Phase 30 · matched progress

Same progress, less damage?

AT-0231 → AT-0238. A fixed-coupling nonlinear baseline and a simple state-dependent gate are forced to travel the same relational distance, then to spend the same action. Only then do we compare identity damage — and we retire the progress metric that let overshoot count as arrival.

Core objective, unchanged

The target is unchanged: substantial genuine relational-location change while preserving identity and invariants, under locality, bounded propagation, finite action, no hidden debt, robustness, scale persistence and nontriviality. A mechanism that looks better only because it overshoots the target, spends more action, or is reproduced by a two-line gate has advanced nothing.

PRIOR INTERNAL RESULT, ENCODED HERE AS A RERUNNABLE BENCHMARK — NOT PHYSICS. Seed 223. A matched-progress comparison of a fixed-coupling nonlinear baseline against simple state-dependent adaptive gating, over 500 matched cases with substantial progress and a progress difference below 0.025, reported median progress 1.0786 in the toy normalised coordinate, median final identity disturbance 0.4537 adaptive against 0.6951 baseline, median advantage 0.2430, adaptive better in 100% of that sampled set, and a median action ratio of 1.0029. This is computational evidence that the chosen gating rule reshapes the tradeoff at matched progress and matched action. It is not physical evidence of anything. The normalised progress exceeds 1 because the toy address coordinate overshoots its target, which is why this page reports destination error alongside it.

Plain language

Last time the protective rule looked good only because it moved less. So this time we force both mechanisms to travel the same distance, and then to spend the same effort, and only then ask which one damaged the object less. We also stopped trusting the distance number: the toy address can sail past its target, which made the old score read above 100%. Now we report how far off the destination each run finished, and no mechanism is paid for overshooting.

Controls

SeedSeed 223 reproduces the recorded benchmark ensemble.Candidate systems — 240Cases that cannot be matched inside tolerance are discarded, not scored.Progress tolerance — 0.025Maximum permitted difference in relational progress between the two mechanisms.Baseline effort — 0.90Only the actuator is tuned in matched comparisons; no mechanism law is retuned.

Running the matched-progress package — bisecting actuator effort per system, then re-running every audit. This takes a few seconds.