Historical task source. The Eve task board holds current task state.
Phase 106: Kalika Integration — Cross-Domain, SDK, Database, Apps, Benchmarks & Research Workflows#
Final integration phase: connects all Kalika libraries into a cohesive platform, builds the application layer (web UI, API services, SDK), establishes the database schema, implements end-to-end research workflows, and creates comprehensive benchmarks proving Kalika meets or exceeds every SOTA system identified in the design research.
106.1 Database & Persistence (@kalika/database)#
106.1.1 Schema Design#
- 106.1.1.1 Design and implement PostgreSQL schema for
kalikadatabase: tables for notebooks (id, title, owner, cells JSONB, metadata, created_at, updated_at), computations (id, notebook_id, cell_id, input_expr, output_expr, derivation_chain JSONB, verification_status, execution_time_ms, created_at), citations (id, bibtex, doi, arxiv_id, metadata JSONB), knowledge_graph_nodes (id, type, name, description, properties JSONB), knowledge_graph_edges (source_id, target_id, relationship_type, properties JSONB), user_preferences (user_id, settings JSONB), and computation_cache (input_hash, output, created_at, access_count) - 106.1.1.2 Implement database migrations: use Drizzle ORM or Kysely for type-safe schema management, with forward-only migrations, seed data for physical constants and particle data, and rollback support
- 106.1.1.3 Implement vector search integration: store embeddings of theorems, concepts, and paper abstracts in Qdrant (via @shared/infrastructure or Sophia integration), enabling semantic search across the knowledge base — "find theorems similar to the Atiyah-Singer index theorem"
- 106.1.1.4 Implement computation cache with content-addressed storage: hash the input expression + parameters + library version -> if cache hit, return stored result immediately (with verification that cached result is still valid) — dramatically speeding up repeated computations in notebooks
106.1.2 Data Import/Export#
- 106.1.2.1 Implement LMFDB data import: L-functions and Modular Forms Database — import elliptic curve data, modular form data, number field data, and L-function zeros for use in @kalika/number-theory computations
- 106.1.2.2 Implement HEPData import: import published experimental results (cross-sections, distributions, correlation matrices) from HEPData.net for comparison with theoretical predictions
- 106.1.2.3 Implement SXS waveform catalog import: import numerical relativity gravitational waveforms from the SXS collaboration for use in @kalika/gravitational-waves
- 106.1.2.4 Implement Kreuzer-Skarke database integration: access the database of 473,800,776 reflexive polytopes in 4D for @kalika/string-theory Calabi-Yau computations
- 106.1.2.5 Implement OEIS API client: query the Online Encyclopedia of Integer Sequences for sequence identification and cross-referencing
- 106.1.2.6 Implement KnotInfo/LinkInfo database integration: import knot invariant data for @kalika/topology knot theory computations
106.2 Application Layer (apps/kalika/)#
106.2.1 Compute Service (apps/kalika/svc-compute)#
- 106.2.1.1 Implement compute service: Express/Fastify HTTP server exposing CAS operations as REST/WebSocket APIs — endpoints for: symbolic computation (simplify, differentiate, integrate, solve, series, limit), numerical evaluation, matrix operations, tensor operations, and batch computation (submit multiple expressions, get results as they complete)
- 106.2.1.2 Implement WebSocket streaming: for long-running computations (Groebner basis, large simplification, numerical simulation), stream progress updates and partial results to the client — with cancellation support
- 106.2.1.3 Implement computation queue: Redis-backed job queue for CPU-intensive computations (IBP reduction, lattice MC), with worker pool, priority levels (interactive > batch > background), timeout management, and result caching
- 106.2.1.4 Implement WASM-in-browser mode: for lightweight computations, run the CAS engine directly in the browser via WASM (no server round-trip), falling back to server for heavy computations — providing sub-millisecond response for simple operations
106.2.2 Agent Service (apps/kalika/svc-agents)#
- 106.2.2.1 Implement agent service: manages autonomous research agents (from @kalika/research-agents), handles agent lifecycle (create, run, pause, resume, terminate), resource allocation (API keys for Claude, compute budget), and result persistence
- 106.2.2.2 Implement agent API: REST endpoints for: creating research tasks, monitoring agent progress, reviewing intermediate results, providing human feedback, and retrieving final reports with citations and proofs
- 106.2.2.3 Implement agent rate limiting and cost management: track API usage (Claude tokens, compute time), enforce per-user budgets, and provide cost estimates before starting expensive agent tasks
106.2.3 Notebook Service (apps/kalika/svc-notebooks)#
- 106.2.3.1 Implement notebook service: CRUD for notebooks, cell execution management (queue cells for execution, manage kernel state), collaboration state (CRDT synchronization), and notebook export (HTML, PDF via LaTeX, .ipynb)
- 106.2.3.2 Implement notebook templates: pre-built notebooks for common research workflows — "Getting Started with Quantum Mechanics", "Computing Feynman Diagrams", "Exploring Calabi-Yau Manifolds", "Gravitational Wave Parameter Estimation", "Number Theory Explorations" — each with runnable examples and explanatory text
106.2.4 Web Application (apps/kalika/web)#
- 106.2.4.1 Implement notebook UI: Monaco-based code editor for TypeScript cells, KaTeX-rendered LaTeX for math cells, Markdown editor for prose cells, interactive visualization canvas for plot cells, cell reordering (drag-and-drop), cell type switching, and keyboard shortcuts (Shift+Enter to execute, Ctrl+Enter to execute and advance)
- 106.2.4.2 Implement math input: LaTeX input with real-time preview, palette of common symbols/operators/functions, handwriting recognition (via ML model for math symbols), and natural language to expression conversion ("integrate sin(x) from 0 to pi" -> Integral expression)
- 106.2.4.3 Implement computation explorer: side panel showing the derivation chain for any computed result (expandable tree of steps), verification status badges (proven/conjectured/numerical), and one-click formal verification button
- 106.2.4.4 Implement knowledge graph browser: interactive graph visualization of the mathematical knowledge base, with node expansion, filtering by domain, search, and navigation to related theorems/concepts
- 106.2.4.5 Implement citation sidebar: search and insert citations, view bibliography, manage citation library, and sync with external managers (Zotero via API)
106.2.5 BFF (Backend for Frontend)#
- 106.2.5.1 Implement BFF layer: aggregate compute, agent, and notebook service APIs into a single frontend-optimized API, handle authentication (via @shared/auth), session management, user preferences, and WebSocket connection management for real-time features
106.3 SDK (@kalika/sdk)#
- 106.3.1.1 Implement TypeScript SDK: importable npm package providing the full CAS API (symbolic computation, simplification, differentiation, integration, solving, tensor operations, physics utilities) for use in Node.js/browser applications — zero server dependency (uses WASM engine directly)
- 106.3.1.2 Implement Python SDK: Python package wrapping the TypeScript SDK via subprocess or REST API calls, providing a SymPy-compatible API surface (so researchers familiar with SymPy can switch with minimal code changes), plus Kalika-specific capabilities (derivation chains, formal verification)
- 106.3.1.3 Implement CLI tool: command-line interface for Kalika
computations —
kalika eval "integrate(sin(x)^2, x)",kalika simplify "...",kalika solve "...",kalika notebook run file.kalika-nb— for scripting and pipeline integration - 106.3.1.4 Implement Jupyter kernel: Kalika computation kernel for Jupyter notebooks, enabling Kalika CAS computations from standard Jupyter — for users who prefer the Jupyter ecosystem
106.4 Cross-Domain Integration#
106.4.1 Sophia Integration (Research & Knowledge)#
- 106.4.1.1 Integrate Kalika knowledge graph with Sophia's document ingestion: when Sophia ingests physics/math papers, extract theorems, definitions, and conjectures, and add them to Kalika's knowledge graph — enriching the mathematical knowledge base from literature
- 106.4.1.2 Integrate Kalika citations with Sophia's citation management: share citation database, avoid duplication, and enable cross-domain citation search
106.4.2 Metis Integration (Education)#
- 106.4.2.1 Integrate Kalika CAS with Metis educational platform: enable interactive mathematical exploration in Metis courses — students can symbolically solve equations, visualize functions, and verify solutions within the learning environment
- 106.4.2.2 Implement step-by-step solution mode: given a problem (e.g., "solve x^2 - 5x + 6 = 0"), show the solution step by step with explanations at each step, using the derivation chain to generate pedagogically appropriate explanations — matching Mathematica's Step-by-Step Solutions and Maple's tutors
106.4.3 Nyx Integration (Astronomy)#
- 106.4.3.1 Integrate Kalika orbital mechanics (@kalika/general-relativity geodesics + Newtonian limit) with Nyx ephemeris calculations: provide relativistic corrections to Nyx's orbital computations where needed (Mercury perihelion precession, GPS relativistic corrections, pulsar timing)
- 106.4.3.2 Integrate Kalika gravitational wave computations with Nyx event catalog: when Nyx catalogs a gravitational wave event, provide theoretical waveform comparison and parameter estimation
106.4.4 Nous Integration (Sovereign AI)#
- 106.4.4.1 Register Kalika's research agents with Nous agent framework: enable Nous to dispatch mathematical/physics computation tasks to Kalika agents, and enable Kalika agents to use Nous's general-purpose AI capabilities (language understanding, planning, tool use)
- 106.4.4.2 Integrate Kalika's formal verification pipeline with Nous's safety verification: use Lean 4 formal proofs to verify critical AI system properties (safety constraints, invariant preservation)
106.5 End-to-End Research Workflows#
- 106.5.1.1 Implement "Compute and Verify" workflow: user enters expression -> CAS computes result -> result is numerically spot-checked -> optionally formally verified via Lean 4 -> result displayed with verification badge and derivation chain — the core workflow for trusted computation
- 106.5.1.2 Implement "Literature-Informed Computation" workflow: user asks a research question -> literature agent searches arXiv/INSPIRE -> computation agent implements relevant techniques from papers -> results compared with published values -> discrepancies flagged for investigation
- 106.5.1.3 Implement "Explore and Conjecture" workflow: user specifies a mathematical structure -> exploration agent computes invariants and properties -> pattern detection identifies regularities -> conjecture agent formulates precise conjectures -> proof agent attempts to prove/disprove -> results presented as a mini research report
- 106.5.1.4 Implement "Theory-to-Observable" workflow for particle physics: user specifies BSM model parameters -> Lagrangian built -> RGE running -> spectrum computation -> Feynman diagram generation -> cross-section/decay rate computation -> comparison with experimental data (PDG, LHC) -> constraints on parameter space — a complete phenomenology pipeline matching the FeynRules -> SARAH -> SPheno -> MadGraph -> CheckMATE chain
- 106.5.1.5 Implement "Paper Generation" workflow: user selects notebook cells for publication -> Kalika generates LaTeX document with equations, figures, tables, derivations, bibliography -> user reviews and edits -> export to compilable .tex file or submit directly to arXiv via API
106.6 Comprehensive Benchmarks#
106.6.1 CAS Benchmarks (vs Mathematica, SageMath, SymPy, FriCAS)#
- 106.6.1.1 Create symbolic integration benchmark: 500+ integrals from Bondarenko, Charlwood, Timofeev, Rubi test suites — measure correctness (percentage of integrals correctly evaluated) and performance (time per integral) — target: exceed SymPy and FriCAS correctness, within 10x of Mathematica performance
- 106.6.1.2 Create polynomial algebra benchmark: Groebner basis (cyclic-n for n=4..8, Katsura-n for n=5..12), polynomial factorization (degree 50/100/500 over Z), polynomial GCD (sparse multivariate, dense univariate) — target: match SINGULAR/Magma for GB, match FLINT for GCD
- 106.6.1.3 Create simplification benchmark: 300+ expressions requiring trigonometric, exponential, radical, and mixed simplification — compare final expression complexity (node count) against Mathematica's FullSimplify output — target: equal or better simplification via e-graph saturation
- 106.6.1.4 Create ODE/PDE solver benchmark: 200+ differential equations from Kamke's catalog (ODEs) and standard PDE test problems — measure solvability rate and correctness — target: match Maple for ODEs, exceed SymPy for PDEs
106.6.2 Physics Benchmarks (vs FeynCalc, xAct, Package-X, FORM)#
- 106.6.2.1 Create QFT benchmark: one-loop QED corrections (vertex, vacuum polarization, self-energy), QCD beta function (one/two-loop), top quark decay width at NLO — compare analytic results against published values and FeynCalc output
- 106.6.2.2 Create GR benchmark: compute all curvature tensors for 20+ exact solutions (Schwarzschild, Kerr, FLRW, Godel, pp-wave, etc.), verify Einstein equation satisfaction, compare against xAct output — target: match xAct correctness with comparable or better performance
- 106.6.2.3 Create tensor canonicalization benchmark: expressions with 10/20/50/100 indices and various symmetry groups — measure canonicalization time against xPerm — target: within 2x for all cases, match for simple symmetries
106.6.3 Numerical Benchmarks (vs LAPACK, PETSc, SUNDIALS)#
- 106.6.3.1 Create dense linear algebra benchmark: eigenvalue computation for matrices of size 100/500/1000/5000, SVD, matrix exponential — compare against OpenBLAS/LAPACK — target: within 3x for native, 10x for WASM
- 106.6.3.2 Create sparse solver benchmark: Poisson equation on N x N grid (N = 100, 500, 1000, 5000) with CG + AMG preconditioner — compare against PETSc — target: within 5x of PETSc for moderate problems
- 106.6.3.3 Create ODE solver benchmark: stiff and non-stiff test problems from Hairer-Norsett-Wanner test suite — compare against SUNDIALS CVODE — target: match accuracy with comparable performance
106.6.4 AI & Verification Benchmarks#
- 106.6.4.1 Create formal verification benchmark: attempt to formally verify 100 CAS computation results in Lean 4 — measure success rate (percentage verified), average verification time, and types of failures — target: >80% automatic verification rate for standard algebraic identities
- 106.6.4.2 Create symbolic regression benchmark: AI Feynman benchmark (100 physics equations) — measure correct recovery rate and computational time — target: match or exceed PySR's published results
- 106.6.4.3 Create conjecture generation benchmark: given data for known mathematical sequences/structures, measure rate of rediscovering known results (e.g., Euler's partition theorem, quadratic reciprocity patterns, Catalan number formulas) — establishing a baseline for conjecture quality
106.7 Documentation & Onboarding#
- 106.7.1.1 Write API reference documentation: auto-generated TypeDoc for all public APIs in all @kalika/* libraries, with usage examples for every major function
- 106.7.1.2 Write tutorial notebooks: 10 tutorial notebooks covering — (1) Getting Started (basic CAS operations), (2) Calculus (differentiation, integration, series), (3) Linear Algebra, (4) Quantum Mechanics, (5) Classical Mechanics (Lagrangian/Hamiltonian), (6) Electrodynamics, (7) General Relativity, (8) Quantum Field Theory (Feynman diagrams), (9) Number Theory, (10) Formal Verification with Lean 4
- 106.7.1.3 Write domain guide: comprehensive guide to Kalika's capabilities, organized by physics/math domain, with pointers to relevant libraries and functions for each topic — serving as a "which tool do I use for X?" reference
- 106.7.1.4 Write migration guides: for users coming from Mathematica (function name mapping, notation differences), SymPy (API comparison), SageMath (workflow comparison), and FeynCalc (QFT workflow comparison) — reducing the barrier to adoption
106.8 Final Integration Tests#
- 106.8.1.1 End-to-end test: compute hydrogen atom energy levels from scratch (define Hamiltonian, solve Schrodinger equation, extract eigenvalues) and verify E_n = -13.6/n^2 eV — testing QM + CAS + numerical integration
- 106.8.1.2 End-to-end test: compute anomalous magnetic moment of the electron at one loop (generate Feynman diagram, evaluate amplitude, take trace, evaluate loop integral, renormalize) and verify a_e = alpha/(2*pi) — testing QFT + Feynman engine + loop integrals
- 106.8.1.3 End-to-end test: compute Schwarzschild geodesic precession and verify Mercury perihelion advance formula delta_phi = 6piGM/(c^2a*(1-e^2)) — testing GR + tensor calculus + geodesic solver
- 106.8.1.4 End-to-end test: discover Newton's law of gravitation (F = GMm/r^2) from synthetic orbital data via symbolic regression — testing AI + symbolic regression pipeline
- 106.8.1.5 End-to-end test: formally verify in Lean 4 that the integral of sin(x) from 0 to pi equals 2 — testing CAS + formal verification bridge
- 106.8.1.6 End-to-end test: generate a complete notebook computing Ising model critical temperature, visualize phase transition, and export as LaTeX paper with bibliography — testing notebooks + statistical mechanics + visualization + typesetting + citations