Design and performance
The handbook and reference say what SSPUR does. This section says why, and how well it works.
Design documents
These describe the system as a whole. They were written before most of the compiler, so where they and the language reference disagree, the reference describes what is implemented.
| Document | Contents |
|---|---|
| Vision | Why SSPUR exists and the principles behind it |
| Core semantics | Types, effects, contracts, memory and concurrency |
| Graph model | Content-addressed definitions, operations and the query API |
| Text projection | The token-optimized source format |
| Deploy model | Services, stores, and effects as permissions |
| Systems layer | Systems programming and the C++ parity matrix |
| AI-native constructs | Guess, taint types, decision tables and other agent-oriented types |
| Roadmap | The phases and their exit criteria |
Decisions
The ADR index lists all 44 architecture decision records, from the foundations to proving F64 to Int conversions safe.
Measurements
| Benchmark | What it measures |
|---|---|
| Native performance | Run time and peak memory of native SSPUR against idiomatic C++ with the same safety checks |
| Agent benchmarks | Tokens, calls and hidden-test pass rates of Claude agents doing feature work in SSPUR, Python, TypeScript and Go |
The Phase 1 token count, which compares hand-written samples in six languages, is in bench/tokens/.