MONTREAL.AI / SKILLOS
Autonomous RSI Metamaterials Discovery Proof
Recursive self-improvement on lightweight structural metamaterial discovery.
Current status
PASSED_AUTONOMOUS_RSI_METAMATERIALS_DISCOVERY_MARKET_PROOF
No human review. No emails. No invoices. No CloudOps, cyber, or silicon reuse. No customers. No private data. No API keys. Deterministic holdout benchmark.
+100.0 ptsfeasible-rate gain
100.0%final feasible rate
95.6%design-time reduction
1new Pareto candidates
Recursive self-improvement curve
Discovery frontier
Before / after on holdout design briefs
| Metric | Baseline | SkillOS RSI |
|---|---|---|
| Feasible design rate | 0.0% | 100.0% |
| Average score | 1.09 | 100.0 |
| Material efficiency | 0.89 | 3.504 |
| Manufacturing failure rate | 100.0% | 0.0% |
| Average design days | 15.94 | 0.7 |
| Average cost | $768315.46 | $33787.2 |
| Pareto candidates | 1 | 2 |
Final learned discovery skills
- skill_load_path_topology — Select lattice topology from load mode instead of using a generic grid.
- skill_manufacturing_constraints — Respect minimum feature size, overhang, and node-count manufacturability.
- skill_hierarchical_stiffness — Use hierarchical subcells when stiffness target is high under mass constraint.
- skill_graded_density_load_paths — Move density toward high-stress load paths and remove density elsewhere.
- skill_mass_pruning — Prune mass in low-stress regions once feasibility is achieved.
- skill_pareto_frontier_selection — Select candidates by constrained Pareto score, not a single naive metric.
- skill_buckling_safety_margin — Raise slenderness safety on compression and torsion cases.
- skill_triangular_compression_tension — Use triangular or octet truss families for axial compression and tension.
- skill_cross_braced_shear_torsion — Add cross-bracing and closed loops for shear and torsion.
- skill_thermal_channel_separation — Reserve aligned thermal channels while preserving load paths.
- skill_gyroid_multiaxis_isotropy — Use gyroid-like continuous lattices for multi-axis isotropy.
- skill_auxetic_impact_absorption — Use auxetic cells for impact-energy absorption.
Proof gates
- ✅ not email workflow
- ✅ not invoice workflow
- ✅ not cloudops workflow
- ✅ not cyberdefense workflow
- ✅ not silicon verification workflow
- ✅ scientific discovery workflow
- ✅ no human review required
- ✅ no emails sent
- ✅ no customers contacted
- ✅ no private data used
- ✅ no api keys required
- ✅ deterministic reproducible benchmark
- ✅ recursive self improvement releases at least 6
- ✅ rsi validation improves monotonically
- ✅ train cases at least 400
- ✅ validation cases at least 200
- ✅ holdout cases at least 800
- ✅ final skills at least 12
- ✅ feasible rate gain at least 90 points
- ✅ final feasible rate at least 98 percent
- ✅ manufacturing failure rate zero
- ✅ design time reduction at least 85 percent
- ✅ cost reduction at least 85 percent
- ✅ new pareto candidates positive
- ✅ synthetic cost avoided positive