Roadmap

Roadmap

The blueprint is released at ABI v0.2 with a runnable reference. This page sketches the direction. It is intentionally modest — this is a reference, and the project's own discipline (Governance and Constitution) is that capability is added deliberately, with evidence, not for its own sake.

Status legend: ✅ shipped · 🔜 planned · 💡 idea / discussion welcome

Shipped (v0.2)

• ✅ Layered architecture (7 layers) with conformance status
• ✅ Cognitive Kernel + ABI v0.2 (load-time conformance gate, budget guarantee)
• ✅ Policy Hook Surface (most-restrictive-wins, fail-closed, audited) + seed policies
• ✅ Capability Provider model + deterministic router + sovereignty constraint
• ✅ Reality Grading Loop (stake → grade vs evidence → validated episode / belief revision)
• ✅ Governance model (propose/ratify ceremony, state-ownership tripwire concept)
• ✅ Runnable, dependency-free reference + 13 offline conformance tests + CI

Planned

• 🔜 Reference tripwire — a shippable test_state_ownership example that fails on shadow state.
• 🔜 rg-2 judge example — content-quality grading + a worked Migration Note.
• 🔜 Provider evaluation matrix — a template for recording which provider mode carries which mission type, filled from evidence.
• 🔜 More seed policies — irreversibility guard, tool allow-lists, disclosure variants.
• 🔜 Porting guides — implementing the specs in TypeScript / Go / Rust, with the Python reference as the conformance oracle.

Ideas (discussion welcome)

• 💡 A minimal MCP / external-tool provider example.
• 💡 A projection layer demonstrating conversation/episodic memory as a read-only view over the canonical store (never a second source of truth).
• 💡 A visual audit-trace viewer for the telemetry explain() output.

How to influence it

Open an issue describing the problem or the decision. Architectural changes start as an ADR (copy ADR-TEMPLATE.md) so the why is recorded before the code. See Contributing.