Subtractive Optimization

The Forgetting Engine

An experimental search method built around strategic elimination: remove low-value candidates, preserve selected alternatives, and continue exploring from a smaller but more informative search space.

How the method is tested

The implementation varies by domain, but the research loop follows the same general pattern.

Step 1

Generate

Create or receive a population of candidate solutions.

Step 2

Evaluate

Score candidates against the stated objective and constraints.

Step 3

Forget selectively

Remove low-value candidates while retaining selected alternatives and contradictory signals.

Step 4

Repair and explore

Use the surviving structure to propose new candidates and continue the search.

Reported benchmark results

30,800 controlled trials in the locked optimization sweep

4,000-trial 3D protein-folding comparison: 25.8% success versus 3.9% for the stated Monte Carlo baseline

Vehicle-routing comparisons reported up to 89.3% improvement at the largest tested scale

Quantum-compilation experiments reported 27.8% gate reduction and 3.7% fidelity gain under the documented simulator setup

Interpretation limits

Results are benchmark-specific and use different objectives and baselines.

Internal validation does not equal independent replication or peer review.

The observed scaling pattern is a research hypothesis, not a universal complexity law.

Performance on a benchmark does not establish effectiveness in clinical, financial, or production deployment.

Search, not memory

“Forgetting” refers to candidate elimination inside an optimization process. It is not a claim about human awareness or machine consciousness.

Baseline matters

Every reported percentage depends on the comparison algorithm, objective, scale, budget, and success definition.

Open hypothesis

Larger relative gaps at larger tested scales are promising. They still require independent and stronger-baseline replication.

Review the benchmark record, not a universal claim.

The Evidence page separates the locked sweep, domain studies, reported comparisons, and remaining limitations.

Review Forgetting Engine Evidence