Phase 7 — Self-Hosting Paradigm

Phase 7 transitions pnxt from a verified prototype to a self-modifying, LLM-programmable system. It follows five milestones (M1–M5) that progressively move execution from TypeScript host code into the pnxt paradigm itself.

Milestone Overview

Milestone	Focus	Status
M1	Self-Description — pnxt describes its own pipeline as VPIR	Complete (Sprint 9)
M2	External Task Expression — real-world tasks in pure VPIR JSON	Complete (Sprint 11)
M3	LLM-Native Programming — reliable LLM-driven VPIR generation	Complete (Sprint 12)
M4	Self-Modification — pnxt modifies its own pipeline through VPIR	Planned
M5	Self-Hosting — pnxt core components expressed in pnxt	Planned

Sprint 10: Handler Library + Tool Registry (Complete)

The foundation for external task execution — standard handlers and a declarative registry.

Standard Handler Library — 8 pre-built tool handlers: http-fetch, json-transform, file-read, file-write, string-format, math-eval, data-validate, unit-convert
Declarative Tool Registry — Operation-to-handler mapping with alias support, auto-registration, discovery API, and trust pre-validation
DPN Supervisor — Supervisor actor pattern with bounded restart strategies (one-for-one, all-for-one), priority mailbox (high > normal > low), and full event log
DPN Runtime Integration — Action and inference nodes resolve handlers from the tool registry with ACI gateway fallback

Sprint 11: VPIR Authoring + External Tasks — M2 Complete (Complete)

The M2 bridge: LLM output flows directly into executable graphs without TypeScript.

VPIR Graph Builder — Fluent API (addObservation, addInference, addAction, addAssertion, addComposition) and fromJSON() for constructing validated graphs from pure JSON. Auto-computes roots/terminals, runs structural validation, validates tool availability via registry
External Task Runner — TaskRunner class orchestrating: JSON spec → fromJSON() → tool discovery → trust validation → DPN compile → DPN execute → TaskExecutionResult
Task-Aware Bridge Grammar — Enhanced LLM generation with system prompts listing all available handlers and their input schemas. Post-generation validation rejects graphs referencing non-existent handlers
External Task Benchmarks — Temperature Conversion (98.6F → 37C) and Math Expression (2*(3+4)-1 = 13) running end-to-end through DPN runtime

Sprint 12: Reliable Bridge Grammar + Error Recovery — M3 Foundation (Complete)

Making LLM-generated VPIR reliable enough for production use.

Bridge Grammar Error Taxonomy — BridgeErrorCategory enum with 6 categories (schema, semantic, handler, topology, truncation, confidence), repair hints, and formatDiagnosisForLLM() for structured retry feedback
Auto-Repair Engine — 6 repair strategies: truncated JSON closure, missing field injection, fuzzy enum matching (Levenshtein), duplicate ID renaming, auto-computed roots/terminals, default security labels
Confidence Scorer — 4-dimension scoring (structural 0.30, semantic 0.25, handler coverage 0.25, topological 0.20) with per-node low-confidence flagging
Z3 Graph Pre-Verification — 4 formal properties via Z3: acyclicity, input completeness, IFC monotonicity, handler trust compatibility
Reliable Generation Pipeline — 7-stage orchestration: LLM generation → schema validation → diagnosis → auto-repair → re-validation → confidence scoring → Z3 verification. Full pipeline stage tracing with timing
Error Recovery Benchmark — 7 scenarios: truncated JSON, missing fields, invalid handlers, cyclic graphs, wrong enums, mixed validity, duplicate IDs

What’s Next: M4 and M5

M4: Self-Modification

pnxt modifies its own pipeline through VPIR:

Pipeline stages become VPIR-editable graphs
Agents propose optimizations as graph transformations
Changes verified via Z3 before application
Transport ensures verified properties survive modification

M5: Self-Hosting

pnxt’s core components expressed in pnxt itself:

Categorical syntax replaces JSON intermediate format
Bootstrap compiler validates the paradigm end-to-end
Full transition from TypeScript host to pnxt-native execution

Advisory Panel Score Trajectory

Sprint	Score	Key Advancement
Phase 4	3.0/10	Infrastructure only — no paradigm components
Sprint 5	7.5/10	HoTT + Knowledge Graph + end-to-end pipeline
Sprint 9	9.2/10	Self-hosting PoC + categorical tokenization
Sprint 10	9.25/10	Handler library + tool registry
Sprint 12	9.35/10	Reliable bridge grammar + M3 foundation