CHANGELOG.md

# Changelog

## Unreleased

- Add bounded live parameterized lab replay and recorded-case rewrite verification;
  preserve duplicate/type-sensitive result comparisons without exporting raw values.
- Collect live context fingerprints and check recommendation validity from current
  metadata; qualify real index drift detection on both supported engines.
- Correlate explicit cross-layer process spans, expose ambiguous links, collect
  configured HTTPS trace exports and compare scoped business process outcomes.
- Add fail-closed central approval-status and external intent-hash witness adapters.
  Live customer identity, full workload coverage and production writes stay unqualified.

- Add persistent diagnostic sessions with scoped test outcomes, counterhypothesis
  review, model revision, optimistic concurrency and verification transitions.
- Add transparent business-process prioritization and a joint repeated-performance
  and business-correctness gate. Supplied evidence is never labeled attested.
- Add recommendation validity hashes, expiration and explicit invalidation.
- Add trusted Ed25519 issuer approvals bound to exact operation and connection
  scope, separate allowlisted reviewers and atomic single-use nonce consumption.
  Production apply stays blocked; live IdP integration is not claimed.

- Validate migration signatures before connecting or executing; bind SQL, scope,
  profile, actor and expiry. Respect unsigned dry-run policy without authorizing apply.
- Limit live migration application to scoped signed lab index changes and exact
  index rollback. Fix SQL Server index syntax and prevent mock/error applied claims.
- Add installed CLI migration workflow qualification with disposable databases,
  rejected signatures, real index/row checks, rollback and cleanup on both engines.

- Add opt-in installed SQL Server/PostgreSQL temporary write rehearsals: CRUD,
  key-conflict recovery, savepoints, full rollback and cleanup verification.
  These qualify synthetic session mechanics, not productive migration authorization.

- Add live-only admin_preflight with explicit database scope, independent collector
  checks, bounded error reporting and no write or production-readiness approval.
- Correct live SQL Server/PostgreSQL statistics: preserve measured means, distinguish
  elapsed time from CPU, report unavailable p95/regression as null, scope queries to
  the selected database, and stop relabeling means as percentiles in workload signals.
- Require complete validated backup/restore evidence, explicit RPO/RTO policy and
  successful restore assertion; missing evidence no longer qualifies as ready.
- Extend installed live qualification with admin preflight for both database engines.

- Fix ambiguous SQL Server column metadata query found during live adapter testing;
  qualify c.name and cover it in regression and installed live qualification checks.

- Add next_evidence_test with worst-case hypothesis discrimination, explicit outcome
  branches, prerequisite/scope checks and bounded supplied cost; no execution or
  probability claims. Integrate test selection into the guided advisor workflow.

- Make agreed business controls mandatory for confirmed advisory-case closure at
  verification and follow-up; bind definition, expected groups and benchmark snapshot.
- Add advisor_workflow as the main guided entry and causal_evidence_review with
  cross-layer trace/time correlation, explicit contradictions and no causality claim.
- Retrieve confirmed history only within exact version/workload/customization scope;
  exclude legacy cases without business verification. Preserve existing history.
- Test case persistence through the installed runtime and document a prospective
  customer-pilot acceptance protocol without fabricating external qualification.

- Add exact-decimal business control reconciliation per company, currency, unit
  and metric, with row-count, missing-group and snapshot/definition scope checks.

- Attribute the plugin to Aivana GmbH and the user-supplied company website.
- Add advisory_case with scoped SQLite persistence, revision checks, proposal-bound
  test approvals, repeated verification, later follow-up and reviewed quality counts.
- Add repeated_benchmark_review and release_qualification_matrix; missing external
  qualification stays not_run and supplied receipts never authorize publication.
- Add business-case deadlines and per-case regression checks with policy fingerprints.
- Remove fixed counterfactual timing and keyword incident probability predictions;
  expose legacy confidence as uncalibrated heuristic scores. Require explicit ROI
  unit costs instead of hard-coded monetary assumptions.
- Extend live qualification with warm-up, alternating benchmark repetitions and
  explicitly opted-in Dataverse/Salesforce collection checks.

- Move audit and advisor memory to SQLite transactions with checked legacy import,
  ordered replay verification and process-crash recovery; require Node 22.16+.
- Reserve one SQL Server session for SHOWPLAN and cleanup; add optional Windows
  Integrated Authentication through msnodesqlv8 and an installed Microsoft ODBC driver.
- Import Azure Monitor tables, normalized diagnostic events, Dataverse plug-in
  traces and Salesforce plans; add two scoped read-only API collectors.
- Compare benchmark samples only when workload, data, parameters, concurrency and
  result proofs match; reject correctness failures even when timings improve.
- Qualify extracted packages and isolated Codex install/update, with optional live
  SQL Server cleanup checks and a PostgreSQL measured benchmark fixture.

- Add ERP/CRM catalog, evidence-plan and risk-analysis tools covering 13 explicit
  vendor profiles and 43 checks, with numeric diagnostics, scope/freshness checks,
  conflict handling and no automatic vendor-system modifications.

- Derive PostgreSQL live EXPLAIN cost and findings from the returned plan.
- Serialize advisor-memory updates across processes and surface persistence errors.
- Build a checked standalone candidate archive with lockfile and SHA-256.

- Preserve live-connection enforcement when a secret provider is configured.
- Isolate test worker state and support external state with CODEXDB_STATE_DIR.
- Move CI to the repository root and add Windows/Linux and Node 22/24 coverage.
- Validate the distributable package, including required files and secret exclusions.
- Repair ten skill frontmatter headers rejected by the Codex plugin validator.
- Accept CLI JSON through stdin and reject non-object arguments.
- Correct installation, environment loading, and AI evidence documentation.

## 1.0.0

- Added production no-mock gates for live SQL advisory workflows.
- Added AI SQL performance advisor orchestration with diagnosis, ranking, benchmarking, workload impact, and consultant response tools.
- Added deep execution-plan diagnostics for cardinality, scan, statistics, and spill risks.
- Added advisor USP layer for index ROI simulation, plan regression watch, parameter sensitivity, maintenance windows, and SLO impact.
- Added advanced advisor USPs for query fingerprint clustering, wait-event root cause, data skew, cache efficiency, and capacity headroom.
- Added end-to-end consultant USPs for evidence packs, tuning experiments, migration prediction, workload replay, index portfolio optimization, incident timelines, confidence grading, change tickets, cost-performance advice, and schema evolution guardrails.
- Added maximum competitive AI layer for autonomous DBA copilot, performance PR review, cross-tool DevOps analysis, guarded SQL generation, synthetic workload lab, vendor-neutral DevOps brain, compliance query assistant, migration twin, policy-gated self-healing, and advisor memory recommendations.
- Added extra AI USP layer for SQL code review, schema diff intelligence, query contracts, rollback rehearsal, release train risk, observability routing, fleet health, index hypotheses, developer coaching, and environment drift.
- Added operating-excellence USPs for plan narration, SLO policy compilation, index retirement, hotfix risk, noisy-neighbor detection, partition strategy, statistics health, connection pools, backup/restore readiness, and vendor feature mapping.
- Added redacted connector request preparation for Prometheus, Grafana, Neo4j, pgvector, PostgreSQL, and SQL Server.
- Hardened missing production USP paths: configured pgvector/Neo4j grounding, pgvector upsert contracts, OpenTelemetry metric normalization, plan-metric cost evidence, and tenant-required RLS blocking.
- Added disruptive closed-loop autonomous operator USP layer for objective decomposition, dry-run experiments, counterfactual risk, confidence budgets, autonomy boundaries, safe next actions, and executive operations briefings.
- Added AI cognitive control-plane USP layer for strategy synthesis, cognitive schema mapping, LLM prompt risk audits, AI decision simulation, autonomous learning backlogs, knowledge-gap detection, trust scorecards, semantic incident prediction, cross-agent consensus, and executive ROI narratives.
- Added release contracts and marketplace demo scenarios for the disruptive autonomous and AI USP layers.
- Added curated killer demos for unsafe prompt blocking, board-level operator briefings, cross-agent disagreement, and CIO/CTO ROI narratives.
- Added launch-ready marketing kit and sales playbook with positioning, ICP, personas, USP matrix, competitive framing, listing copy, demo flow, objection handling, and pilot criteria.
- Added root marketplace manifest and compacted marketplace skill files to satisfy scanner requirements for `.codex-plugin/plugin.json` and the 128-file artifact limit.
- Added enterprise proof USP layer for live evidence, formal contracts, governance proof, benchmark ROI, pilot success, visual executive reporting, enterprise security proof, and competitive battlecards.
- Added readiness, smoke, audit, and CI entry points for production release checks.