Research Overview Tracks Agenda
Docs › Research › SkeinDB True Status Matrix

SkeinDB True Status Matrix

Last updated: 2026-06-13

This is the short truth surface. It is intentionally not a changelog. Use it to answer: what is real today, what is partial, and what should not be claimed yet.

Current Truth Snapshot

  • Compatibility: SkeinDB does not claim full MySQL or PostgreSQL compatibility. MySQL coverage is broad and corpus-backed; PostgreSQL support is a partial PG v3 baseline. See docs/MYSQL_COMPAT.md and docs/PG_COMPAT.md.
  • Core roadmap: docs/PROJECT_BACKLOG.md has 140 done / 0 open top-level roadmap checkboxes. That does not mean every phase is production-complete; several phases remain partial or prototype-strength.
  • Research roadmap: docs/RESEARCH_BACKLOG.md has 109 done / 0 open research checkboxes. R01-R17 and R20 are hardened; R18 and R19 remain prototype implemented.
  • Latest verified slice: PostgreSQL CSV COPY now honors single-byte ESCAPE markers in addition to QUOTE across export and import on the live PG path, while the recent catalog parity work still includes pg_catalog.pg_description alongside pg_catalog.pg_am.

Do Not Overclaim

  • No "100% MySQL compatibility" or "100% PostgreSQL compatibility" claim.
  • No production-complete LSM/MANIFEST/WAL storage claim for the whole engine.
  • No claim that CDC has external sinks, cluster-wide fanout, broader predicates, or binary/columnar event encodings.
  • No claim that R18 performance replay or R19 Wasm query operators are hardened.
  • No claim that distribution packaging is fully published until signing/release secrets are configured.

Current Partial Areas

Area Truth today Remaining gap
Storage core The core on-disk primitives exist and are unit/integration tested in isolation: typed MANIFEST (append + replay), WAL (committed-only recovery + torn-tail truncation), RowSeg/RV1 + FilePtr MVCC chains, ValueStore + DELTA + learned indexes, Run/LSM blocks, rowdir, encryption envelopes, and the Wasm catalog/UDF sandbox (crates/skeindb-core/src/{manifest,wal,rowseg,valuestore,run,rowdir,mvcc,encrypted_valuestore}.rs). The live engine persists rows via JSON/segment (.rseg) snapshots with full in-memory table materialization (engine.rs open_with_storage_mode, persist_table, storage_stats_snapshot); Engine::core_lsm_files_active reports whether the core MANIFEST + wal-*.log files are present as observability for this gap. Storage-mode decisions are centralized in storage_mode.rs. The core primitives are not yet wired as the primary always-on row-persistence path: no MANIFEST/WAL/LSM-backed primary writes, no streaming/large-table reads (tables fully materialize in memory), and crash safety between persists is best-effort.
PostgreSQL compatibility PG listener, startup/auth, many SQL rewrites, catalog probes (pg_database, pg_namespace, pg_am, pg_description, pg_proc, pg_type, pg_settings, pg_stat_activity etc via virtual tables), COPY text/csv/binary with NULL / HEADER / HEADER MATCH / DELIMITER / QUOTE / ESCAPE, parenthesized and legacy bare WITH CSV|TEXT|BINARY format aliases, extended query (Parse/Bind/Execute), binary params for common OIDs, SQLSTATE mapping, and corpus coverage exist. Code: pg_wire.rs (framing, auth, COPY encode/decode), server.rs PG paths + virtual catalog, cluster_rpc.rs pg_*_roundtrip tests, tests/compat/pg_corpus.sql. Micro D (2026-06-11 prior + this round): pg_class + new pg_stat_user_tables (empty typed virtual table) in server.rs catalog handler + sql_exec test coverage (SELECT * returns typed cols + 0 rows). Not full PostgreSQL; broader COPY option coverage beyond current (e.g. more formats, OIDs, portal suspension), wider catalog parity (more pg_* tables), production driver matrices (Django, Rails, full psycopg etc.), broader dialect.
CDC/changefeeds Local table/query subscriptions work with polling, SSE, WebSocket, durable cursors, row images, objects_json/plain_json, op/pk/range/column filters, pause/resume, backpressure, and resnapshot signaling. Dependency expansion over views/CTE/union/set-ops. External sink config struct exists. Code: engine.rs cdc_* (event delivery, project, subscriptions v9), server.rs CDC RPC + sink drain. Tests in cluster_rpc and engine. (D micro also touches shared server PG catalog surface used by CDC consumers.) Broader predicates, binary/columnar encodings, external sink connectors implementation, cluster-wide fanout remain.
Compaction scheduler Live policy/status controls, safe-mode write backpressure, and a pressure-driven worker exist. Deeper multi-level/file-count compaction across many live tables remains future storage-engine work.
Distribution Debian metadata, Homebrew formula, and tag-driven release workflow scaffolding exist. Signed apt publication depends on configured secrets and release operations.
R18 performance replay Replay bundles carry performance profiles, deterministic replay rehydrates cache hints + runs variance, timing injection primitive + exercised in replay run + cache/LSM stats re-compute on rehydrate now implemented. Evidence: engine::tests::replay_timing_injection_simulates_deterministic_pacing, engine::tests::replay_pacing_applies_injected_delays_for_replay_exec_fidelity (new unit for pacing), engine::tests::replay_bundle_run_rehydrates_cache_hints (enhanced), engine::tests::replay_bundle_export_import_run_roundtrip, server::tests::maintenance_replay_run_rehydrates_cache_hints, cluster_rpc t183/t188 (updated for replay/RPC + timing pacing assert). Timing injection wired to simple pacing mechanism (apply_simulated_pacing called in maintenance_replay_run internal exec path using injected delays for deterministic sim); exercised in unit + RPC integration; stronger cache/LSM/timing fidelity. CI compare harness pre-exists in main.rs. Gaps: full real pacing in runner exec + .github CI distribution gates.
R19 Wasm query operators Compile/inspect/run/perf/edge-package surfaces exist with generated fixed-width Wasm artifacts and host fallback. Production SIMD-lowered codegen and hardened operator breadth are not claimed. Micro (2026-06-11): engine test now explicitly exercises wasm_plan_inspect + wasm_plan_run (host dispatch to execute_select) + row result parity for host_interpreted_v1 fallback artifact (str projection over inserted data). Evidence: engine::tests::wasm_plan_compile_falls_back_for_unsupported_types (expanded).

Recent Verified Changes

  • 2026-06-13: Removed dead "hardening" scaffolding that was carried only to be cited as evidence: the scan_rows_streaming_stub and select_streaming_row_path_stub / primary_row_extension / should_bootstrap_core_lsm_for_replay_materialize no-op helpers, the discarded ManifestReader::replay_state() and EncryptedValueStore::new() constructions in storage_stats_snapshot, and the replay-workspace RowSeg/MANIFEST/WAL bootstrap whose output nothing read back. The genuinely-wired pieces are retained and tested: storage_mode.rs mode decisions (uses_segment / expects_core_lsm_files / lsm_pipeline_files_active), Engine::core_lsm_files_active + its core_lsm_active stat, the compaction worker's uses_segment-based mode checks in server.rs, and the real replay export/import/run roundtrip. Evidence: engine::tests::storage_mode_decision_helpers, engine::tests::core_lsm_files_active_validates_manifest_header_with_core_primitive, engine::tests::replay_bundle_export_import_run_roundtrip, cluster_rpc::t183_replay_bundle_export_import_run_roundtrip. No on-disk format change; cargo fmt/clippy -D warnings/test --all clean.
  • 2026-06-11: R18 Performance Replay micro-slice (pacing wiring + fidelity): wired inject_replay_timing into simple pacing mechanism (apply_simulated_pacing using delays for deterministic sim in maintenance_replay_run internal replay exec; no real sleeps/IO). Added new unit test engine::tests::replay_pacing_applies_injected_delays_for_replay_exec_fidelity; updated engine::tests::replay_bundle_export_import_run_roundtrip + replay_bundle_run_rehydrates_cache_hints comments + cluster_rpc::t188_replay_run_rehydrates_cache_hints (integration exercising replay/RPC path + timing pacing assert on report). Evidence added to R18 row. Net <<200 LOC (small AGENTS slice), unit+RPC integration tests, no on-disk/format change. Deep dive done (engine replay fns, main CLI compare, tests, profiles). Followed by cargo fmt/clippy/test (replay), matrix+backlogs update, site rebuild (python3 scripts/build_docs_site.py), hard self code review, commit (ref review/matrix/R18/AGENTS), push. Per AGENTS.md. Gaps remain: full pacing in runner + CI workflow gates.
  • 2026-06-11: R18 Performance Replay micro-slice (timing injection + fidelity): added inject_replay_timing primitive (deterministic delay sim from profile p50/p95/p99 for pacing/variance) + exercised unconditionally in maintenance_replay_run (plus storage_stats_snapshot post-rehydrate for LSM fidelity) + new unit test engine::tests::replay_timing_injection_simulates_deterministic_pacing + enhancements/asserts in roundtrip + cache_hints tests. Covers timing injection gap + strengthens cache/LSM recon; pre-existing CI harness (replay compare + thresholds in main.rs) referenced. Net <50 LOC, unit+integration tests, no format change. Followed by fmt/clippy/test (replay-filtered + full relevant), matrix+backlog update, site rebuild, hard review, commit (ref review/matrix/R18). Per AGENTS. Gaps now: full injection pacing during exec + CI workflow distribution gates.
  • 2026-06-11: R19 Wasm query operators (host fallback hardening micro-slice): extended engine::tests::wasm_plan_compile_falls_back_for_unsupported_types to insert rows + call wasm_plan_inspect + wasm_plan_run (exercises the if artifact.execution == ... GENERATED else execute_select dispatch in wasm_plan_run for host_interpreted_v1 artifacts) and assert projected row results. This adds concrete coverage for compile/inspect/run/perf/edge-package surfaces on the host fallback path (in addition to generated_filter_project_v1). Still prototype: SIMD-lowered codegen and broader operator support (joins etc) remain gaps. Updated TRUE_STATUS_MATRIX + RESEARCH_BACKLOG; followed by fmt/clippy/test/site-rebuild/review/commit (ref review/matrix/R19). Per AGENTS: tiny, tests included, no on-disk/format change.
  • 2026-06-11: ABCD round complete (this entry): A (replay_state in stats + R18 interleave), B (lsm_pipeline_files_active extract), C (EVS in stats path), D (pg_stat_user_tables catalog). See new top Recent bullet + table rows. Per AGENTS tiny slices + tests + matrix/backlog/site + review + fmt/clippy/test + commit/push.
  • 2026-06-11: Full workspace verification + small storage hardening: added Engine::core_lsm_files_active() + EngineStorageStats.core_lsm_active (engine.rs) with test evidence in dedup_stats_roundtrip_across_restart. Confirmed core LSM files (MANIFEST + wal-*.log) for non-JSON modes. Matrix Storage core / Phase 1 gap description + evidence updated with code locations and this observability step toward primary pipeline.
  • 2026-06-11: A/B micro (Storage Pipeline + Monolith Split): added TableStorageMode::expects_core_lsm_files() helper in storage_mode.rs (B: decision logic centralized in extracted module per reviewer rec, avoiding dup in engine/server); refactored Engine::core_lsm_files_active + its test to use ManifestReader::open (and ManifestWriter for test setup) from core (A: more of the Manifest pipeline exercised in primary engine stats/open/restart path for gap tracking; read-only, no row persist/format change). Updated test engine::tests::core_lsm_files_active_validates_manifest_header_with_core_primitive + exercised via dedup_stats_roundtrip_across_restart. Interleaved R18 start (R18 replay uses storage_stats_snapshot for post-rehydrate LSM fidelity; this strengthens the observability used there). Per AGENTS: small net change, tests, fmt/clippy/test verified, matrix update. No ON_DISK change.
  • 2026-05-28: PostgreSQL CSV COPY now honors single-byte ESCAPE markers on supported CSV forms, allowing custom quote and escape characters to round-trip values containing both characters over COPY ... TO STDOUT and COPY ... FROM STDIN. Coverage: server::tests::pg_copy_csv_encode_and_parse_rows_handle_custom_escape_marker, server::tests::pg_parse_copy_to_stdout_recognizes_table_and_query_sources, server::tests::pg_parse_copy_from_stdin_and_decode_rows, cluster_rpc.rs::pg_simple_query_copy_to_stdout_with_csv_escape_roundtrip, cluster_rpc.rs::pg_simple_query_copy_from_stdin_with_csv_escape_roundtrip.
  • 2026-05-28: PostgreSQL catalog parity now includes pg_catalog.pg_description as an empty but correctly typed virtual table, so SELECT * FROM pg_catalog.pg_description returns PostgreSQL-shaped row descriptions and zero rows over both sql.exec and the live PG listener. Coverage: server::tests::sql_exec_pg_catalog_virtual_tables_roundtrip, cluster_rpc.rs::pg_simple_query_pg_catalog_virtual_tables_roundtrip.
  • 2026-05-28: PostgreSQL CSV COPY now honors single-byte QUOTE markers across export and import on supported CSV forms, including round-tripping quoted delimiters, empty strings, and doubled custom quote characters. Coverage: server::tests::pg_copy_csv_encode_and_parse_rows_handle_custom_quote_marker, server::tests::pg_parse_copy_to_stdout_recognizes_table_and_query_sources, server::tests::pg_parse_copy_from_stdin_and_decode_rows, cluster_rpc.rs::pg_simple_query_copy_to_stdout_with_csv_quote_roundtrip, cluster_rpc.rs::pg_simple_query_copy_from_stdin_with_csv_quote_roundtrip.
  • 2026-05-28: PostgreSQL catalog parity now includes pg_catalog.pg_am, exposing heap and btree access-method rows with stable OIDs aligned to the current pg_class.relam values over both sql.exec and the live PG listener. Coverage: server::tests::sql_exec_pg_catalog_virtual_tables_roundtrip, cluster_rpc.rs::pg_simple_query_pg_catalog_virtual_tables_roundtrip.
  • 2026-05-28: PostgreSQL text/csv COPY now honors custom NULL '...' markers across export and import, quotes literal "NULL" CSV cells so they round-trip distinctly from unquoted nulls, and enforces CSV HEADER MATCH on COPY ... FROM STDIN by rejecting mismatched header names before insert assembly. Coverage: server::tests::pg_copy_text_encode_and_parse_rows_handle_custom_null_marker, server::tests::pg_copy_csv_encode_and_parse_rows_handle_custom_null_marker, server::tests::pg_parse_copy_to_stdout_recognizes_table_and_query_sources, server::tests::pg_parse_copy_from_stdin_and_decode_rows, cluster_rpc.rs::pg_simple_query_copy_csv_with_custom_null_string_roundtrip, cluster_rpc.rs::pg_extended_query_copy_csv_with_custom_null_string_roundtrip, cluster_rpc.rs::pg_simple_query_copy_from_stdin_with_csv_header_match_rejects_mismatch.
  • 2026-05-28: PostgreSQL COPY now accepts PostgreSQL-style WITH (CSV|TEXT|BINARY) format aliases inside COPY option lists, and the live PG harness now explicitly covers extended-query CSV HEADER MATCH mismatch rejection. Coverage: server::tests::pg_parse_copy_to_stdout_recognizes_table_and_query_sources, server::tests::pg_parse_copy_from_stdin_and_decode_rows, cluster_rpc.rs::pg_simple_query_copy_to_stdout_with_csv_keyword_format_alias_roundtrip, cluster_rpc.rs::pg_extended_query_copy_from_stdin_with_csv_header_match_rejects_mismatch.
  • 2026-05-28: PostgreSQL COPY now also accepts legacy bare WITH CSV|TEXT|BINARY syntax, and the live PG harness covers keyword-alias copy-in roundtrips in addition to copy-out. Coverage: server::tests::pg_parse_copy_to_stdout_recognizes_table_and_query_sources, server::tests::pg_parse_copy_from_stdin_and_decode_rows, cluster_rpc.rs::pg_simple_query_copy_from_stdin_with_csv_keyword_format_alias_roundtrip, cluster_rpc.rs::pg_simple_query_copy_to_stdout_with_legacy_with_csv_header_roundtrip.
  • 2026-06-01: PostgreSQL binary COPY ... FROM STDIN now round-trips over both simple-query and extended-query flows, decoding the standard PostgreSQL binary stream (signature, flags, tuple framing, per-field values) and committing rows through the shared write path. Coverage: server::tests::pg_parse_copy_from_stdin_and_decode_rows, cluster_rpc.rs::pg_simple_query_copy_from_stdin_with_binary_format_roundtrip, cluster_rpc.rs::pg_extended_query_copy_from_stdin_with_binary_format_roundtrip.
  • 2026-05-27: PostgreSQL split_part(text, delimiter, n) now works through the shared evaluator with positive and negative field indexes, returns text metadata on the PG wire path, and is listed in pg_catalog.pg_proc. Coverage: engine::tests::eval_pg_split_part, server::tests::sql_exec_pg_catalog_virtual_tables_roundtrip, cluster_rpc.rs::pg_simple_query_split_part_roundtrip.
  • 2026-05-27: stats.snapshot.alerts now supports settings-backed route matching plus stats.snapshot-driven HTTP(S) webhook delivery for newly active matched alerts, exposing per-alert route delivery counters and top-level routing.delivery.{delivered,suppressed,failed,unsupported} metadata. Coverage: server::tests::stats_snapshot_routes_operator_alerts_from_settings, server::tests::stats_snapshot_delivers_http_alert_routes_once_per_active_alert.
  • 2026-05-27: encryption profile metadata and the redacted audit ring now persist across reopen in data/encryption.json, preserving mode / active_key_id status without persisting master keys. Coverage: engine::tests::encryption_status_persists_profiles_and_audit_across_restart.
  • 2026-05-29: encrypted-at-rest cell payloads are now wired through the engine row-persistence path. When a database has an active encryption profile (ENC_RANDOM/ENC_MLE_DB), encryptable cells (Str/Json/Bytes/Uuid/Embedding) are stored as "$skein_enc" envelopes in table-row files (format_version bumped 3→4), scalar/key cells stay plaintext, and ENC_MLE_DB stays deterministic for equal plaintext. Master keys are still never persisted: tables loaded without their key are marked locked (zero rows, persist refused so ciphertext is preserved) and transparently reload/unlock when the key is registered. Coverage: engine::tests::encrypted_at_rest_cells_roundtrip_and_lock_without_key, engine::tests::encrypted_at_rest_mle_is_deterministic_for_equal_plaintext. See docs/ON_DISK_FORMAT.md §11.8.1.
  • 2026-05-27: advisor.evaluate latency benchmarking now covers join-key samples, multi-range ordered samples, multi-column grouped workloads, and grouped range+order workloads. Coverage: engine::tests::advisor_evaluate_reports_latency_benchmark_for_join_key_workload, engine::tests::advisor_evaluate_reports_latency_benchmark_for_multi_range_order_workload, engine::tests::advisor_evaluate_reports_latency_benchmark_for_multi_group_workload, engine::tests::advisor_evaluate_reports_latency_benchmark_for_range_group_order_workload, cluster_rpc.rs::r16_index_advisor_evaluate_reports_join_key_latency_benchmark, cluster_rpc.rs::r16_index_advisor_evaluate_reports_multi_range_order_latency_benchmark, cluster_rpc.rs::r16_index_advisor_evaluate_reports_multi_group_latency_benchmark, cluster_rpc.rs::r16_index_advisor_evaluate_reports_range_group_order_latency_benchmark.
  • 2026-05-27: CDC plain_json format added and persisted in cdc_subscriptions.json format v8. Coverage: engine::tests::cdc_table_subscription_plain_json_format_persists_and_serializes, server::tests::cdc_table_subscription_plain_json_format_roundtrip, cluster_rpc.rs::cdc_table_subscription_plain_json_format_roundtrip.
  • 2026-05-27: CDC query dependency extraction now handles CTE definitions while ignoring CTE aliases as physical tables. Coverage: engine::tests::cdc_query_subscription_over_cte_query_invalidates_on_base_table_changes, cluster_rpc.rs::query_subscribe_over_cte_reports_base_table_keys_and_emits_sse_on_base_changes.
  • 2026-05-27: CDC query dependency extraction handles set-operation branches and view-expanded base tables. Coverage includes the cdc_query_subscription_over_union... and cdc_query_subscription_over_view... engine tests plus matching cluster_rpc.rs SSE tests.
  • 2026-05-27: CDC primary-key range filters are supported on single-column primary keys and persisted in subscription state.
  • 2026-05-26: CDC source-op, exact primary-key, and changed-column filters are live for table and query subscriptions.

Core Roadmap Status

Phase Current status Short truth
Phase 0 Repo setup Implemented Primitive file/record/value-id building blocks have runtime tests.
Phase 1 Storage core Partial Prototype persistence exists; full production storage pipeline is not complete. Monolith split progress: TableStorageMode extracted to storage_mode.rs + helpers + expects_core_lsm_files + lsm_pipeline_files_active (B); core_lsm_files_active + stats now use ManifestReader::open + replay_state (A, more core in stats path). This round: streaming stub + replay materialize path routed to core RowSegmentWriter+Manifest+Wal (A); more centralization in storage_mode (B). Evidence in Recent + Storage row (replay_bundle_export... + t183 + storage_mode_helpers... tests, "this round").
Phase 2 SQL + metadata Partial Catalog, DDL/DML subset, and compatibility metadata exist.
Phase 3 MySQL protocol Implemented baseline MySQL listener and broad SQL compatibility corpus pass; not full MySQL.
Phase 4 Web console Partial advanced SkeinAdmin is embedded with broad live panels; console remains evolving.
Phase 5 SkeinQL API Implemented baseline Typed RPC/API surface exists for core operations.
Phase 6 ETag cache coherence Implemented baseline ETags, prepared GET, and dependency notifications exist.
Phase 7 Delta values Implemented prototype Delta storage behavior is covered by ValueStore tests.
Phase 8 Wasm extensions Implemented baseline Wasm catalog/UDF sandbox exists; query operators are R19 prototype.
Phase 9 Audit WAL Implemented baseline Forensic hash chain, anchors, verify/status, and export tooling exist.
Phase 10 Row/column snapshots Partial Snapshot build/read/optimizer coverage exists for selected shapes.
Phase 11 Compat telemetry + migration Implemented baseline Telemetry and migration intent/rewrite/report surfaces exist.
Phase 12 SkeinAdmin Implemented baseline Admin UI covers major runtime surfaces and research panels.
Phase 13 Observability Partial advanced Stats, metrics, latency, CDC telemetry, basic alerts, settings-backed alert routing, and stats.snapshot-driven HTTP(S) webhook delivery exist; standalone escalation automation remains.
Phase 14 Cluster scale-out Implemented baseline Node identity, replication, shard movement, CAS transfer, and routing hints exist. HA is hardened: automated fenced failover (whole-cluster + per-shard) with quorum fencing, leadership epoch, and a Raft-style vote round; data-safe election on true (term, index) log positions; self-healing replication (op-log (term, seq), idempotent/gap-aware apply, leader-driven catch-up); and a commit index (majority-ack) propagated cluster-wide with per-node commit_lag. Remaining consensus follow-ons (read-committed replica reads, automated snapshot transfer for re-sync) are enhancements, not failover-safety gaps — see docs/CLUSTERING.md §2.5.
Phase 15 Perf improvements Implemented baseline Interning, late materialization, batch scan, and MVCC visibility cache exist.
Phase 16 Query coalescing Implemented baseline Coalescing is live for prepared GET and patch paths.
Phase 17 CAS-aware replication Implemented baseline Object need/missing/fetch/pull and shard-manifest transfer exist.
Phase 18 Index advisor Partial advanced Advisor synthesis/apply/retire/evaluate exists; non-grouped range+order layouts without the same leading key still fall back.
Phase 19 Time travel + replay Implemented baseline MVCC as_of, history, replay export/import/run, and admin tooling exist.
Phase 20 Encryption Partial baseline Crypto primitives, envelope helpers, rotation helpers, settings controls, and persisted mode/active-key/audit metadata exist. Engine row persistence now writes encrypted-at-rest cell envelopes (format_version 4, $skein_enc) for encryptable cell kinds when a DB profile is active, with deterministic ENC_MLE_DB and a locked-table guard when keys are absent. Master key bytes are still not persisted, scalar/key cells remain plaintext. Code: core/encrypted_valuestore.rs, core/encryption.rs, engine.rs encryption_* + key_manager. Core tests + engine integration tests. Micro C (2026-06-11 prior + this round): EncryptedValueStore use in skeindb crate test + extended to non-primary stats snapshot path (storage_stats_snapshot constructs it over ValueStore + key_manager for secondary/dedup routing baseline). storage_mode B split also.
Phase 21 Compaction scheduler Partial Live scheduler controls and worker exist; deeper storage compaction remains.
Phase 22 Autoparam + plan cache Implemented baseline Autoparam classifiers/feedback/metrics and plan-cache controls exist.
Phase 23 CDC/changefeeds Partial Local CDC is strong; external sinks/fanout and richer encodings remain.
Phase 25 PostgreSQL compat Partial advanced PG v3 baseline is substantial but not full PostgreSQL.
Phase 26 Distribution Partial Packaging scaffolding exists; signed publication requires release configuration.

Research Track Status

Track Status Primary runtime surface
R01 Learned indexes Hardened ValueStore learned-index reports, lookup traces, refresh policy, and benchmark probes.
R02 Adaptive row/column Hardened Snapshot optimizer, adaptive replacement, dependency refresh, and hybrid execution scaffolds.
R03 Delta topology Hardened Delta-chain policy, skip patches, compaction, and topology analysis.
R04 Differential privacy Hardened dp.* aggregates, budgets, audit, accuracy evaluation, and RDP composition.
R05 Oblivious execution Hardened oblivious.policy.*, padded execution, explain/evaluate, and privacy controls.
R06 Forensic WAL queries Hardened forensic.query, forensic.verify, forensic.export, Merkle proofs, and bundles.
R07 Client-side merge funcs Hardened merge.*, Wasm merge policies, offline queue spec, and conflict evaluation.
R08 Incremental views Hardened view.* create/refresh/evaluate/status/explain_deps with dependency tracking.
R09 QUIC-native protocol Hardened QUIC RPC transport, 0-RTT write rejection, migration/rebind, and transport benchmark.
R10 Vector embeddings Hardened vector.insert/search/benchmark/index.status, HNSW/LSH, cache metadata, and RAG sample.
R11 LLM/autoparam Hardened ai.autoparam.* classifier catalog, labels, feedback, and metrics.
R12 NL -> SkeinQL Hardened ai.nl.translate/explain/execute, approval tokens, and eval harness.
R13 Causal ETag consistency Hardened min_causality, vector clocks, causal validators, and replication watermarks.
R14 Replay bundles Hardened edge.bundle.* plus replay bundle redaction/export/import/run.
R15 Schema evolution Hardened schema.propose_change/merge_status/simulate_rollout/apply_merge.
R16 Auto index synthesis Hardened advisor.* synthesize/apply/retire/evaluate/history/dismiss.
R17 Intent inference Hardened migration.intent_report, migration.rewrite_preview, and migration.report_export.
R18 Perf regression replay Prototype implemented (micro progress) Performance profiles, variance reports, cache-hint rehydrate, timing injection primitive (inject_replay_timing + dedicated test), LSM stats fidelity re-compute in replay run exist; see recent verified + engine replay tests. CI distribution harness (compare) pre-exists.
R19 Wasm query operators Prototype implemented Wasm plan compile/inspect/run/perf/edge-package exists; production SIMD-lowered codegen and hardened operator breadth remain open. 2026-06-11 micro: host fallback run/inspect exercised in engine test (see Recent Verified + partial areas).
R20 Energy-aware compaction Hardened Energy-aware compaction policy, external energy signals, status/stats, and eval harness.

Compatibility Guardrail

SkeinDB coverage is measured against tests/compat/corpus.sql and live integration tests. If marketing copy says "100% MySQL compatibility" or "100% PostgreSQL compatibility," treat that as false until this file and the compatibility docs explicitly say otherwise.

Truth-Maintenance Rule

When a task is promoted from prototype to hardened behavior, update:

  1. The corresponding entry in docs/PROJECT_BACKLOG.md or docs/RESEARCH_BACKLOG.md.
  2. This matrix row, including any remaining gap.
  3. At least one test reference proving the claim.

Edit this page on GitHub  ·  View source