Skip to content

v1.8.0 — Three Frontiers

Choose a tag to compare

View all tags
@eidetic-works eidetic-works released this 03 Apr 10:51
· 170 commits to main since this release
v1.8.0
a4de087

Three Frontiers: AI reliability that compounds

Every AI output verified. Every correction recorded. Every mistake trains the system.

GROUND — Machine verifies

5-tier execution verification: diff, syntax, imports, tests, runtime. Receipts logged to .brain/verification_log.jsonl.

ALIGN — Human corrects

One-call corrections via nucleus_align. Each correction writes a verdict, records a delta, creates a DPO training pair, and emits an event. The missing piece that closes the loop.

COMPOUND — System learns

Deltas measure the gap between intent and reality. Recurring patterns become strategy engrams. Negative deltas become training signal.

What's new in 1.8.0

  • Three Frontiers loop fully wired — GROUND/ALIGN/COMPOUND events fire and feed each other
  • Ambient health footer — every tool response shows [frontiers: GROUND N | ALIGN N | COMPOUND N]
  • brain://health resource — IDE sidebar visibility for frontier counts
  • 7 compounding arteries — brief→engram, heartbeat→tasks, orchestrator←engrams, triggers→events
  • Delta primitives — Jaccard similarity, pattern extraction, reinvestment into strategy
  • Quality grades — copper/silver/gold/platinum weighting on training pairs
  • 6 business events — growth gates, content performance, distribution signals
  • Weekly synthesis — unified convergence view across all dimensions
  • Claude Code hook — GROUND Tier 1-2 on every file edit, pre-commit hook blocks broken code

Install

pip install nucleus-mcp==1.8.0
nucleus init --recipe founder

Or via npm: npx nucleus-mcp@1.8.0

Assets 2
Loading