CHANGELOG

v1.2.0-dev - 2026-02-25

PHASE_55 — KPI Cube

KPI-1/2: KpiCubeBuilder + 10 canonical patterns:

• lib/ai/kpi-cube.ts (new) — buildCubeForDataset(datasetId, schema) builds Redis-backed cube at upload time; parallel Promise.all over 10 patterns with 7-day TTL (604800s)
• CUBE_PATTERNS array: metric_by_day, metric_by_week, metric_by_month, top_n_by_metric, period_over_period, distribution_by_dim, latest_value, min_max_avg, cumulative_metric, rolling_7d_avg
• Column detection helpers: detectMetricColumns (numeric non-id), detectDateColumn (type=date or name heuristic)
• Coverage manifest stored at kpi:cube:coverage:{datasetId}; TTL patterns skip gracefully on null SQL

KPI-1 (integration): Dataset upload triggers cube build:

• app/api/datasets/route.ts — fire-and-forget buildCubeForDataset(id, schemaJson).catch(warn) after DB insert

KPI-3: KpiCubeLookup in orchestrator:

• lib/ai/orchestrator.ts — lookupCube(datasetId, intent) called before PlannerAgent; cache hit bypasses all LLM calls
• Hit/miss counters: kpi:cube:hits:{datasetId} / kpi:cube:misses:{datasetId} incremented fire-and-forget

KPI-4: Delta-aware invalidation:

• lib/ai/kpi-cube.ts — invalidateCubePartitions(datasetId, changedDateRange?): partial range deletes time-partitioned patterns; full invalidation deletes all 10 + coverage

KPI-5: Ops hit-rate metrics:

• app/api/ops/metrics/route.ts — kpi_cube: { hit_rate, total_hits, total_misses } via Redis key scan; null when no data or Redis unavailable

Tests:

• tests/unit/lib/ai/kpi-cube.test.ts (new) — 24 tests: column detection, all 10 buildSQL outputs, matchFn intent routing, buildCubeForDataset Redis calls + TTL, lookupCube hit/miss, invalidateCubePartitions full+partial

PHASE_50 — Elite Conversion Polish

P5: Welcome Wizard:

• components/onboarding/WelcomeWizard.tsx (new) — 3-step onboarding modal: Welcome (value prop + checklist), Upload (drag-drop zone + 3 sample presets), Ask (4 example questions)
• localStorage key insights-ai:welcome-wizard-complete prevents re-showing after first completion
• AnimatePresence backdrop + modal; per-step slide transitions; progress dots navigation; skip on every step
• useReducedMotion compliance: static opacity transitions when reduced motion preferred

P7: First-insight celebration:

• components/chat/ChatInterface.tsx — toast.success("Your first insight! Keep exploring your data.") fires once per user on first completed query result
• localStorage flag insights-ai:first-insight-done ensures single-fire; reduced-motion: toast only (no sparkle)

P13: Error/404 page redesign:

• app/error.tsx — cancelable countdown: Cancel button alongside 5s auto-retry timer clears setInterval; human-readable error.message surface above digest; dev-only <details> stack trace (process.env.NODE_ENV === "development"); animated entrance with useReducedMotion
• app/not-found.tsx (converted to "use client") — animated entrance with useReducedMotion; wandering-path SVG (bezier curves + endpoint dot, no compass); new copy "This page went exploring and got lost"; inline search bar routes to /dashboard?q=...

P14: Scroll-to-bottom FAB:

• components/chat/ChatInterface.tsx — ArrowDown floating action button, AnimatePresence fade in/out; appears when distFromBottom > 200px; scrolls to bottom on click

P17: Layout meta tags:

• app/layout.tsx — canonical URL <link rel="canonical" href={NEXT_PUBLIC_APP_URL} />; explicit <meta property="og:image" content=".../api/og" />; <link rel="apple-touch-icon" href="/icon.svg" sizes="any" />; <meta name="twitter:card" content="summary_large_image" />

P19: Pricing toggle wired to real prices:

• app/pricing/page.tsx (converted to "use client") — MONTHLY_PRICES record; TierPrice component using useAnimatedPrice from PricingToggle.tsx; annual state wired to PricingToggle onChange; default annual selected

P1-P4, P6, P8-P12, P15-P16, P18: Confirmed already implemented (logo cloud, usage counter, TestimonialCarousel, scroll progress, sample datasets, error taxonomy, agent labels, typing indicator rotation, command palette + shortcut overlay, focus rings, placeholder styling, PWA manifest)

Tests:

• tests/unit/components/premium-ui-batch3.test.ts — updated assertion to match new not-found.tsx copy ("went exploring and got lost")

PHASE_51 — Agent Correctness

AC-1: Claim-evidence linking:

• lib/ai/schemas.ts — added evidence_query_id: z.string().optional() to InsightResultSchema
• lib/types/index.ts — added evidence_query_id?: string to InsightResult interface
• lib/ai/agents/insight-synthesis.ts — deriveQueryId(sql) computes first 16 hex chars of SHA-256(sql); injected post-parse

AC-2: Evidence coverage enforcement:

• lib/ai/verification.ts — EvidenceCoverageReport interface + checkEvidenceCoverage(insights, expectedQueryId?) function
• Returns { passed: boolean, violations: string[] }; violation on missing or mismatched evidence_query_id

AC-3: Formal orchestrator fallback modes:

• lib/ai/orchestrator.ts — OrchestratorMode const (FULL=0 | REDUCED=1 | SINGLE=2 | DETERMINISTIC_ONLY=3)
• OrchestratorModeContext interface + selectOrchestratorMode() pure function
• Mode 3 upfront check in runQueryPipeline: when both circuit breakers OPEN, falls back to deterministic or returns 503

AC-4: Statistical significance gate:

• lib/utils/statistics.ts (new) — zScore, isStatisticallySignificant (|z|≥1.96, n≥30), hasStrongClaimLanguage, hedgeStatisticalLanguage (single-pass combined regex — no chain substitution)
• lib/ai/agents/insight-synthesis.ts — applyStatisticsGate hedges summary+insights when rowCount < 30

AC-5: Ops metrics stubs:

• app/api/ops/metrics/route.ts — added claims_without_evidence_count and fallback_mode_distribution to response

AC-6: Tests (54 new, all passing):

• tests/unit/lib/utils/statistics.test.ts (22 tests) — zScore, significance, hasStrongClaimLanguage, hedgeStatisticalLanguage
• tests/unit/lib/ai/agents/insight-synthesis.test.ts (+4 tests) — evidence_query_id shape, determinism, statistical hedging gate
• tests/unit/lib/ai/orchestrator.test.ts (+7 tests) — selectOrchestratorMode all 4 modes + Mode 3 integration

AC-7: Per-agent token cap:

• lib/ai/config.ts — getInsightModel capped at min(1024, MAX_LLM_OUTPUT_TOKENS) (was 2048)
• lib/ai/orchestrator.ts — AGENT_MAX_OUTPUT_TOKENS const documents per-agent caps

PHASE_52 — DAG Planner

DAG-1: TaskGraph types + Zod schema:

• lib/ai/task-graph.ts — AgentTaskType, AgentLifecycle, AgentTask, TaskGraph, TaskTrace types
• Zod schema validates: no circular deps (DFS), all depends_on refs exist

DAG-2: PlannerAgent (Haiku):

• lib/ai/agents/planner-agent.ts — calls claude-haiku-4-5-20251001, emits TaskGraph JSON, falls back to default graph on invalid output
• lib/ai/prompts/planner-agent.ts — planning prompt with minimal graph instructions

DAG-3: Budget allocator:

• allocateBudgets() — 10% planner / 75% execution / 15% verifier; exec tasks split proportionally by complexity_weight
• buildDefaultGraph() — standard 4-task fallback: DATA_PROFILE → QUERY_GEN → INSIGHTS → VERIFY
• MAX_TOTAL_TOKENS_PER_REQUEST env var (default: 60,000)

DAG-4: Orchestrator refactor:

• lib/ai/orchestrator.ts — runQueryPipeline now calls PlannerAgent before the LangGraph pipeline; taskGraph added to QueryPipelineState
• Topological ordering preserved via existing LangGraph node sequence

DAG-5: Per-task trace logging:

• emitTaskTrace() — writes to usage_logs table (reuses existing schema); falls back to structured log on DB error
• topoSort() utility exported from task-graph.ts
• Trace emitted for each completed pipeline node in stream loop

DAG-6: Tests:

• tests/unit/lib/ai/task-graph.test.ts — DAG validation, cycle detection, budget allocation, topoSort, emitTaskTrace
• tests/unit/lib/ai/agents/planner-agent.test.ts — valid parse, fallback on no JSON, fallback on cycle, fallback on API error
• tests/unit/lib/ai/orchestrator.test.ts — added planner + task-graph mocks

PHASE_49 — Elite Interactions

M1: Motion presets module — Added slideInLeft, slideInRight, slideInDown, blurIn, scaleRotateIn variants to lib/utils/motion-presets.ts

M2: Page transitions — components/layout/PageTransition.tsx scroll-to-top on route change

M3: Chart animation choreography — animationBegin=200ms delay in ScatterPlot, AreaChart; BigNumber adaptive duration 300 + digits×180ms

M4: Animated AI avatar — Pulsing ring during streaming (avatarPulse CSS), agent icon differentiation (Search/CheckCircle/Zap/Sparkles) based on pipeline stage

M5: Message action toolbar — Copy + Save query buttons in hover toolbar in ChatMessage

M6: Enhanced confidence ring — Already implemented in prior work (ConfidenceBadge SVG ring)

M7: Optimistic UI — Already implemented in chat-store (user message + placeholder appear instantly)

M8: Content-aware skeletons — Skeleton component variant="shimmer"|"pulse", role="status" on loading.tsx

M9: will-change audit — willChange: "opacity" on streaming cursor + avatar pulse; Framer Motion handles transforms automatically

M10: ARIA live regions — role="progressbar" + aria-valuenow/min/max added to AgentProgress; ChatInterface already had role="log" + aria-live="polite"

M11: Rich chart tooltip — glass-md styling, tooltipIn CSS enter animation, smart format detection (currency/percentage/int/float)

M12: Code block enhancements — Line number gutter, SQL syntax highlighting (keyword/string/comment/number tokens via CSS classes), diff mode (+/- line coloring)

M13: AnimatedSection direction system — direction prop (up/down/left/right/scale/blur), staggerChildren prop using stagger factory, springs.gentle transition, delay clamped to 0.8s

PHASE_48 — Elite Design Tokens

T7: Wordmark gradient variant — gradient variant added to components/ui/Wordmark.tsx; aria-label on icon-only; hover animation via group

T9: Input token migration — duration-normal added to input.tsx transition

T10: Opacity scale — --opacity-subtle/ghost/muted/dim/strong CSS vars + Tailwind opacity extend

T11: Shimmer + glass utilities — shimmer keyframe + animation in Tailwind; .glass-sm/md/lg CSS utilities; streamCursor + avatarPulse keyframes

v1.1.0-dev - 2026-02-25

PHASE_47 — Code Quality

ENG-7: Standardized error responses:

• lib/api/error-response.ts — new errorResponse(), unauthorizedError(), forbiddenError(), notFoundError(), validationError(), internalError() and more helpers
• Applied to 10 routes: alerts, notifications, saved-queries, datasets, conversations, organizations
• Fixed pre-existing type errors: app/pricing/page.tsx, lib/payments/credits.ts (Stripe v20 pricing.price), components/landing/AnimatedSection.tsx

ENG-13: Split ChatMessage.tsx (1,248 → 564 lines):

• Extracted MarkdownCodeBlock → components/chat/MarkdownCode.tsx
• Extracted ReactionBar → components/chat/ReactionBar.tsx
• Extracted CommentsSection → components/chat/CommentsSection.tsx
• Extracted PinToDashboard → components/chat/PinToDashboard.tsx

ENG-16: 3 new DB indexes:

• idx_activity_feed_user on activityFeed.userId
• idx_scheduled_reports_next_run on scheduledReports.nextRunAt
• idx_msg_comments_user on messageComments.userId

ENG-18: Removed unused wget from Dockerfile runner stage (reduces attack surface ~500KB)

ENG-20: Added OPS_OWNER_IDS and BYPASS_AUTH to .env.example with documentation

ENG-21..26: Critical bug fixes from audit:

• JSON.parse wrapped in try-catch in 6 agent files (builder, planner, insight-synthesis, query-generation, intent-understanding, schema-analysis)
• Dummy API key fallback → proper 503 response with LLM_NOT_CONFIGURED code
• Saved-queries: DB try-catch + .safeParse() instead of .parse()
• v2/conversations: consistent v2Error() + runtime type validation
• DomainId: validated against allowlist before casting in packs route
• WORD_BOUNDARY_CACHE: capped at 1000 entries with FIFO eviction

PHASE_36 — Intelligence Credit System & Pricing Migration

Credit Cost Model (P1):

• lib/ai/config.ts — CREDIT_COSTS map, intentToCreditTier(), creditCostForIntent(); Simple: 2cr, Standard: 5cr, Complex: 15cr, Dashboard: 25cr, Forecast: 40cr

Credit DB Tables (P2):

• lib/db/schema.ts — userCredits and creditLedger (append-only audit) tables
• drizzle/0010_credits.sql + drizzle/0010_down.sql

Credit Deduction Logic (P3):

• lib/payments/credits.ts — getCreditBalance(), grantMonthlyCredits(), deductCredits() (monthly-first drain), applyTopUpCredits(), checkCreditBalance(), triggerAutoTopUpIfNeeded()
• app/api/chat/route.ts — credit pre-flight after intent; fire-and-forget deduction after success

Credit Top-Up Purchase (P4 + P5):

• app/api/credits/topup/route.ts — Stripe Checkout mode: "payment" for 250/1000/5000cr packs
• app/api/webhooks/stripe/route.ts — one-time top-up + monthly credit grants on subscription events

Pricing Page (P6):

• app/pricing/page.tsx — 6-tier ladder (Free/Starter/Pro/Growth/Enterprise/API) + top-up packs section

Usage Banner (P7):

• components/layout/UpgradeUsageBanner.tsx — dual-path credit/query-count display

Stripe Price IDs (P8):

• lib/payments/stripe.ts — PlanType, CREDIT_MONTHLY_ALLOTMENT, STARTER_PRICE_ID (backwards-compat alias), NEW_PRO_PRICE_ID, GROWTH_PRICE_ID, TOPUP_PRICE_IDS

Migration Script (P9):

• scripts/migrate-pro-to-starter.mjs — batch Clerk metadata update with --dry-run + --limit=N flags

Domain Pack Add-On Checkout (P10):

• app/api/packs/[packId]/subscribe/route.ts — Stripe subscription checkout for all 7 domain packs

Plan Types + Feature Flag + Tests (P11):

• lib/feature-flags.ts — enableCreditSystem flag (default: false, gate: 1,000+ paying users)
• app/api/checkout/route.ts — expanded plan enum; default changed "pro" → "starter"
• tests/unit/lib/payments/credits.test.ts — 15 unit tests for credit tier mapping + ALLOTMENT values

Result: 1254 passing (11 test regressions fixed: checkout default plan, chat/cache credits mock)

PHASE_37 — Technical Debt

ISR / Cache Revalidation:

• export const revalidate = 300 added to 4 dashboard RSC page shells (dashboard, history, alerts, pinned)
• revalidatePath('/dashboard') called on dataset create/update/delete
• revalidatePath('/dashboard/history') called on conversation create
• Architecture documented in ARCH_ISR_CACHE.md

Event Bus:

• lib/events/types.ts — typed AppEvent + AppEventType union (7 event types)
• lib/events/bus.ts — Redis list queue (LPUSH/RPOP), publish() / drainBatch() / on() / dispatch()
• /api/cron/process-events — consumer cron; routes dataset.created/deleted + conversation.created → activity feed
• Events wired from dataset POST/DELETE and conversation POST routes

Billing Idempotency:

• lib/billing/idempotency.ts — Redis SETNX-based guard, 60s TTL, billingIdempotency(userId, action, discriminator)
• /api/checkout — idempotency guard added; all 3 plan branches commit on success
• /api/billing/cancel — idempotency guard added for save_offer / cancel / reactivate actions

DB Tracked Migrations:

• npm run db:migrate now runs drizzle-kit migrate (tracked, safe for staging/prod)
• npm run db:push added as local-dev alias for drizzle-kit push:pg (dangerous — local only)
• ARCH_DB_MIGRATIONS.md — documents workflow, rollback strategy, and CI integration
• All 10 existing migrations (0000–0009) already have _down.sql rollback scripts

Background Jobs:

• lib/jobs/runner.ts — registerJob() / runJob() / getLastJobResult() / listJobs(); timeout enforcement + Redis result store (24h TTL)
• ARCH_BACKGROUND_JOBS.md — documents cron schedule, job pattern, security, failure handling

Result: 1243 passing (7 test regressions fixed)

v1.1.0-dev - 2026-02-24

PHASE_35 — CI/CD Performance

CI pipeline restructured for parallel execution (~25min → target <12min on PRs):

• O1 Bundle analyzer (npm run analyze) gated to main pushes only — saves ~10–15 min on every PR check
• O2 Parallel jobs: secret-scan, lint, type-check, test all run concurrently; build starts only after all pass
• O3 public/duckdb/ cached in GitHub Actions by package-lock.json hash; DuckDB WASM copy skipped on warm cache; npm ci --ignore-scripts used in all jobs
• S1 Removed redundant prebuild script from package.json — postinstall already handles DuckDB WASM copy
• O4 timeout-minutes: 3 on Run load smoke test step — prevents infinite hang if server fails to start
• O5 TruffleHog secret scan extracted to standalone job — fails immediately without waiting for lint/test/build
• O6 Vitest uses pool: 'threads' with maxThreads: 2 in CI; auto-detects locally

Test fixes (pre-existing failures surfaced by O6):

• schema-analysis.test.ts — updated to match new runSchemaAnalysis(intent, userQuery, schema) signature
• rate-limit.test.ts — Redis mock updated from incr/expire/ttl to eval (Phase 33 Lua script migration)
• alerts-route.test.ts — replaced stub file with proper placeholder test suite
• conversations-routes.test.ts, datasets.test.ts, notifications-route.test.ts, notification-center.test.ts — added .offset() to DB mock chains; added Request arg to GET() calls

Result: 1129 passing, 0 failing (was 1098 + hidden failures)

v1.0.0-dev - 2026-02-18

Phase 17 launch prep completion batch.

• Closed remaining Phase 17 launch tasks:
  • A3: full API route integration matrix coverage complete across app/api/**/route.ts families
  • A4: launch-critical E2E matrix rebuilt in tests/e2e/launch-readiness-flows.spec.ts with deterministic mocked flows:
    • full chat flow (query, follow-up, export, delete)
    • conversation management (create, rename, search, navigate, delete)
    • dataset lifecycle (upload path, rename, delete)
    • settings/API key and billing portal flows
    • mobile dashboard navigation/chat coverage
  • A5: coverage gate moved to 80% for launch-critical API scope in vitest.config.ts
• Added public status page at /status (app/status/page.tsx) and linked it from landing/pricing/footer/sidebar/settings; sitemap updated.
• Added pricing validation reference (docs/40_reference/30_business/BIZ_PRICING_VALIDATION.md) with margin scenarios and guardrail decisions.
• Added load-test assets:
  • k6 script: scripts/load-tests/chat-load-test.js
  • npm command: npm run test:load:chat
  • baseline runbook: docs/40_reference/20_engineering/LOAD_TEST_BASELINE.md
• Added conversation rename support (API + UI):
  • PATCH /api/conversations/[id] now supports title updates
  • history UI inline rename action in components/chat/ChatHistory.tsx
• Pricing copy aligned with enforced free-tier query limit (50/month) on landing and pricing pages.

v0.2.8-dev - 2026-02-18

Phase 17 launch prep testing expansion batch.

• Added integration coverage for connection endpoint matrix (tests/integration/api/connections-routes.test.ts):
  • /api/connections
  • /api/connections/[id]
  • /api/connections/test
  • /api/connections/[id]/test
  • /api/connections/[id]/schema
  • /api/connections/[id]/query
  • Includes auth, validation, not-found, success, and server-error assertions
• Added integration coverage for conversation endpoint matrix (tests/integration/api/conversations-routes.test.ts):
  • /api/conversations
  • /api/conversations/[id]
  • /api/conversations/[id]/messages
  • Includes auth, validation, ownership/not-found, and success assertions
• Added integration coverage for utility endpoints (tests/integration/api/utility-routes.test.ts):
  • /api/usage
  • /api/transform
  • /api/health
  • /api/ready
• Added integration coverage for organization endpoints (tests/integration/api/organizations-routes.test.ts):
  • /api/organizations
  • /api/organizations/[id]
  • Includes auth, validation, conflict, role/owner checks, and success assertions
• Added integration coverage for API key management (tests/integration/api/api-keys-routes.test.ts):
  • /api/settings/api-keys
  • Includes auth, permission, validation, not-found, and success assertions
• Added integration coverage for organization collaboration subroutes (tests/integration/api/organization-collaboration-routes.test.ts):
  • /api/organizations/[id]/members
  • /api/organizations/[id]/members/[userId]
  • /api/organizations/[id]/invitations/[invitationId]
  • /api/organizations/[id]/invitations/[invitationId]/resend
  • /api/organizations/[id]/datasets
• Added integration coverage for invitation token flows (tests/integration/api/invitations-token-route.test.ts):
  • /api/invitations/[token] GET/POST/DELETE
  • Includes preview, accept, decline, expired invite, and email-mismatch assertions

v0.2.7-dev - 2026-02-18

Phase 17 launch prep progress batch (AI-owned items).

• Added feature flag framework (lib/feature-flags.ts) and wired key surfaces:
  • connector UI gating
  • team/API settings gating
  • public API middleware gating (NEXT_PUBLIC_FF_PUBLIC_API)
  • anomaly detection gating in insights API
• Added launch-readiness UX and public pages:
  • new /changelog page rendering CHANGELOG.md
  • changelog/status links in public and dashboard navigation areas
  • dashboard billing past-due banner
  • dashboard NPS survey modal (10-query trigger, monthly cooldown, nps_submitted event)
• Added transactional email infrastructure upgrades:
  • templates: welcome, weekly digest, payment failure
  • resend helpers for welcome/digest/payment-failure delivery
  • Clerk webhook now sends welcome email on user.created
  • dunning flow now tracks failure count and stages reminder/final warning emails
• Added API versioning policy and implementation:
  • /api/v1/* responses now include X-API-Version
  • deprecation policy docs added (docs/40_reference/20_engineering/API_VERSIONING.md)
  • contracts and ops docs updated
• Expanded testing foundations:
  • fixture-based LLM response library + helper
  • new AI agent unit tests (intent, schema analysis, query generation, insight synthesis)
  • new dunning unit coverage
  • new Clerk webhook integration coverage
  • broader API auth integration matrix
  • new Playwright launch-readiness flow spec
• Validation executed:
  • npm run lint
  • npm run type-check
  • npm test -- --run
  • npm run build
  • npm run test:e2e -- --project=chromium --grep "Launch readiness critical flows"

v0.2.6-dev - 2026-02-18

Phase 16 performance foundations completion batch.

• Added high-row-count virtualization for chat results tables using @tanstack/react-virtual with sticky headers, preserved sorting/CSV copy, and null-value highlighting
• Added input debouncing improvements:
  • Chat composer now keeps immediate local input state while debouncing external updates and flushing on submit
  • Conversation history and dataset list filtering now use debounced search state
• Added bounded LRU behavior to in-memory Zustand caches:
  • Conversations capped at 50 entries (MAX_CONVERSATIONS)
  • Datasets capped at 100 entries (MAX_CACHED_DATASETS)
  • Added unit coverage for both eviction helpers
• Added client rendering performance optimizations via contain: content, content-visibility, and targeted will-change hints in chat/visualization-heavy surfaces
• Added root-layout resource hints (preconnect + dns-prefetch) for configured third-party origins
• Added Web Vitals reporting pipeline (web-vitals + WebVitalsReporter) to emit web_vitals analytics events
• Added streaming CSV load path for DuckDB files above 5MB with progress callbacks and upload-stage UI updates
• Added build/dev tooling updates:
  • next dev --turbo enabled for local dev
  • bundle analyzer integration in next.config.ts
  • npm run analyze script and CI artifact upload for .next/analyze
  • analyze script made cross-platform via cross-env
• Validation completed with npm run lint, npm run type-check, npm test -- --run, npm run build, and npm run analyze

v0.2.5-dev - 2026-02-18

Phase 15 onboarding and accessibility completion batch.

• Added built-in sample datasets and schema descriptions under public/sample-data/ for zero-friction onboarding (ecommerce-orders.csv, saas-metrics.csv, marketing-campaigns.csv, column-descriptions.json)
• Added dashboard zero-state redesign with "Try with sample data" flow, sample upload path in dataset store, and sample dataset badge rendering in dataset cards
• Replaced static prompt chips with schema-driven suggestions (lib/ai/generate-example-questions.ts) including per-dataset localStorage caching and fallback questions
• Added reusable trust badge UI and surfaced it above the fold across dashboard/upload/chat contexts
• Added activation and product analytics instrumentation:
  • Activation funnel events: dataset_uploaded, first_query, aha_moment, return_visit
  • Standard usage events: query_submitted, query_success, query_failed, chart_exported, feedback_given
  • Added missing events: conversation_created, dataset_deleted, sql_edited, chart_filter_clicked
• Enabled @vercel/analytics in root layout (<Analytics />)
• Added WCAG-focused accessibility improvements:
  • Skip-to-content link and #main-content target
  • aria-live status announcements for query processing and SSE completion summaries
  • role="status" and role="alert" wiring on progress and validation/error surfaces
  • aria-describedby on validated form inputs
  • Auto-generated chart alt text with SVG aria-label + injected <desc> summaries
• Verified with npm run lint, npm run type-check, and npm test -- --run (all passing)

v0.2.4-dev - 2026-02-18

Phase 14 cost tracking completion batch.

• Added robust token usage extraction across Anthropic/OpenAI LangChain response metadata (lib/ai/token-usage.ts) and switched routing costs to authoritative model pricing (MODEL_COSTS, calculateCost)
• Added per-agent usage attribution in orchestrator (intent_understanding, schema_analysis, query_generation) with total token/cost rollups in pipeline outputs
• Added monthly LLM budget controls with fail-open Redis behavior (lib/ai/budget.ts) and enforced caps in chat/API query routes with 402 BUDGET_EXCEEDED responses
• Added cost persistence schema and migrations:
  • messages columns: input_tokens, output_tokens, cost_usd, model_used
  • new usage_logs table + index (drizzle/0007_cost_tracking.sql, drizzle/0007_down.sql)
• Added usage persistence flow from chat UI to conversation message API and agent-level usage log writes for chat + API paths
• Added internal cost dashboard endpoint (GET /api/admin/costs) with period aggregation and cost-per-query metrics
• Added cost telemetry and alerts: query-level llm_cost PostHog events, daily spend warn threshold ($50), and daily alert event threshold ($100)
• Updated free-tier query limit defaults to env-configurable 50 (FREE_TIER_QUERY_LIMIT / NEXT_PUBLIC_FREE_TIER_QUERY_LIMIT)
• Added unit coverage for token usage extraction and model cost calculation (tests/unit/lib/ai/token-usage.test.ts)

v0.2.3-dev - 2026-02-18

Phase 13 security and compliance completion batch.

• Added security foundations: centralized app errors (lib/errors.ts), RBAC middleware (lib/middleware/rbac.ts), input sanitization helpers (lib/middleware/sanitize.ts), and audit logging helper (lib/audit.ts)
• Added DB/security schema updates: audit_logs table, datasets.deleted_at soft-delete column, connections.password_key_version, plus migration/down scripts (drizzle/0006_security_compliance.sql, drizzle/0006_down.sql)
• Hardened APIs: Stripe webhook idempotency, CSP policy headers, soft-delete + dataset restore flow, 30-day dataset purge cron endpoint, and standardized rate-limit limit headers on 429s
• Improved encryption and validation: connector key-version support, broad API request sanitization/validation, and transform-agent LLM output validation via Zod
• Completed code quality cleanup: removed remaining as unknown as assertions and eslint-disable suppressions in app/lib code paths
• Verified with npm run lint, npm run type-check, and npm test -- --run

v0.2.2-dev - 2026-02-18

Phase 12 observability and reliability completion batch.

• Added structured server logging foundation (pino) and migrated app/api + lib server paths off console.*
• Added request context + correlation ID propagation and request logging middleware (x-request-id, Sentry breadcrumb/tag wiring)
• Added /api/ready endpoint and expanded /api/health with per-check latency and overall status reporting
• Added OpenTelemetry bootstrap/instrumentation plumbing with manual spans in the AI orchestrator pipeline
• Added graceful shutdown handling (signal hooks, SSE drain notifications, shutdown callbacks, DB/telemetry cleanup hooks)
• Added Redis retry/exponential backoff improvements and circuit-breaker on-open alert callbacks
• Added ops docs: docs/40_reference/10_ops/ALERTING_RULES.md, docs/40_reference/10_ops/BACKUP_STRATEGY.md, docs/40_reference/10_ops/LOG_AGGREGATION.md

v0.2.1-dev - 2026-02-18

Phase 11 completion batch (AI-owned tasks).

• Compliance and privacy updates: cookie consent banner, expanded privacy policy (GDPR/CCPA/subprocessors/retention), account deletion API and settings flow
• Reliability and monitoring updates: circuit_breaker_open telemetry, enhanced rate-limit telemetry, Stripe dunning handling for failed payments and past-due subscriptions
• AI infrastructure updates: prompt version hashing + metadata persistence, confidence calibration telemetry on feedback, low-confidence clarification rerun with preComputedIntent
• Testing updates: new cache integration suite, webhook and chat enhancements, accessibility expansion, keyboard navigation coverage, chat history/export E2E coverage
• Data model update: added message metadata column and migration (drizzle/0005_prompt_metadata.sql)

v0.2.0-dev - 2026-02-18

Phase 11 hardening updates (security, telemetry, and testing).

• Rate limiting now fails closed when Redis is unavailable (lib/rate-limit.ts)
• Added rate_limit_hit PostHog server event capture on limit enforcement and Redis outages (lib/rate-limit.ts)
• Chat API now applies user plan-aware rate limits and includes real token usage in model usage logging (app/api/chat/route.ts)
• Added integration tests for rate-limit edge cases and expanded chat rate-limit assertions (tests/integration/api/rate-limit.test.ts, tests/integration/api/chat.test.ts)
• Fixed token usage metadata typing in query generation (lib/ai/agents/query-generation.ts)

v0.2.0-plan - 2026-02-18

Roadmap expansion and delivery planning (v0.2 through v2.0).

• 680-item feature roadmap research (docs/90_sandbox/research/FEATURE_ROADMAP_RESEARCH.md)
• 13 new PHASE files (PHASE_12-PHASE_24) covering observability, security, cost tracking, UX, connectors, collaboration, platform intelligence
• Updated ROADMAP, DELIVERY_INDEX, PHASE_INDEX, STATE, CONTRACTS, BUILD_PACK
• Audit findings added to SEC_SECURITY, PERF_PERFORMANCE, OPS_RELIABILITY, TEST_TESTING
• No code changes - documentation and planning only

v0.1.0 - 2026-02-18

Initial MVP code-complete (Phases 01-08). Not yet deployed or validated against real services.

• Next.js 15 + TypeScript strict mode, Tailwind CSS + shadcn/ui
• DuckDB WASM client-side SQL execution
• 9-agent LangGraph AI pipeline: intent -> schema -> query -> validation -> execution -> synthesis + anomaly + forecast + transform
• Chat interface with SSE streaming, agent progress, follow-up chips
• 11 chart types with auto-selection and PNG export (Recharts)
• Clerk auth + Stripe checkout + webhooks + billing portal
• PostgreSQL (Drizzle ORM, 10 tables, 5 migrations) + Upstash Redis + Pinecone semantic cache
• Organizations/teams with invitations and role-based permissions
• Database connectors (PostgreSQL, MySQL)
• Public API v1 (datasets, conversations, query) + API key management
• Chart sharing, embed page, shared conversation page
• Sentry + PostHog monitoring configured
• 16 test files: unit, integration, E2E (Playwright)
• GitHub Actions CI workflow