{ "schema": "hypercurve.performance-checkpoint.v1", "captured_on": "2026-08-02", "checkpoint": "exact parallel-pair components and residual replay", "source_revision": "fa838bd22e210dcf1c125fa96ec909101f6d788f", "previous_checkpoint_revision": "8fc9a70", "historical_binary_revision": "ea99fe6", "hypersolve_revision": "339d4d8b6109da2c814febce3da9f725c44d8e56", "scope": { "included": [ "general BezierParallel2 versus BezierParallel2 exact contacts", "same, reversed, and partial exact source-overlap transport", "positive-dimensional parameter-component saturation", "residual isolated-contact recovery after exact component division", "algebraic-pair replay with refined tensor-Bernstein rejection", "exactly preconditioned Poincare-Miranda box existence proofs", "STRICT and APPROXIMATE_512 policy matrices", "historical wall-time, release-size, heap, and call-graph guards" ], "excluded": [ "CurveRegion2 traversal of algebraic-endpoint parallel fragments", "raw offset join construction", "offset arrangement regularization", "fillet and chamfer", "hypermesh" ] }, "mathematical_contract": { "carrier_pair": "q1(t)=p(t)+d*Np(t), q2(u)=r(u)+e*Nr(u), with selected signed unit normals and exact source/rational differential data", "source_overlap_selection": "same-direction source overlap is selected exactly iff d=e; reversed overlap is selected exactly iff d=-e", "component_rule": "a common parameter component is removed only after exact division from both equations; the divided residual system remains authoritative for isolated contacts", "non_ph_rule": "source-overlap component disposal is allowed only after both regular rational sources are decisively non-PH", "projected_pairing": "each retained coordinate interval isolates exactly one projected resultant root; exact box existence plus equation replay fixes the coupled pair", "box_certificate": "affine restriction, strict tensor-Bernstein face signs, and an exact nonzero midpoint-Jacobian adjugate preconditioner establish Poincare-Miranda existence without sampling", "independent_tangent_shortcut": "when source tangents are provably independent, the two selected signed projection equations uniquely determine the separation vector, so their squared norm eliminant is redundant", "policy": "STRICT and APPROXIMATE_512 execute the same exact projection, saturation, isolation, and replay graph; only an inherited equality predicate may terminate at the 512-bit policy boundary, and returned geometry remains exact" }, "implementation": { "hypersolve_substrate": [ "fde0d28: accelerated bounded rational-curve resultants", "339d4d8: equal-degree eliminated systems through degree 12 use an n-by-n polynomial Bezout matrix instead of a 2n-by-2n Sylvester matrix" ], "exact_replay": [ "structural u-t and t+u-1 source components", "generic Hypersolve bivariate component extraction and repeated exact division", "residual projection and coupled algebraic replay", "progressive exact isolator refinement shared by all feature modes", "selected-branch box sign before heavier algebraic lift replay" ], "probe_safety": "bivariate_pair_may_have_component remains a nonauthoritative fast probe and is permitted false negatives; a degenerate resultant still enters exact component authority", "compatibility_shims": 0, "public_carrier_growth_bytes": 0 }, "benchmark": { "method": "all-feature release offset benchmark pinned to CPU 11; timings are elapsed wall time per exact operation", "representative_complete_group": { "general_quadratic_candidates_milliseconds": 30.012359, "general_quadratic_contacts_milliseconds": 31.676382, "partial_source_overlap_candidates_microseconds": 292.304, "partial_source_overlap_contacts_microseconds": 292.014, "false_source_component_candidates_microseconds": 210.474, "false_source_component_contacts_milliseconds": 1.450538, "residual_algebraic_contact_candidates_milliseconds": 2.706617, "residual_algebraic_contact_contacts_milliseconds": 730.318452 }, "matched_historical_guard": { "method": "three alternating pinned pairs with byte-identical benchmark source and a shared Hypersolve substrate", "historical_curve_revision": "ea99fe6", "general_candidates_historical_median_milliseconds": 29.346, "general_candidates_current_median_milliseconds": 29.851, "general_candidates_change_percent": 1.72, "general_contacts_historical_median_milliseconds": 30.402, "general_contacts_current_median_milliseconds": 31.015, "general_contacts_change_percent": 2.02, "interpretation": "the already-supported isolated general contact remains near-neutral; the new exact component probe and box replay add about two percent while closing component and overlap classes the historical revision could not return completely" } }, "memory": { "tool": "Heaptrack direct application trace", "artifact": "/tmp/hypercurve-pair-current-direct.heaptrack.zst", "workload": "complete mixed-iteration bezier-pair-intersection benchmark group under instrumentation", "runtime_seconds": 90.706, "allocation_calls": 187853454, "peak_heap_mib": 7.33, "recorder_inclusive_peak_rss_mib": 20.37, "interpretation": "peak live memory is bounded, but exact resultant/component replay still creates substantial short-lived allocation churn and remains a measured optimization target" }, "release_size": { "artifact": "all-feature release offset benchmark with byte-identical benchmark source", "historical_file_bytes": 7488504, "current_file_bytes": 7533792, "file_change_bytes": 45288, "file_change_percent": 0.6048, "historical_text_bytes": 5720702, "current_text_bytes": 5757274, "text_change_bytes": 36572, "text_change_percent": 0.6393, "historical_data_bytes": 266216, "current_data_bytes": 267024, "data_change_bytes": 808, "historical_bss_bytes": 3080, "current_bss_bytes": 2576, "bss_change_bytes": -504, "interpretation": "the accepted completion class adds 0.64 percent text and no per-carrier storage" }, "static_callgraph": { "tool": "tools/hyper-callgraph", "command": "hyper-callgraph --root /home/tim/Documents/GitHub/workspace --out-dir /tmp/hypercurve-parallel-pair-callgraph --crate-name hypercurve,hypersolve --include-examples --include-tests --include-bench --include-fuzz --format json", "discovered_crates": 9, "previous_nodes": 30231, "current_nodes": 30433, "node_change": 202, "previous_edges": 54571, "current_edges": 55096, "edge_change": 525, "artifact": "/tmp/hypercurve-parallel-pair-callgraph/callgraph.json" }, "competitive_context": { "direct_guarantee_equivalent": false, "reason": "floating curve-offset libraries do not expose exact procedural parallel-pair contacts, positive-dimensional component saturation, or policy-identical algebraic replay", "historical_capability_gap": "ea99fe6 times the already-supported isolated general contact but cannot completely return the new partial-overlap and residual-after-component classes", "retained_controls": "the existing capsule and open-cubic floating offset comparisons remain end-to-end speed references; their weaker guarantees do not make them correctness oracles" }, "verification": { "all_features": "passed: 334 library tests plus every integration, adversarial, fuzz-corpus, documentation, and feature target", "no_default_features": "passed: 298 library tests plus every available integration and documentation target", "clippy_all_features": "passed with -D warnings", "clippy_no_default_features": "passed with -D warnings", "bench_check": "passed with all features", "format": "passed", "diff_check": "passed", "policy_regressions": [ "partial source overlap and exact reparameterization", "false same-source component removal", "residual isolated contact after component saturation", "coupled algebraic parallel/rational branch replay", "rotated transverse Poincare-Miranda preconditioning" ] }, "remaining": [ "retain represented and algebraic parameter ranges as canonical parallel fragments", "admit exact parallel carriers to CurveRegion2 traversal and Boolean dispatch", "construct round, bevel, and bounded-miter raw pieces", "regularize raw expansion, erosion, collapse, and self-contact topology through the sole arrangement", "solver-driven fillet and chamfer", "reduce exact resultant/component allocation churn", "remaining scalar towers and caller cutover" ] }