Experiments

The fee ideas on the test bench right now, and how each one did against the current fee.

Each candidate is replayed on past swaps, run in shadow (worked out, never charged) and attacked by a red team. Only then may it charge real trades in a live test, within 1 point of the current fee. Rejected ones stay here with the reason.

On the bench

3 benches · uncalibrated simulation (replay + assumed elasticity)

Bench 1: adaptive-fee plate A

sim #76

rejectedE-0076

project ETH/day (ε=1): −0.32%
swap retention (ε=1): +2.48%
fee at cap: +0%

base 2.00%, multiplier 0.20×, cap 2.80% · gates failed: project ETH/day at ε=1 8.336719 < baseline 8.363663

Bench 2: adaptive-fee plate A

sim #77

rejectedE-0077

project ETH/day (ε=1): +1.38%
swap retention (ε=1): −9.87%
fee at cap: +0.46%

base 2.20%, multiplier 0.40×, cap 3.20% · the daily maximum of 12 routine red-team runs is reached

Bench 3: adaptive-fee plate A

sim #78

rejectedE-0078

project ETH/day (ε=1): +0.25%
swap retention (ε=1): −1.68%
fee at cap: +0%

base 2.10%, multiplier 0.20×, cap 3.00% · the daily maximum of 12 routine red-team runs is reached

The path every change takes

current stage · simulate

Research
Simulate
Red team
Canary
Measure
Rollback
Memory

A module only moves right. Any kill condition sends it to rollback, and every outcome, kept or reverted, is written to memory.