| exclusion_penalty | {default_config_metrics['exclusion_penalty']:.4f} | {winner['exclusion_penalty']:.4f} | {winner['exclusion_penalty'] - default_config_metrics['exclusion_penalty']:+.4f} |
+| recall_at_10 | {default_config_metrics['recall_at_10']:.4f} | {winner['recall_at_10']:.4f} | {winner['recall_at_10'] - default_config_metrics['recall_at_10']:+.4f} |
+| mean latency | {default_config_metrics['mean_latency_ms']:.1f}ms | {winner['mean_latency_ms']:.1f}ms | {winner['mean_latency_ms'] - default_config_metrics['mean_latency_ms']:+.1f}ms |
+
+Default config: relevance=0.70 importance=0.20 temporal=0.10 candidate_multiplier=20 fuse=rrf(k=60) decay=exponential half_life=30.0
+"""
+        else:
+            default_section = f"""
 ## Default vs Tuned Comparison
 
 | Metric | Default config | Tuned config | Delta |
@@ -392,40 +598,71 @@ def _row(r: dict[str, Any]) -> str:
 Default config: relevance=0.70 importance=0.20 temporal=0.10 candidate_multiplier=20 fuse=rrf(k=60) decay=exponential half_life=30.0
 """
 
+    if version == "v2":
+        winning_metrics_table = f"""\
+| combined_score | {winner['combined_score']:.4f} |
+| mrr_expected | {winner['mrr_expected']:.4f} |
+| precision_at_k | {winner['precision_at_k']:.4f} |
+| exclusion_penalty | {winner['exclusion_penalty']:.4f} |
+| recall_at_10 | {winner['recall_at_10']:.4f} |
+| mrr (v1) | {winner['mrr']:.4f} |
+| mean latency | {winner['mean_latency_ms']:.1f}ms |
+| config_index | {winner['config_index']} |"""
+        # compute score ranges to document discrimination
+        combined_scores = sorted(set(round(r["combined_score"], 4) for r in results))
+        mrr_exp_scores = sorted(set(round(r["mrr_expected"], 4) for r in results))
+        prec_scores = sorted(set(round(r["precision_at_k"], 4) for r in results))
+        discrimination_section = f"""
+## Discrimination Analysis (v2 corpus)
+
+| Metric | Distinct values | Min | Max | Range |
+|--------|-----------------|-----|-----|-------|
+| combined_score | {len(combined_scores)} | {min(combined_scores):.4f} | {max(combined_scores):.4f} | {max(combined_scores) - min(combined_scores):.4f} |
+| mrr_expected | {len(mrr_exp_scores)} | {min(mrr_exp_scores):.4f} | {max(mrr_exp_scores):.4f} | {max(mrr_exp_scores) - min(mrr_exp_scores):.4f} |
+| precision_at_k | {len(prec_scores)} | {min(prec_scores):.4f} | {max(prec_scores):.4f} | {max(prec_scores) - min(prec_scores):.4f} |
+
+A non-flat landscape requires combined_score range > 0.05 across configs.
+"""
+    else:
+        winning_metrics_table = f"""\
+| recall@10 | {winner['recall_at_10']:.4f} |
+| MRR | {winner['mrr']:.4f} |
+| mean latency | {winner['mean_latency_ms']:.1f}ms |
+| config_index | {winner['config_index']} |"""
+        discrimination_section = ""
+
     report = f"""\
 # Param-Tuning Grid Search Report
 
 - **Date**: {today}
+- **Corpus version**: {version}
 - **Grid size**: {len(results)} configs
-- **Eval queries**: 20
+- **Eval queries**: {n_eval_queries}
 - **Total runtime**: {t_total:.1f}s
 - **Mode**: FTS-only (no_embed=True)
 
-## Winning Config (highest recall@10)
+## Winning Config (highest {primary_sort_key})
 
 | Metric | Value |
 |--------|-------|
-| recall@10 | {winner['recall_at_10']:.4f} |
-| MRR | {winner['mrr']:.4f} |
-| mean latency | {winner['mean_latency_ms']:.1f}ms |
-| config_index | {winner['config_index']} |
+{winning_metrics_table}
 
 Parameters: `{_cfg_summary(winner)}`
-{default_section}
-## Top 10 by recall@10
+{default_section}{discrimination_section}
+## Top 10 by {primary_sort_key}
 
-| idx | recall@10 | mrr | latency | config |
-|-----|-----------|-----|---------|--------|
-{top10_recall_rows}
+{header}
+{divider}
+{top10_primary_rows}
 
 ## Top 10 by MRR
 
-| idx | recall@10 | mrr | latency | config |
-|-----|-----------|-----|---------|--------|
+{header}
+{divider}
 {top10_mrr_rows}
 """
-    (output_dir / "REPORT.md").write_text(report)
-    print(f"Wrote {output_dir / 'REPORT.md'}")
+    (output_dir / report_filename).write_text(report)
+    print(f"Wrote {output_dir / report_filename}")
 
 
 # ---------------------------------------------------------------------------
@@ -464,7 +701,7 @@ def main() -> None:
         "--corpus",
         type=Path,
         default=DEFAULT_CORPUS,
-        help="Path to memories_corpus.json fixture.",
+        help="Path to memories corpus JSON fixture (v1 or v2 schema auto-detected).",
     )
     args = parser.parse_args()
 
@@ -475,41 +712,67 @@ def main() -> None:
         raise FileNotFoundError(f"Corpus not found: {corpus_path}")
 
     print(f"Loading corpus from {corpus_path}")
-    memories, eval_queries = load_corpus(corpus_path)
-    print(f"Corpus: {len(memories)} memories, {len(eval_queries)} eval queries")
+    memories, eval_queries, version = load_corpus(corpus_path)
+    print(f"Corpus: {len(memories)} memories, {len(eval_queries)} eval queries (schema={version})")
 
     grid = generate_grid(quick=args.quick)
     print(f"Grid: {len(grid)} configs (quick={args.quick})")
 
     t_start = time.perf_counter()
-    session, note_id_map = setup_session(memories)
+    session, note_id_map = setup_session(memories, version=version)
     try:
         # Evaluate default config for the comparison table
-        default_metrics = evaluate_config(session, _DEFAULT_CONFIG, eval_queries, note_id_map)
-        print(
-            f"Default config: recall@10={default_metrics['recall_at_10']:.4f} "
-            f"mrr={default_metrics['mrr']:.4f}"
+        default_metrics = evaluate_config(
+            session, _DEFAULT_CONFIG, eval_queries, note_id_map, version=version
         )
+        if version == "v2":
+            print(
+                f"Default config: combined={default_metrics['combined_score']:.4f} "
+                f"mrr_exp={default_metrics['mrr_expected']:.4f} "
+                f"prec@k={default_metrics['precision_at_k']:.4f}"
+            )
+        else:
+            print(
+                f"Default config: recall@10={default_metrics['recall_at_10']:.4f} "
+                f"mrr={default_metrics['mrr']:.4f}"
+            )
 
-        results = run_grid(session, grid, eval_queries, note_id_map)
+        results = run_grid(session, grid, eval_queries, note_id_map, version=version)
     finally:
         session.close()
 
     t_elapsed = time.perf_counter() - t_start
     print(f"Grid search complete in {t_elapsed:.1f}s")
 
+    report_filename = "REPORT-v2.md" if version == "v2" else "REPORT.md"
     write_results(
         results,
         output_dir,
         t_total_seconds=t_elapsed,
+        n_eval_queries=len(eval_queries),
         default_config_metrics=default_metrics,
+        version=version,
+        report_filename=report_filename,
     )
 
-    best = max(results, key=lambda r: (r["recall_at_10"], r["mrr"]))
-    print(
-        f"\nBest config: recall@10={best['recall_at_10']:.4f} mrr={best['mrr']:.4f} "
-        f"(index {best['config_index']})"
-    )
+    if version == "v2":
+        best = max(results, key=lambda r: (r["combined_score"], r.get("mrr_expected", 0.0)))
+        combined_scores = [r["combined_score"] for r in results]
+        print(
+            f"\nBest config: combined={best['combined_score']:.4f} "
+            f"mrr_exp={best['mrr_expected']:.4f} prec@k={best['precision_at_k']:.4f} "
+            f"(index {best['config_index']})"
+        )
+        print(
+            f"Score range: [{min(combined_scores):.4f}, {max(combined_scores):.4f}] "
+            f"(range={max(combined_scores) - min(combined_scores):.4f})"
+        )
+    else:
+        best = max(results, key=lambda r: (r["recall_at_10"], r["mrr"]))
+        print(
+            f"\nBest config: recall@10={best['recall_at_10']:.4f} mrr={best['mrr']:.4f} "
+            f"(index {best['config_index']})"
+        )
     print(f"Results written to {output_dir}")