Designated hot paths

Index of designated hot paths with Criterion bench targets, baseline docs, allocation/memory contracts, and owning crates.

Hot path	Owner crate	Bench	Baseline	Allocation / memory contract
Sample gather	antecedent-kernels	gather	gather.md	Dispatch entry; no per-index heap
Kernel reductions	antecedent-kernels	reductions	kernel_reductions.md	Reuse out/table buffers; scalar↔portable differential
Graph reachability	antecedent-graph	traversal	graph_traversal.md	Reusable GraphWorkspace
d-separation	antecedent-graph	dseparation	dseparation.md	Batch / witness APIs; workspace reuse
Adjustment search	antecedent-identify	adjustment	adjustment.md	Minimal-set enumeration budgets
Partial correlation batch	antecedent-kernels / antecedent-stats	partial_correlation	partial_correlation.md	Reusable ParCorrWorkspace
PCMCI discovery	antecedent-discovery	pcmci	pcmci.md	LaggedFrame + DiscoveryWorkspace; no per-CI plan rebuild
CI / orientation	antecedent-stats / antecedent-discovery	ci_framework, orientation	ci_orientation.md	Batch CI; mask complete-case
Propensity bootstrap	antecedent-estimate	propensity_bootstrap	propensity.md	Workspace buffer reuse across replicates
Progressive estimate execute	causal	(conformance) latency_tiers	—	StageClock + ProgressSink + StageResultSink payloads; effort on ExecutionPerformanceRecord
Cancel mid-bootstrap	antecedent-estimate / causal	(conformance) latency_tiers::cancel_mid_bootstrap	—	Soft partial SE; cancelled flag; no silent full result
Adaptive bootstrap	antecedent-estimate / causal	(conformance) latency_tiers::adaptive_bootstrap_pin	—	SE relative early-stop; early_stopped + actual bootstrap_replicates_ok
Adaptive Bayesian draws	antecedent-estimate / causal	(conformance) latency_tiers::adaptive_draws_pin	—	Quantile-width early-stop; early_stopped + actual n_draws
Prepared re-estimate	causal	(conformance) prepared_analysis	—	Compile-once Ready plan; schema-gated refresh; 2nd shot skips compile
Discover-once / estimate-many	causal / Python	(conformance) latency_tiers::interactive_refuses_inline_discovery, test_accepted_graph, test_discovery_interactive_guard	—	Interactive refuses Discover*; AcceptedGraph version stable across estimate clicks
Shared estimate→refute workspace	causal	(conformance) shared_workspace	—	StaticEstimateWorkspaces for linear / propensity / AIPW across estimate→refute
Interactive graph×effect subsample	antecedent-prob / causal	(unit) envelope::interactive_subsample_mass_accounting_honest	—	Leftover identified mass → unidentified_mass; approximate diagnostic
Arrow CDI interactive estimate	Python / antecedent-data	(conformance) test_arrow_interactive_smoke	—	Prefer CDI borrow under latency=interactive; pandas correct but not latency default
Post-ID column projection	antecedent-data / causal	(conformance) projection_wide	—	Wide sheet → gather T/Y/Z only; ATE matches; exec.project.columns diagnostic
Batch multi-query	causal / Python	(conformance) batch_analysis, test_analyze_many	—	One table ingest, N AverageEffect queries; match solo ATE
Refute second click	causal / Python	(conformance) refute_second_click, test_refute_second_click	—	Prepared estimate then refute(suite); ATE frozen; validation replaced
Matching index	antecedent-stats	matching	matching.md	Exact path ≤ 10k; retain index on compatible fits
m-separation / PAG orient	antecedent-graph / antecedent-discovery	mseparation, pag_orientation	pag.md	Sparse + stress fixtures
RPCMCI / temporal mediation	antecedent-discovery / antecedent-estimate	rpcmci, temporal_mediation	regime_mediation.md	Multi-env plans must not clone sibling series
Shapley attribution	antecedent-attribution	shapley	shapley.md	Coalition cache; exact size gates
Design ranking / state append	antecedent-design / antecedent-state	design_rank, state_append	design_state.md	MonteCarloBudget; CacheBudget refuse

Smoke commands

Criterion smokes used by feature gates (--test):

cargo bench -p antecedent-kernels --bench gather -- --test
cargo bench -p antecedent-kernels --bench reductions -- --test
cargo bench -p antecedent-graph --bench traversal -- --test
cargo bench -p antecedent-graph --bench dseparation -- --test
cargo bench -p antecedent-identify --bench adjustment -- --test
cargo bench -p antecedent-kernels --bench partial_correlation -- --test
cargo bench -p antecedent-discovery --bench pcmci -- --test
cargo bench -p antecedent-attribution --bench shapley -- --test
cargo bench -p antecedent-design --bench design_rank -- --test
cargo bench -p antecedent-state --bench state_append -- --test

Absolute timings in baseline files are machine-class references (Apple M1). Unexplained regressions beyond documented budgets block merge.