ruvector/examples/benchmarks
Claude d8906ed416 feat(agi-contract): multi-dimensional IQ with cost, robustness, and AGI contract
Redefine intelligence measurement as a falsifiable contract with three
equal pillars: graded outcomes (~34%), cost efficiency (~33%), and
robustness under noise (~33%). This addresses the fundamental critique
that accuracy-only IQ saturates at the ceiling.

New modules:
- agi_contract.rs: AGI contract definition (5 core metrics), autonomy
  ladder (5 levels gated by sustained health), viability checklist
- acceptance_test.rs: 10K-task holdout harness with frozen seed,
  multi-dimensional improvement tracking, deterministic replay
- bin/agi_proof_harness.rs: nightly proof runner publishing success
  rate, cost/solve, noise stability, policy compliance, autonomy level

Changes to existing modules:
- intelligence_metrics.rs: Add CostMetrics, RobustnessMetrics as
  first-class dimensions; add noise_tasks, contradictions, rollbacks,
  policy_violations to RawMetrics; rebalance overall_score weights
- superintelligence.rs: Track noise accuracy, contradiction rate,
  rollback correctness, and policy violations across all 5 levels

Contract metrics: solved/cost, noise stability, contradiction rate,
rollback correctness, policy violations (zero tolerance).

https://claude.ai/code/session_01RnwD4x5cbpB7FPvoyYQz8G
2026-02-15 20:43:31 +00:00
..
src feat(agi-contract): multi-dimensional IQ with cost, robustness, and AGI contract 2026-02-15 20:43:31 +00:00
tests style: apply rustfmt across entire codebase 2026-01-28 17:00:26 +00:00
Cargo.toml feat(agi-contract): multi-dimensional IQ with cost, robustness, and AGI contract 2026-02-15 20:43:31 +00:00