AU native numeric call-site gap report

This is the NE-153 production audit for the xp imports in agent_utilities/. It covers 36 production files and 75 unique xp attribute surfaces. The names below are the historical audit inventory; every bounded production surface now has an explicit disposition: a native-kernel call, a caller-owned builtin-list loop, a deterministic random adapter, or a versioned artifact seam. The AU adapter does not grow a replacement array runtime.

Exact audited surfaces

Production file	Surfaces observed
domains/finance/alpha_factors.py	isnan, log, spearmanr
domains/finance/composite_backtest.py	abs, array, asarray, cumprod, full, maximum, maximum.accumulate, mean, min, ones, prod, sqrt, std, zeros
domains/finance/cross_market_arb.py	exp, linalg, linalg.lstsq, log, ones, sqrt, std, vstack
domains/finance/execution.py	ks_2samp
domains/finance/kronos_forecaster.py	abs, any, array, maximum, minimum, percentile, where
domains/finance/market_data.py	cumsum, exp, random, random.default_rng
domains/finance/microstructure.py	mean
domains/finance/profit_attribution.py	corrcoef, cumprod, max, maximum, maximum.accumulate, mean, min, prod, sqrt, std, sum
domains/finance/research_autopilot.py	array, cumsum, maximum, maximum.accumulate, mean, min, random, random.default_rng, sqrt, std, sum
domains/finance/risk_manager.py	mean, norm_pdf, norm_ppf, percentile, random, random.default_rng, sort, std
domains/finance/signal_fusion.py	abs, asarray, atleast_2d, cov, eye, linalg, linalg.LinAlgError, linalg.pinv, linalg.solve
domains/finance/visual_ta.py	arange, max, mean, min, sum
graph/test_time_diversity.py	asarray, dot, float32, linalg, linalg.norm
harness/assimilation_benchmark.py	array, linalg, linalg.norm, log2, random, random.default_rng, vstack
harness/latent_efficiency_benchmark.py	eye, float32, linalg, linalg.norm
harness/superhuman_gate.py	asarray, float64, quantile, random, random.default_rng
knowledge_graph/assimilation/concept_matcher.py	argsort, asarray, float32, linalg, linalg.norm
knowledge_graph/core/analogy_engine.py	array, dot, linalg, linalg.norm
knowledge_graph/core/engine_tasks.py	dot, linalg, linalg.norm, mean
knowledge_graph/core/formal_reasoning_core.py	int64, linalg, linalg.matrix_power, linalg.norm, ones, random, random.default_rng, zeros
knowledge_graph/core/hypergraph.py	concatenate, dot, maximum, random, random.default_rng, sqrt, zeros
knowledge_graph/core/markov_regime.py	argsort, array, concatenate, cumsum, full, inf, prod, random, random.randint, zeros
knowledge_graph/core/optimal_execution.py	array, float64, mean, std, sum, tanh, zeros
knowledge_graph/core/semantic_subsumption.py	array, dot, linalg, linalg.norm
knowledge_graph/core/spectral_navigator.py	argmax, argmin, array, clip, diag, diff, eigsh, empty, eye, fill_diagonal, float64, linalg, linalg.eigh, linalg.norm, random, random.default_rng, sort, sqrt, where
knowledge_graph/core/world_model.py	asarray, concatenate, eye, float64, linalg, linalg.norm, linalg.solve, stack, zeros
knowledge_graph/distillation/deduplicator.py	array, float32, linalg, linalg.norm
knowledge_graph/distillation/lsh_index.py	array, dot, float32, linalg, linalg.norm, random, random.RandomState
knowledge_graph/extraction/fact_extractor.py	asarray
knowledge_graph/memory/optimization_engine.py	abs, arange, argmin, argsort, array, clip, dot, exp, float64, linalg, linalg.LinAlgError, linalg.norm, linalg.svd, log2, max, mean, newaxis, random, random.default_rng, sqrt, std, sum, triu_indices, var, where, zeros
knowledge_graph/orchestration/engine_finance.py	array, zeros
knowledge_graph/retrieval/capability_index.py	argsort, array, asarray, dot, float32, linalg, linalg.norm, load, save, stack, zeros
knowledge_graph/retrieval/generative_recommender.py	argsort, array, empty, float64, linalg, linalg.norm, vstack, zeros
knowledge_graph/retrieval/semantic_retrieval_engine.py	array, dot, linalg, linalg.norm, mean
knowledge_graph/retrieval/temporal_semantic_id.py	argmin, asarray, float64, linalg, linalg.norm, nan_to_num, random, random.default_rng, sum, vstack
mcp/tools/analysis_tools.py	cumprod, random, random.normal, roll

Final semantic dispositions

Gap family	Representative callers	Required owner
constructors/dtypes/constants	all constructor-heavy callers, including capability index and spectral clustering	Removed from AU; bounded builtin-list conversion and explicit loops
stateful/random module	market-data, research, harness, LSH, temporal IDs, analysis tools	_DeterministicRandom over seeded one-shot EG exports, including native batch choice/permutation; fixed seeds and list outputs
array arithmetic/indexing/broadcasting	capability index, memory optimization, recommender, spectral/world models	Bulk native matmul/kmeans where applicable, with caller-owned bounded mapping loops; no array runtime
covariance/persistence/object dtype	finance signal paths and capability index save/load	Bulk native covariance/matrix operations; save_numeric_artifact/load_numeric_artifact JSON seam with exact schema, bounds, and verified digest
missing scalar/vector helpers	roll/cumulative/quantile/log2-style callers	Explicit list loops or existing native reduction/elementwise calls
linalg compatibility	regression, fusion, spectral, world-model callers	Native solve/pinv/lstsq/svd/eigh; unsupported names remain typed failures

The focused facade tests cover both sides of this report: direct native calls (sum, sqrt, norm, where), deterministic random/list behavior including batch choice/permutation, bounded artifact round-trips with digest verification, and typed rejection of representative unsupported gaps (array, asarray, zeros, stack, save, load).