{ "schema": "hypercurve.performance-checkpoint.v1", "captured_on": "2026-08-02", "checkpoint": "policy-complete algebraic parameter ordering", "historical_hypercurve_revision": "35cf435403280bf2e0e41178008899c3177a65de", "candidate_hypercurve_revision": "5d2ad176c4667f70ed8a6447daa3deb1fe9a9a14", "historical_hypersolve_revision": "de63d0a8a7c2d70ecec7de87dc22c7a800ae5e89", "policy_hypersolve_revision": "78ec34c0f45fa60cfc56bcbcb211e6d2c5aed255", "candidate_hypersolve_revision": "1a857d780496ce5af8450e8ffddb7aea1487f956", "previous_checkpoint": "2026-08-02-topology-backed-carrier-aliases.json", "scope": { "included": [ "explicit STRICT or APPROXIMATE_512 policy on Hypersolve represented-root arithmetic", "exact translated-root difference proof beyond 512 bits", "policy-aware Hypercurve comparison and arithmetic adapters with aggregate certainty observation", "one shared represented-root comparison authority for arrangement, tangent, and rational-coordinate consumers", "STRICT algebraic parameter comparison through exact difference construction after bounded interval refinement", "APPROXIMATE_512 parameter equality only after its 512-bit terminal", "exact sign proof for unit algebraic isolators whose endpoint is zero", "matched historical runtime, allocation, release-size, and static-callgraph evidence" ], "excluded": [ "nonrational algebraic coefficient-field arithmetic beyond Hypersolve's bounded rational package", "remaining algebraic-cusp arrangement admission and distinct-carrier join work", "finite projection or export", "fillet and chamfer solving", "hypermesh" ] }, "mathematical_contract": { "strict": "STRICT never reaches an implicit approximate terminal. Overlapping parameter isolators receive 64 exact bisections, then exact represented-root difference construction; unsupported evidence remains explicit Ordering uncertainty.", "approximate_512": "APPROXIMATE_512 first traverses the strict proof path, continues interval refinement to the policy's 512-bit terminal only when exact proof remains unavailable, and records Approximate512Consumed exactly when that terminal supplies the decision.", "translation_proof": "For equal-degree exact-rational defining polynomials, normalized next-leading coefficients propose one root translation. The proposal is accepted only after affine transformation of the selected isolator and an exact common-root proof; the returned rational offset is therefore evidence rather than a coefficient heuristic.", "arithmetic_policy": "Every represented-root arithmetic construction receives the caller's explicit policy. Hypercurve retries policy-dependent undecided or validation results under APPROXIMATE_512 only after STRICT fails and records the weakened certainty before consuming a successful result.", "zero_endpoint_sign": "A unit isolator touching zero is strictly signed when the defining polynomial's exact constant coefficient is nonzero. If that coefficient is zero, unit isolation identifies the selected root as zero. A crossing isolator with nonzero constant remains undecided.", "unsupported_field": "Parameters with nonrational polynomial coefficients are not smuggled into the rational algebraic package. A nested-radical pair separated by 2^-600 remains uncertain under STRICT and reaches approximate equality only through APPROXIMATE_512." }, "capability": { "historical": "Two translated roots of rational quadratics separated by 2^-600 exhausted Hypercurve's local 64-step comparison without an authoritative exact difference. Four consumers also called Hypersolve with policy outcomes that Hypercurve could not account for in aggregate certainty.", "candidate": "The translated pair orders Less under both policies without consuming the approximate terminal. The deliberately unsupported nested-radical pair remains uncertain under STRICT and returns Equal with Approximate512Consumed under APPROXIMATE_512.", "policy_validation_fixture": "Unary negation of a valid represented root whose transformed isolator endpoints are distinguishable only beyond 512 bits fails STRICT validation, succeeds only under APPROXIMATE_512, and marks the enclosing curve operation accordingly.", "callers": [ "BezierParameter2 ordering", "Bezier algebraic coordinate comparison", "arrangement endpoint equality and tangent negation", "algebraic tangent scalar arithmetic and sign", "general rational Bezier coordinate equality" ] }, "benchmark": { "host": "AMD Ryzen 7 5800X3D", "toolchain": "rustc 1.97.0, LLVM 22.1.6, x86_64-unknown-linux-gnu", "matched_analytic_squares": { "method": "Callgrind on matched all-feature production benchmark artifacts; one warmup plus ten measured shared all-four CurveRegion2 Boolean batches", "command": "taskset -c 11 env HYPERCURVE_CURVE_REGION_BATCH_FIXTURE=analytic-squares HYPERCURVE_CURVE_REGION_BATCH_ITERATIONS=10 HYPERCURVE_CURVE_REGION_BATCH_MODE=shared valgrind --tool=callgrind ", "historical_instructions": 71626820, "candidate_instructions": 71661655, "instruction_change_percent": 0.048634, "interpretation": "The policy-complete algebraic path adds 34,835 deterministic instructions, below one twentieth of one percent, on the ordinary analytic Boolean guard." }, "matched_line_arc_adaptive": { "method": "three alternating all-feature processes per revision, each with eleven calibrated 100 ms samples pinned to CPU 11; values are medians of process medians", "historical_exact_batch_ns": 383712.9, "candidate_exact_batch_ns": 377705.3, "adaptive_change_percent": -1.565650, "interpretation": "The common native line/arc authority is neutral-to-faster; the policy additions are cold on this path." }, "competitive_line_arc": { "method": "one current all-feature process with eleven calibrated 100 ms samples pinned to CPU 11", "hypercurve_exact_batch_ns": 376728.6, "cavalier_four_finite_calls_ns": 9019.3, "exact_to_finite_ratio": 41.769162, "interpretation": "Cavalier is retained only as a throughput sentinel. Its finite polyline contract does not provide canonical-Real geometry, algebraic witnesses, exact degeneracy handling, or STRICT/APPROXIMATE_512 semantics." }, "exact_translation_fixture": { "method": "Callgrind process sentinel for the release unit regression, including the Rust test harness, construction, one STRICT comparison, and one APPROXIMATE_512 comparison", "process_instructions": 12711033, "result": "both comparisons return Less and the approximate operation remains Certified", "historical_result": "the exact translated-root order was unresolved, so this is a completion sentinel rather than a like-for-like timing" }, "hard_rational_replay": { "method": "serialized optimized integration regression with /usr/bin/time", "historical_test_seconds": 32.55, "candidate_test_seconds": 32.31, "test_time_change_percent": -0.737327, "historical_maxrss_kib": 233652, "candidate_maxrss_kib": 236464, "interpretation": "The multi-minute debug test becomes a practical optimized sentinel and shows no runtime regression; single-process maximum RSS is recorded but is too noisy to treat as a heap claim." } }, "memory": { "analytic_squares": { "tool": "Heaptrack direct release traces", "workload": "one warmup plus 25 measured shared all-four batches", "historical_profile": "/tmp/hypercurve-algebraic-policy-parent.heaptrack.zst", "candidate_profile": "/tmp/hypercurve-algebraic-policy-current.heaptrack.zst", "historical_allocation_calls": 132323, "candidate_allocation_calls": 132323, "historical_temporary_allocations": 9596, "candidate_temporary_allocations": 9596, "historical_peak_heap": "154.91 KiB", "candidate_peak_heap": "154.91 KiB", "interpretation": "The common production guard is exactly allocation- and peak-neutral." } }, "release_size": { "representative_production": { "method": "matched all-feature curve_region_boolean_batch release artifacts", "historical_file_bytes": 7849192, "candidate_file_bytes": 7875552, "file_change_percent": 0.335831, "historical_stripped_bytes": 6222552, "candidate_stripped_bytes": 6247912, "stripped_change_percent": 0.407550, "historical_aggregate_section_bytes": 6224474, "candidate_aggregate_section_bytes": 6249058, "aggregate_change_percent": 0.394957, "historical_text_section_bytes": 4428519, "candidate_text_section_bytes": 4451975, "text_change_percent": 0.529658, "historical_sha256": "82c3d21b9358ddda1d047dfdb21ce976181751d30e9d97e2240d5b8f915769ac", "candidate_sha256": "fdc3a0a90c6281b650318f6687a097c7031672e7ecc9ddb16a89b58bd17f19f2", "interpretation": "The bounded exact arithmetic, policy bridge, and regressions add 23,456 text bytes (0.53 percent) without growing any public carrier." } }, "static_callgraph": { "tool": "tools/hyper-callgraph", "command": "tools/hyper-callgraph/target/release/hyper-callgraph --root /home/tim/Documents/GitHub/workspace --out-dir /tmp/hypercurve-callgraph-algebraic-policy-5d2ad17 --crate-name hypercurve,hypersolve --include-examples --include-tests --include-bench --include-fuzz --format json", "discovered_crates": 10, "nodes": 32268, "previous_nodes": 32204, "edges": 59743, "previous_edges": 59606, "confirmed_paths": [ "BezierParameter2::cmp_by_refinement -> cmp_by_refinement_once -> compare_distinct_parameters -> compare_parameters_by_exact_difference", "compare_parameters_by_exact_difference -> Hypersolve::compare_algebraic_root_representations_by_difference -> represented_root_translation_difference", "compare_algebraic_representations_with_policy -> Hypersolve difference comparison -> CurveContext::observe_approximate_512", "arithmetic_algebraic_representations_with_policy -> Hypersolve represented-root arithmetic -> CurveContext::observe_approximate_512", "arrangement, tangent-order, and rational-Bezier consumers -> shared Hypercurve policy adapters" ] }, "verification": { "hypersolve_all_features": "passed: 210 unit, 2 README, 131 property, and 209 smoke tests", "hypercurve_all_feature_library": "passed: 375 tests serially", "hypercurve_affected_integrations": "passed: 9 algebraic-image, 26 algebraic-parameter, 58 arrangement, 9 tangent-order, 48 rational-Bezier, and 29 CurveRegion2 Boolean tests; the hard rational replay also passed separately in release", "clippy_all_features": "passed all targets with -D warnings in both crates", "clippy_no_default_features": "passed all targets with -D warnings in both crates", "builds": "passed all-target all-feature and no-default checks plus matched release benchmark builds", "format": "passed", "diff_check": "passed" }, "engineering_decisions": { "strict_refinement_budget": "Retained the historical 64 exact bisections for STRICT before symbolic difference proof. Raising STRICT to 512 made ordinary exact failures pay an arbitrary terminal-sized cost even though STRICT has no finite terminal.", "approximate_refinement_budget": "Only APPROXIMATE_512 continues through 512 refinements and then returns its policy-defined equality with recorded consumption.", "no_compatibility_shim": "Hypersolve's arithmetic function was changed in place to require policy at every caller; no old implicit-policy overload remains.", "next_authority": "Admit remaining algebraic-cusp subfragments and exact distinct-carrier joins through one CurveRegion2 arrangement and raw-offset topology engine." }, "remaining": [ "admit every retained algebraic-cusp subfragment through CurveRegion2 arrangement splitting and traversal", "replay selected positive-dimensional analytic-parallel/circle components into exact overlap ranges", "route general algebraic U-turn and distinct-carrier joins through the authoritative raw-offset graph", "complete finite projection/export and solver-driven fillet/chamfer work", "continue reducing exact polynomial and bigint allocation churn" ] }