BENCHMARKS / NO INVENTED NUMBERS

Efficiency means nothing if the result is wrong.

Every comparison begins with the same correctness gate, exact fixtures, repetitions, observable artifacts, confounders, and explicit unavailable data.

From the original research notebook. Earlier concepts are preserved here as history. See the current theory map and dispositions →

CORRECTNESS→MUTATION→CONTEXT→TOKENS / COST / TIME

CROSS-PROJECT EXPERIMENTS

01

Semantic edit → trajectory profiler

Does bounded semantic editing reduce mutation amplification without lowering correctness?

UNMEASURED
02

Context firewall → decision evidence

How much context can deterministic reduction remove before correctness drops?

UNMEASURED
03

Verifiable handoff → fresh review

Does source destruction reduce correlated verification errors?

UNMEASURED