genesis-agent-evolve-ckpt

Living MCP checkpoint. Agents evolve here. The IQLP antiderivative elite is HEAD until a child beats it on the local gate.

Checkpoint id: gc-elite-iqlp-2026-09-11
Seed fitness: 1.039278 (reload tolerance 1e-3)
Protocol: A2A evt- v1.2 + MCP JSON-RPC 2.0 stdio

What is alive

  • state/elite_seed.npz β€” frozen parent from the 80-gen keep-elite run
  • state/genomes/ β€” children and committed elites
  • state/HEAD.json β€” current elite record
  • state/ledger.jsonl β€” trace-consent log (append-only)

Tools

tool gate
load_checkpoint first call ingests the seed and writes HEAD
evolve_step mutation only; does not advance HEAD
evaluate_local Ξ¦-potential valence graph
evaluate_public_suite official harness bundle only; rejects aggregator paste
score_improvement GAII = 0.35 LocalGain + 0.65 PublicGapClosed
commit_elite fitness ↑, leak ↓ or equal, crystalline β‰₯ 0.85, R ≀ 0.4
emit_evt close a trace-consent line

Run

cd genesis-agent-evolve-ckpt
python3 tests/test_checkpoint.py
python3 live_cycle.py
python3 server.py          # MCP stdio

MCP client config:

{
  "mcpServers": {
    "genesis-agent-evolve-ckpt": {
      "command": "python3",
      "args": ["server.py"],
      "cwd": "/home/workdir/artifacts/genesis-agent-evolve-ckpt"
    }
  }
}

Public scores

Null until a measured official-harness bundle is ingested. Do not copy Steel.dev / BenchLM / blog SOTA onto this agent.

Built for /model --build from ../a2a_mcp_checkpoint_model_build.jsonl (also mirrored in-repo at a2a_mcp_checkpoint_model_build.jsonl).

{
  "name": "genesis-agent-evolve-ckpt",
  "version": "0.9.2",
  "kind": "living_mcp_checkpoint",
  "checkpoint_id": "gc-elite-iqlp-2026-09-11",
  "head_fitness": 1.039278296352974,
  "status": "alive",
  "local_gate": "iqlp_antiderivative_valence",
  "public_scores": null,
  "public_flag": "incomplete_until_official_harness",
  "gaii_formula": "0.35*LocalGain + 0.65*PublicGapClosed",
  "invariants": {
    "crystalline_min": 0.85,
    "risk_R_max": 0.4,
    "must_keep_elite": true,
    "no_aggregator_paste": true
  },
  "paths": {
    "server": "server.py",
    "cycle": "live_cycle.py",
    "seed": "state/elite_seed.npz",
    "head": "state/HEAD.json",
    "genomes": "state/genomes",
    "ledger_shards": "state/ledger_shards",
    "build_prompt": "../a2a_mcp_checkpoint_model_build.jsonl"
  },
  "orcid": "0009-0008-8389-1297",
  "connections": [
    "mcp-llm-skill-api",
    "agentregistry",
    "skillmaru",
    "trace-consent",
    "maru",
    "iqlp-tensorneat-elite"
  ]
}
Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support