Skip to main content
The per-package config matrices each describe one package. This is the cross-package view. The anchor is the Rust / Arrow-native / fused -core kernel: it is the reference implementation, and the Python and TypeScript surfaces are ports that either dispatch to it (the fast path) or run a byte-identical pure-language fallback. So the sections below read against that reference — how the packages line up, which scorers a kernel actually backs vs which are fallback-only, where the two language ports diverge, and what’s missing where. Everything below the line is generated from the parity manifests + AST (scripts/gen_suite_matrix.py) and verified in CI, so it can’t drift.

​
Per-package surface

Live counts (introspected from each package) side by side. MCP / Exports / A2A are the same numbers the llms.txt/README gate locks; CLI is the Python command count from the parity manifest.

​
Compute substrate — the Rust -core kernel is the reference

The Rust / Arrow-native / fused -core kernels are the source of truth for scoring; each language surface either dispatches to the kernel (the fast path) or runs a byte-identical pure-language fallback. This tracks which scorers a kernel actually backs vs which are fallback-only. Sources: Python _NATIVE_SCORER_IDS (arrow bucket kernel) and TS WASM_COVERED_SCORERS (-core WASM), gated as the scorer_kernels api_parity surface. 19 of 22 scorers are kernel-backed (the reference fast path); the other 3 are pure-language fallbacks with no -core kernel.

​
Cross-language parity (Python ↔ TypeScript)

Per surface: values shared by both languages vs each language’s exclusives. A non-zero Python-only or TS-only is a declared cross-language gap (parity/<pkg>.yaml), not drift — but it is where the two ports diverge.

​
Gaps & asymmetries (auto-detected)

  • No A2A skills surface: goldenanalysis, infermap.
  • goldenmatch mcp_tools — Python-only: create_domain, documents_ingest, documents_suggest_schema, explain_routing, identity_audit, identity_audit_seal, identity_audit_verify, identity_profile, identity_show, identity_stats, identity_worklist, lint_routing, list_domains, list_plugins, plan_routing, pprl_auto_config, pprl_link, test_domain.
  • goldenmatch mcp_tools — TS-only: build_clusters, find_exact_matches, find_fuzzy_matches, read_file, score_pair, score_strings, server_info, write_csv.
  • goldenmatch cli_commands — Python-only: analyze-blocking, anomalies, autoconfig, compare-clusters, config, explain, incremental, ingest-docs, init, interactive, label, lineage, pprl, review, rollback, runs, schedule, sensitivity, serve-ui, setup, sync, unmerge, watch.
  • goldenmatch cli_commands — TS-only: tui.
  • goldenmatch a2a_skills — Python-only: analyze_blocking, compare_clusters, configure, documents_ingest, documents_suggest_schema, identity_audit, identity_audit_seal, identity_audit_verify, identity_show, incremental, list_runs, pprl, quality, retrieve_similar, review, rollback, schema_match, sensitivity.
  • goldenmatch a2a_skills — TS-only: agent_approve_reject, agent_deduplicate, agent_explain_cluster, agent_explain_pair, agent_match_sources, agent_review_queue, fix_quality, memory_import, scan_quality, score.
  • goldenmatch scorers — Python-only: date_diff, geo_haversine, numeric_diff.
  • goldencheck mcp_tools — Python-only: install_domain.
  • goldencheck cli_commands — Python-only: agent-serve, denial-constraints, init, learn, refs, review, scan-db, schedule, serve.
  • goldenflow cli_commands — Python-only: agent-serve, init, interactive, mcp-serve, schedule, serve, watch.
  • goldenpipe cli_commands — Python-only: interactive.
  • infermap cli_commands — Python-only: mcp-serve.