GuideArch

Modern fuzzy multi-criteria decision analysis for software architecture, shipped as three concurrent implementations sharing one spec.

GuideArch helps software architects pick between competing technology stacks by modeling a decision space — decisions, alternatives, quality properties with priority weights, and three families of constraints — and ranking the resulting candidate architectures using fuzzy TOPSIS. It also reports which decisions matter most (sensitivity analysis) and which constraints are most binding (elimination counting).

1. Status

v1.0.0 released — M5 complete — release artifacts wired. All three implementations (TypeScript, C#, Python) are feature-complete: full domain + TOPSIS engine, editors, results, analysis charts, and release-ready build configs. Tauri installers, Avalonia self-contained binaries, Python wheel + Docker image, and a SvelteKit static web bundle are all produced by the GitHub Actions release workflow on every v*.*.* tag push.

Milestone tags: v0.0.0-bootstrap (M0), v0.1.0-m1 (M1 — domain + TOPSIS), v0.2.0-m2 (M2 — ViewModels + skeleton UI), v0.3.0-m3 (M3 — full editors), v0.4.0-m4 (M4 — analysis + charts), v1.0.0 (M5 — release artifacts).

2. What's in the box

Three implementations of the same application, kept in conformance by a shared spec and a cross-impl test corpus.

All three are built on the VMx MVVM framework, included as a git submodule at vendor/vmx/.

3. Documentation hub

Read most-to-least essential. Each link includes when to use it.

3.1 Architecture & design

• Design specification — the full architectural spec: domain model, TOPSIS algorithm, per-impl design, conformance strategy, GitHub plan, v1 scope, M0–M5 roadmap. Read this first if you want to understand the project, contribute to design, or evaluate the architecture.

3.2 Milestone plans

• M0 — Repo bootstrap — the 27-task implementation plan that produced the bootstrap state. Read this if you want to understand how the scaffolding was assembled.
• M2–M5 plans land in docs/plans/ as each milestone begins. M1 was executed directly from the spec without a separate plan document because the algorithm was already fully specified in spec/algorithms/topsis.md.

3.3 Specification & conformance

• spec/ — the language-neutral source of truth that every implementation must satisfy. Populated at M1 with:
  • spec/algorithms/topsis.md — the canonical TOPSIS pipeline with magic-number table and tie-break rule.
  • spec/algorithms/critical-decisions.md, critical-constraints.md — reference cards.
  • spec/domain/scenario.schema.json — JSON Schema 2020-12 for the input format.
  • spec/domain/glossary.md, invariants.md — vocabulary and load-time validation rules.
  • spec/conformance/ — the seed corpus: scenarios/sas.json (10 decisions, 25 alternatives, 7 properties), scenarios/eds.json (same shape), and the matching expected/*.json outputs the three impls must reproduce within tolerances.json (1e-9 absolute on scalars; ranking exact).

3.4 Architecture decision records

Numbered rationale for each non-obvious design choice. Read when questioning why something is the way it is.

• ADR-0001 — Three implementations sharing one spec; VMx as submodule
• ADR-0002 — JSON Schema for scenario files (not legacy XML)
• ADR-0003 — TOPSIS as in-repo code; no Microsoft Solver Foundation
• ADR-0004 — MIT License
• ADR-0005 — Single monorepo version; all three impls release together
• ADR-0006 — NiceGUI 3.x as the Python view layer (not Shiny, not Streamlit)

3.5 Contributing & governance

• CONTRIBUTING.md — feature workflow (spec-first, all three impls in lockstep), local development, test layout, code style per language. Read before opening a PR.
• CODE_OF_CONDUCT.md — Contributor Covenant 2.1. Report violations to kaveh.razavi@gmail.com.
• SECURITY.md — private vulnerability reporting.

4. Repository layout

spec/                language-neutral spec (schema, algorithms, conformance corpus, ADRs)
vendor/vmx/          VMx submodule (do not edit directly; PR upstream)
langs/typescript/    TS + Tauri 2 + Svelte 5 implementation
langs/csharp/        C# + Avalonia 12 implementation
langs/python/        Python + NiceGUI 3.x implementation
tools/               cross-cutting scripts (VMx mode switch, legacy XML import)
docs/                design specs and milestone plans
.github/             CI workflows, issue/PR templates, dependabot config

5. Quickstart

5.0 Clone with submodules

VMx lives at vendor/vmx/. Without it initialized, every impl below fails because VMx imports won't resolve.

git clone --recurse-submodules https://github.com/thekaveh/GuideArch.git
# or, if already cloned without --recurse-submodules:
cd GuideArch
git submodule update --init

5.1 Prerequisites

5.2 TypeScript (Tauri 2 + Svelte 5)

cd langs/typescript
pnpm install
pnpm dev          # web mode — browser at http://localhost:1420
pnpm tauri dev    # desktop mode — native Tauri window (first build is slow)

Ctrl-C (or close the window) to stop.

5.3 C# (Avalonia 12)

cd langs/csharp
dotnet build
dotnet run --project src/GuideArch.View   # desktop Avalonia window

Close the window to stop. For the WebAssembly build see langs/csharp/README.md.

5.4 Python (NiceGUI 3.x)

cd langs/python
uv sync
uv run guidearch           # web mode — browser at http://localhost:8080
uv run guidearch --native  # desktop mode — native pywebview window

Ctrl-C (or close the window) to stop.

5.5 Try the sample scenarios

Each app ships SAS (Service-Oriented Architecture, 10 decisions / 25 alternatives / 7 properties) and EDS (Enterprise Decision Space, similar shape) as bundled samples. After launching any flavor, click the toolbar button Open Sample SAS (or Open Sample EDS) — the candidates table populates immediately. The legacy *.xml versions converted to JSON via tools/import-legacy-xml.py live under spec/conformance/scenarios/ if you want to inspect them.

The recommended exploration flow:

1. Open Sample SAS in the toolbar.
2. Results tab — top candidate's score should be 0.031180695179944085. The bar chart on the right shows the top 30; click any bar to jump to that candidate.
3. Properties tab — change one property's weight (e.g., bump Reliability to 9). Watch the candidates table refresh within ~100 ms.
4. Critical decisions tab — see which architectural choices drive the result most.
5. Critical constraints tab — see which constraints eliminate the most candidates.
6. Save As… to a temp file; New; then Open… the file you just saved — the edit should round-trip.

5.6 Run the conformance suite

Each impl ships a runner that solves every scenario in spec/conformance/scenarios/ and compares against spec/conformance/expected/ within 1e-9 absolute. CI fails on divergence.

# Python
cd langs/python && uv run python -m guidearch.conformance.runner

# C#
cd langs/csharp && dotnet run --project src/GuideArch.Conformance

# TypeScript
cd langs/typescript && pnpm conformance

5.7 Run the unit + integration test suites

Each impl ships VM-layer integration tests (load scenario → exercise ViewModel → assert results — no UI mounted) that prove MVVM separation works.

cd langs/python     && uv run pytest tests/ -q
cd langs/csharp     && dotnet test --nologo
cd langs/typescript && pnpm test

5.8 Troubleshooting

6. License

MIT — see LICENSE.