ZeroProofML v0.7.0a1 Documentation Hub¶

This documentation set tracks the breaking v0.7 typed Signed Common Meadow (SCM) alpha. Historical v0.6 containment reports remain available, while the new normative entry points are the migration guide and the semantics/ guides. The alpha is not experiment or deployment qualified. Published docs track the latest main branch at the site root.

Structure¶

• Files prefixed with 00_…–06_… are the core “read in order” guides.
• Higher-numbered files are reference, add-ons, and engineering notes.
• docs/theory/ contains deeper algebraic notes (also numbered).
• examples/archive_tr/ contains legacy Transreal-era scripts for historical reference.

Core Guides¶

• migration/v0.6-to-v0.7.md — breaking migration checklist and compatibility behavior.
• semantics/profile-registry.md — machine-readable profile and hash contract.
• semantics/result-model.md — orthogonal scalar/tensor result axes.
• semantics/projective-gauge.md — canonical gauge and typed strict decoding.
• semantics/refinement.md — bounded exact-to-floating validation scope.
• semantics/limitations.md — alpha non-claims and unsupported surfaces.
• 00_getting_started.md — install, run the quickstart, and understand the new SCM primitives.
• 01_scm_foundations.md — algebraic background for signed common meadows and the absorptive bottom.
• 02_projective_learning.md — when and how to enable projective (N, D) mode.
• 03_gradient_policies.md — choosing masking/clamping/projective strategies.
• 04_loss_functions.md — implicit, margin, sign consistency, and rejection losses.
• 05_training_guide.md — trainer loop, mixed precision, and coverage heuristics.
• 06_inference_deployment.md — strict SCM semantics, safety margins, and export notes.

Reproduce The Paper¶

• 31_reproduce_the_paper.md — shortest supported paper rerun path, frozen artifact bundle, and what to retain from reruns.

Task Guides¶

• 32_tau_infer_guide.md — choosing and freezing tau_infer from held-out |Q| sweeps.
• 33_namespace_guide.md — when to use zeroproofml.* and when zeroproof.* compatibility imports are still appropriate.
• 34_masks_and_provenance_guide.md — how to interpret bottom_mask, gap_mask, and stable fault/semantic provenance diagnostics.
• 35_deployment_bundles_guide.md — exporting, validating, smoke-testing, and handing off ONNX bundles.
• 36_ros2_integration_guide.md — companion-workspace layout, launch paths, QoS presets, and telemetry.
• 37_visualization_reporting_guide.md — report regeneration, visualization helpers, and live telemetry exports.

Engineering Notes¶

• 42_testing_strategy.md — test classes, markers, determinism, and evidence tiers.
• 07_projective_training_dev.md — developer walk-through of tuple renormalization and SCM decode.
• 15_debug_logging.md — capturing trainer metrics via log_hook.
• 16_verification_report.md — what the v0.6.1 SCM code implements.
• 22_bottom_mask_provenance_design.md — short Q1 design memo selecting the heuristic-first provenance rollout.
• 23_experimental_provenance_guide.md — mask provenance guide covering stable fault/semantic diagnostics and the separate experimental FRU structural-provenance axis.
• 24_bottom_mask_provenance_q2_decision.md — historical Q2 review memo, now superseded for the v0.6.x fault/semantic mask set.
• 26_ros2_repo_layout_decision.md — repository-layout decision for the first ROS 2 beta companion workspace.
• 27_ros2_bundle_loading_decision.md — bundle-loading decision for the first ROS 2 beta node: startup params first, reload service deferred.
• 28_ros2_rmw_selection_decision.md — initial ROS 2 middleware decision: CycloneDDS first, second RMW deferred until the beta path is stable.
• 29_rest_grpc_bundle_service_decision.md — integration-surface decision: optional minimal REST adapter is worthwhile, gRPC deferred.
• 30_triton_inference_server_decision.md — inference-server decision: Triton-style serving is a downstream recipe candidate after ONNX Runtime stability, not a first-party runtime path yet.
• 41_gauge_policy_design_review.md — gauge-policy choices and the opt-in projective normalizer; raw denominator scores remain gauge-dependent across different conventions.
• 43_checkpoint_resume_contract.md — fail-closed checkpoint capability matrix and named resume profiles.

Reference Material¶

• 31_reproduce_the_paper.md — published-docs guide to the canonical paper rerun path; the repo root keeps a compact mirror in REPRODUCIBILITY.md.
• CHANGELOG.md — release notes and breaking changes.
• theory/00_overview.md — SCM algebra and weak signs.
• theory/01_fracterm_flattening.md — strict rational-expression flattening.
• theory/02_mathematical_framework.md — the implemented strict-inference and flattening contract.
• 08_api_reference.md — public API surface summary for SCM, autodiff, and projective helpers.
• 09_bridge_summary.md — scalar IEEE bridge and vectorised SCM masks.
• 10_bridge_extended.md — worked examples for NumPy/Torch masks and projective heads.
• 11_integrations.md — integration patterns for NumPy/Torch/JAX.
• 12_float64_enforcement.md — dtype guidance and singular threshold notes.
• 17_experiment_protocol_v1.md — scientifically accurate paper-suite protocol.

Training Add-Ons¶

• 13_adaptive_loss_guide.md — coverage control and rejection loss wiring.
• 14_adaptive_loss_summary.md — quick summary of coverage/rejection.

Optimization¶

• 19_optimization_summary.md — performance levers and testing summary.
• 20_optimization_guide.md — deeper performance notes and profiling guidance.

Benchmarks & Examples¶

• Updated scripts in examples/ demonstrate SCM operations, projective tuples, and coverage control.
• 38_examples_inventory.md — current label map for every file under examples/.
• 39_examples_tutorials.md — promoted walkthroughs for the quickstart, coverage-control, bundle-export, and FRU strict-check examples.
• examples/fru_strict_check_demo.py shows a composed multi-stage rational head being flattened into one P/Q pair before strict_inference(...) checks the final denominator once. As of v0.6.1 the flatten path is operational-only: evaluate_fru_expression(..., strict=True) refuses float coefficients with FloatingCoefficientRefusal, while the default compatibility path warns; FRUExpression.flatten() refuses expansion above FLATTEN_MAX_WORK = 100_000 with FRUResourceRefusal.
• Scientific benchmark runners live in zeroproofml/benchmarks/; performance microbenchmarks live in perf/.
• Benchmarks overview: 18_benchmarks.md.
• Experiments & reproduction: 21_experiments.md.
• Benchmark artifact glossary: 40_benchmark_artifact_reference.md.

Archives¶

• Transreal and wheel-mode scripts now live in examples/archive_tr/ for historical reference only.

How to Navigate¶

• Start with Getting Started to verify your environment.
• Move to SCM Foundations before implementing custom layers or losses.
• Use the guides as a checklist when wiring projective heads or selecting gradient policies.