ITERATION05: DOES PRE-PROBE RETIMING BEAT AN UNPROBED FULL-BUDGET FIXED PAIR?
October 6, 2026. Numerical, uncalibrated reference-beam experiment.

RATIONALE AND FROZEN EXPERIMENT

Iteration4 added a full-budget unprobed SINGLE. Its fixed-period PAIR still paid
for a probe whose estimate it did not use. Iteration05 adds the direct timing
control: a fixed-period pair with no probe and the entire 10 mJ allocated to its
two moving strikers. This is a fairer test of whether the chosen sensing/retiming
protocol earns back the ideal mechanical energy charged for sensing.

All fifteen iteration4 conditions, including their original seeds, contact
properties, nominal gap-timing errors and estimator outputs are inherited
unchanged. No parameter is optimized or selected after seeing iteration05.
The existing adaptive output is loaded from source-hashed iteration4 records.
No original model, study, estimator, report, or release04 file is edited.

NEW BASELINE: FINITE INITIAL ENERGY, NO STATE RESET

Both protocols have 20 g total moving striker mass and a common 10 mJ ideal
mechanical budget. The new baseline has two 10 g masses, each holding 5 mJ at
t=0, traveling at 1.0 m/s. Contact locations are .9 times reference beam length.
The frozen adaptive pair has two 10 g masses at sqrt(.99) m/s: its 9.9 mJ main
collision budget plus 0.1 mJ prescribed probe energy equals the same 10 mJ.

The new baseline is independently integrated at its actual energy and speed.
It is never obtained by scaling an earlier 9.9 mJ result. Nonlinear contact
dissipation makes such scaling inappropriate. Coarse/fine complete profiles
retain per-source energy, contact storage/loss, positive/negative work, residuals,
and actual first-contact events.

Its second initial gap is:

  gap = initial_speed * (frozen_nominal_period + case_nominal_gap_timing_error).

The adaptive pair instead uses its frozen inferred period and its own speed,
with the SAME additive case timing error. Both strikers already have their
kinetic energy at t=0. There is no later powered reset or hidden energy input.
The code variable flight_time_error_s refers to this NOMINAL gap perturbation.
Actual contact time can differ because the beam moves; source first_contact_s
retains the numerical event. This does not model a real launcher controller.

PROBE, SETTLING AND EXCLUDED DEVICE COSTS

The unprobed baseline uses no probe and no probe-settling wait. Its observation
window is .17 seconds starting with a resting reference beam. The adaptive
protocol retains its ideal prescribed .1 mJ probe, explicit residual/dissipation
accounting, 35-second settling assumption and .17-second main observation.

These modeled intervals are NOT demonstrated device cycle times or throughput.
Electrical power, probe transfer efficiency, sensing, launch, positioning,
handling and reset energy/time remain unknown. The probe is prescribed modal
mechanical energy, not a calibrated electrical input. The target is first-mode
energy in an ideal three-mode cantilever, not a crack path, fracture energy,
stress intensity, fragmentation rate, or diamond/boulder capability.

NUMERICAL AND REPRODUCIBILITY CHECKS

Every new baseline is independently run at 20 and 10 microsecond steps. Its
coarse/fine values are compared with the already refined frozen adaptive result.
Outputs give absolute joule differences, common-10-mJ budget fractions, ratios,
and a timestep sensitivity classification. This screen is not a certified error
bound or physical/statistical significance. Fifteen deterministic conditions,
some with repeated mechanical plants, are not fifteen independent experiments.

Before execution, iteration05 snapshots all imported local source dependencies
and iteration4.py, validates the parent study's captured source hashes, and
checks its summary plus all 15 reports and 15 synthetic CSVs. It snapshots the
same source/data again after execution and rejects drift BEFORE writing an
accepted summary. Each new case report also has a SHA-256 digest. The tests
repeat these checks and verify that no estimator, command, seed or adaptive
profile was recalculated, every baseline has its correct budget/mass/speed,
timing errors are shared, and sensor-only changes cannot affect the unprobed
baseline. Negative cases and unresolved differences are retained.

REPRODUCTION

From custom-poc with Python and NumPy installed, retaining the completed
iteration4 source/report/CSV artifacts:

  python efficiency/iteration05.py
  python -m unittest discover -s efficiency -p test_iteration05.py -v

The script has a 300-second execution limit and no GPU requirement. CPU/runtime
and timestamp metadata vary between executions. Compare numerical outputs and
tolerances rather than requiring identical complete JSON bytes.

Outputs are separate from all earlier releases:

  efficiency/results/ITERATION05_summary.json
  efficiency/results/iteration05/case_*.json
  efficiency/results/iteration05/TEST-report.json
  efficiency/results/iteration05/TEST-log.txt
  efficiency/results/iteration05/AUDIT.json

This module performs no hardware control, network request or publication. A
reviewed public release, if requested, is handled separately from this module.

OBSERVED NUMERICAL OUTCOMES, OCTOBER 6, 2026

All fifteen new baseline profiles completed with both timesteps in 92.507 seconds.
All declared numerical constraints passed. Twelve new tests passed with zero
skips. Source and all 31 inherited data hashes match before/after execution.

The estimated-period protocol beats the unprobed fixed pair in three conditions
and loses in twelve. The relative range is -8.641% to +10.770% peak first-mode
energy. These are magnitudes across selected deterministic conditions, not
performance probabilities or experimental success rates.

At reference, adaptive energy is 3.370081 mJ versus 3.400516 mJ unprobed fixed:
-0.895% relative, -0.030436 mJ, or -0.304 percentage points of the common 10 mJ
budget. At the finer timestep, the relative difference remains -0.892%.
Thus the reference protocol does NOT recover its ideal mechanical probe charge.

The three positive conditions are:

  I4-01: frequency scale .915 only. Adaptive 3.363552 mJ versus fixed
         3.159126 mJ; +6.471% relative, +0.204426 mJ.
  I4-02: frequency scale 1.085 only. Adaptive 3.376525 mJ versus fixed
         3.048221 mJ; +10.770% relative, +0.328304 mJ.
  I4-H-zero: combined H inputs with only nominal gap error removed.
         Adaptive 3.253575 mJ versus fixed 2.950787 mJ;
         +10.261% relative, +0.302788 mJ.

These support a limited conclusion: retiming can avoid enough mistuning loss
to repay this ideal probe charge for some declared conditions. They do not
establish a universal advantage for sensing, a cheapest real device, or the
ability to control or cause fracture. The positive H-zero result also depends
on the counterfactual removal of the specified gap perturbation.

Negative conditions remain explicit: I4-00, I4-03, I4-04, I4-05, I4-06, I4-07,
I4-08, I4-09, I4-10, I4-F-error, I4-F-zero and I4-H-error. The original combined
F and H cases remain worse by 6.843% and 8.641% respectively. Removing the F
gap perturbation still leaves adaptive behind by 1.874%. No post-result tuning
was used to hide or repair these outcomes.
