100%

04 — Mechanism-Class Portfolio / MAP-Elites Archive

Goal

Replace the single-survivor funnel + one-in-flight throttle with a MAP-Elites portfolio of divergent candidates keyed by mechanism-class, sampled as inspiration. Internal persistent diversity is the empirically-decisive factor (EvoLattice).

Implementation Sketch

  • mechanism_class emitted by the generator from a plan-declared enum (LLM-tagged), schema-validated.
  • archive keyed by niche; the one-in-flight throttle becomes a bounded portfolio cap; one-writer + read-API §11.6 unchanged.
  • inspiration sampler injects K archived divergent candidates (incl. failed) into the prompt.

Spec References

  • scripts/ledger_evolution_runtime/{prompts.py:149-196, pipeline.py:78-176, novelty.py:51-97, gates.py:201-203}
  • script-first.md §11 plan-agnostic/domain-agnostic (plan-declared taxonomy via registry)
  • learning-ledger.md §11.6 one-writer + one-read-API (preserved)
  • Prior art: FunSearch island model + AlphaEvolve inspiration sampling + EvoLattice internal persistent diversity

Work Items

  • Implement the plan-declared mechanism-class taxonomy + LLM-tagged behavior_descriptor.mechanism_class emission at generation (prompts.py:149-196 / pipeline.py:78-176), validated against the plan-declared enum (per the descriptor decision; plan-agnostic per script-first.md §11 domain_gates.py registry pattern).
  • Implement the MAP-Elites archive: retain divergent candidates keyed by mechanism-class niche; revise the at-most-one-in-flight throttle (gates.py:201-203 open-evolution no-op) to a bounded portfolio (the deliberate invariant change; one-writer + read-API §11.6 preserved).
  • Inspiration sampling: feed archived divergent candidates (including failed ones) into the generation prompt (prompts.py) as FunSearch/AlphaEvolve-style inspirations; cover with scripts/ledger_evolution_runtime/tests/test_archive.py.

Fresh intel (regenerated)

{ “schema_version”: 1, “target”: { “kind”: “plan-section”, “ref”: “evolution-empirical-redesign/s-3a8644f1” }, “generated_at”: “2026-06-19T23:15:32.345973+00:00”, “graph_state”: { “head_sha”: “e4b7c39a”, “last_code_import_at”: “2026-06-19T23:12:07.437Z”, “embedding_stale”: false, “insights_stale”: false, “cpg_stale”: false }, “surfaces”: {}, “summary”: “intel-package[plan-section:evolution-empirical-redesign/s-3a8644f1] agent-authored dossier”, “degraded”: false, “dossier”: { “objective_symbols”: [], “tiers”: {}, “agent_authored”: true, “difficulty”: { “class”: “routine”, “research_online”: false, “research_mode”: “auto”, “signals”: [], “plan_dir”: “/home/eric/projects/ori_lang/plans/evolution-empirical-redesign” } } }

(full dossier: portfolio-archive—s-3a8644f1.intel.json)