Research Overview Tracks Agenda
Docs › Developer / API › SkeinQL Research Extensions (Experimental)

SkeinQL Research Extensions (Experimental)

Last updated: 2026-05-14

This document lists experimental SkeinQL method families that support the 20-proposal research agenda.

Principles: - These methods are not required for MySQL compatibility. - They should be implemented behind feature flags and exposed only when enabled. - For each family, the canonical research description is in docs/research_agenda/. - Use docs/SKEINQL.md for method payload examples and docs/API_REFERENCE.md for the v0.3.18 runtime method map exposed through system.capabilities.

1. dp.* — Differential privacy

Related: R04

  • dp.aggregate — COUNT/SUM/AVG aggregates with explicit DP parameters, bounded sensitivities, and privacy_etag validators
  • dp.evaluate — seeded accuracy-vs-epsilon and overhead-vs-exact report for bounded DP aggregate plans
  • dp.budget.get / dp.budget.set — persisted per-principal budget tracking and policies
  • dp.audit.log — persisted budget consumption events

2. oblivious.* — Oblivious execution policies

Related: R05

  • oblivious.policy.get / oblivious.policy.set
  • oblivious.explain — show which operators will be padded/shuffled
  • oblivious.evaluate — deterministic trace-based leakage and overhead report for the active table policy

3. forensic.* — Forensic WAL queries with proofs

Related: R06

  • forensic.query — run table/op/id-bounded forensic queries with the SkeinForensic JSON filter grammar and return boundary, checkpoint-anchor, Merkle-root, inclusion-proof, and index-summary metadata
  • forensic.verify — verify contiguous returned record slices against a supplied start hash
  • forensic.export — export skein.forensic.bundle.v1 report bundles with the query manifest, records, proof, and verification summary

4. merge.* — Optimistic writes with merge semantics

Related: R07

  • merge.register — register a per-table merge policy with built-in or values-only Wasm functions
  • merge.apply — apply a policy to an incoming row with ETag/min-causality conflict hooks
  • merge.simulate — test current+incoming row merges without committing
  • merge.evaluate — report conflict rate, resolution success, and merge timing for example workloads
  • merge.wasm.register / merge.wasm.list / merge.wasm.drop — manage values-only Wasm merge modules

5. view.* — Materialized views + incremental maintenance

Related: R08

  • view.create / view.drop
  • view.refresh — incremental, full, or auto refresh for restricted single-table views (including grouped views)
  • view.evaluate — read-only incremental-vs-full correctness oracle and timing report
  • view.status — lag/last refresh
  • view.explain_deps — dependency graph edges

6. transport.* — Protocol negotiation hints

Related: R09

  • transport.capabilities — advertise supported transports (HTTP/1.1, HTTP/2, QUIC); the QUIC runtime has framing, prepared-query, 0-RTT write rejection, rebind coverage, and the companion skeindb transport-bench CLI for comparative p99 checks against HTTP/2 and MySQL/TCP

Related: R10

  • vector.insert — store embedding values (ValueKind::Embedding)
  • vector.search — ANN query (HNSW when available, with LSH bucket filtering as the prototype fallback) with cache validators, table-version dependency metadata, V2 causality tokens, and source-change invalidation
  • vector.benchmark — exact brute-force vs indexed top-k recall and latency report for embedding columns
  • vector.index.status — index health and coverage

See Vector RAG retrieval for a credential-free sample application that uses vector.insert and vector.search to assemble grounded context.

8. ai.* — AI-assisted query workflows

Related: R11, R12

  • ai.autoparam.classify — classify literals (semantic-constant vs parameterizable)
  • ai.autoparam.analyze — extract literals from SQL and classify them
  • ai.nl.translate — prompt packaging + optional rule-based translation (read-only by default)
  • ai.nl.explain — explanation + preview rows for verification
  • ai.nl.execute — execution gated by approval token
  • skeindb nl-eval — JSONL evaluation harness with exact and execution-match metrics

9. causal.* — Causal consistency via ETag chains

Related: R13

  • query.select / query.execute_prepared accept cache.min_causality
  • query results include causality tokens (vector_clock_v2; legacy etag_chain_v1 remains accepted on input)
  • causal.session.begin / causal.session.end (future)

10. edge.* — Replay bundles as edge replication primitive

Related: R14

  • maintenance.replay.export — export snapshot replay bundles with optional primary-key redaction (none, hash_pk, drop_pk) before checksums/performance metadata are computed
  • edge.bundle.request — request bounded WAL slice / replay bundle
  • windows: table + seq bounds + max events
  • redaction: none | hash_pk | drop_pk
  • edge.bundle.apply — apply bundle coverage windows to the edge node
  • edge.bundle.status — coverage + bounded-staleness routing verdict, including gap detection for disjoint windows

11. schema.* (extensions) — Conflict-free schema evolution

Related: R15

  • schema.propose_change — propose schema evolution changeset
  • schema.merge_status — show divergence, merge plan, and proposed resolution guidance
  • schema.simulate_rollout — simulate prepare/mixed/steady-state rollout waves for the current merge plan
  • schema.apply_merge — apply merged schema
  • Prototype change ops: add_column (nullable/default support) and add_index (name/column list/unique flag)

12. advisor.* (extensions) — Dependency-driven index synthesis

Related: R16

  • advisor.index_synthesize — propose indexes based on dependency analysis
  • advisor.apply_index — apply an index suggestion (in-memory secondary index)
  • advisor.dismiss — suppress a suggestion
  • advisor.history — list advisor actions

13. migration.* (extensions) — Intent inference

Related: R17

  • migration.intent_report — detect idioms (pagination, polling, hierarchy self-joins, recursive CTEs, EXISTS membership, COALESCE defaults)
  • migration.rewrite_preview — show suggested SkeinQL migration

Example (intent_report):

{
  "samples": [
    { "query": { "...": "..." }, "args": [] }
  ],
  "limit": 20,
  "window_ms": 60000
}

Example (rewrite_preview):

{
  "samples": [
    { "query": { "...": "..." }, "args": [] }
  ],
  "limit": 10
}

14. replay.* (extensions) — Performance replay

Related: R18

  • maintenance.replay.export — exports data bundles with optional performance profile metadata (lsm_state, cache_warm, and timing).
  • maintenance.replay.import — validates both correctness checksums and performance-profile checksums when present.
  • maintenance.replay.run — returns performance_report variance deltas for performance-annotated bundles, with replay-run checksum parity over reconstructable snapshot state.

15. wasm.* (extensions) — Wasm-native operators

Related: R19

  • wasm.plan.compile — compile plan subset to Wasm (see docs/WASM_OPERATORS.md)
  • wasm.plan.run — run compiled plan (sandbox)

16. energy.* — Energy-aware scheduling

Related: R20

  • energy.policy.set — compaction energy policies
  • energy.status — current energy signals and scheduling decisions

Important: method shapes are intentionally left high-level in this document. Use the proposal docs to define exact parameter schemas.

Edit this page on GitHub  ·  View source