ITERATION06: STRUCTURAL MODE TRUNCATION WITH FROZEN COMMANDS
October 6, 2026. Numerical research only; no fracture/hardware validation.

QUESTION AND PRESPECIFIED DESIGN

Do the iteration05 energy comparisons survive retaining more bending modes in
the same ideal reference cantilever? This is the structural branch of the
previous NEXT-EXPERIMENT-PLAN. Contact-law changes are NOT part of this run.

Six declared cases: I4-00, I4-01, I4-02, I4-F-error, I4-H-error, I4-H-zero.
Three mode counts: 3, 5, 7. Two fixed protocols: adaptive pre-probe pair and unprobed
fixed-period pair. Two common timesteps: 5 and 2.5 microseconds. This is exactly
72 matrix integration rows. The two protocols are vectorized in a shared
integration call but have independent states, energies, speeds and source work.

The estimator, timestamps, inferred delay and fixed delay remain frozen from
the earlier source-hashed study. No parameters are optimized. The adaptive
protocol retains .1 mJ assigned ideal mechanical probe energy and 9.9 mJ main energy;
the unprobed pair has 10 mJ main energy. Both have 20 g total moving mass. The shared
nominal gap-timing perturbation is converted to each protocol's own initial
gap using its own speed. Actual contact remains a separate simulated event.

FIELD INTERPRETATION: each detail includes frozen_command and frozen_estimator
copied as inherited ADAPTIVE CONTEXT, including in unprobed_fixed rows. Those
two fields are NOT the unprobed baseline's applied command or inputs. The actual
command for either protocol is result.candidate.second_arrival_delay_s; its
actual starting gap is result.initial_second_gap_m. For unprobed_fixed, the
delay is NOMINAL_PERIOD_S plus case.plant.flight_time_error_s, and speed is 1 m/s.
The baseline uses no estimate or probe. FIELD-INTERPRETATION.json documents this
distinction without rewriting the frozen detail files; post-run tests verify
the fixed delay and gap for every unprobed row.

The old probe is explicitly the prescribed THREE-MODE probe with its 35-second
settling and bounded residual accounting. Added modes receive no assigned probe
energy. This isolates main-event plant truncation; it is not a new physical
probe calibration or validation of sensing at higher mode count.

ANALYTIC STRUCTURAL BASIS

Cantilever eigenvalues solve cos(beta)+sech(beta)=0 using bracketed bisection
around (n-.5)*pi. Frequencies are beta^2*sqrt(EI/(rho*A*L^4))/(2*pi), with the
same declared frequency scale. Standard bending shapes have integral(phi^2)=L;
division by sqrt(rho*A*L) gives unit generalized modal mass. A stable exponential
expression replaces subtraction of large hyperbolic terms at higher modes.
Mass-weighted orthogonality/normalization are checked with 128- and 256-point
Gauss-Legendre quadrature. Unit tests independently check clamped displacement/
slope and free-end moment/shear residuals.

All retained modes use the same prescribed modal damping ratio. This does not
validate that assumption for a real specimen. The Euler-Bernoulli theory also
omits shear/rotary inertia and plate behavior of the 120 mm-wide strip. Point
modal loading, the support, contact stiffness and loss remain hypothetical.

FINITE ENERGY, CONTACT AND INTEGRATION

The passive unilateral contact law is unchanged:
  force = k*d + a*d*max(d_dot,0), when indentation d>0; zero otherwise.
  elastic storage = .5*k*d^2.
  dissipation rate = a*d*max(d_dot,0)^2.
Track every mode's energy, every striker's kinetic/contact energy and loss,
positive/negative source work, structural damping and total residual. No energy
is added at delayed impact or through a state reset. Extra-mode energy is
reported both per-mode peak and concurrent peak sum above mode3; individual
mode peaks need not occur simultaneously and must not be summed as one event.

The highest retained frequency over the six cases is about 3.50 kHz. The common
5 microsecond step and 2.5 microsecond refinement satisfy the original minimum
30 samples/period screen even for that mode. Requested and actual timesteps and
step counts are recorded. This frequency rule is a screen, not an error bound.
Each profile also has energy/domain checks and factor-two refinement.
The inherited displacement-domain check samples eleven beam locations. Its
saved maximum is explicitly a SAMPLED displacement, not a guaranteed maximum
over the entire continuous beam. This limitation matters when adding modes.

Before the matrix, the expanded three-mode integrator must reproduce the
original separate integrator at 20 and 10 microseconds on both reference protocols
within 1e-8 relative target energy and 1e-10 J final energy. Full outputs from
that four-comparison check are saved separately from the 72 matrix rows.

PRESPECIFIED TRUNCATION SCREEN

Use fine-step target energy: each protocol's 5-to-7-mode change must be <=2%.
A resolved comparison sign flip between 5 and 7 modes fails the screen. Unresolved
signs or failed numerical/refinement checks yield an inconclusive/failed screen,
not a positive convergence claim. This is a working diagnostic inside the
chosen theory; it cannot establish full-elasticity convergence or physical
accuracy. All six cases remain in the output regardless of outcome.

DURABLE EVIDENCE AND EXACT PROGRESS

New outputs are only in iteration06 files:
  results/iteration06/PROGRESS.json         atomic timestamped snapshot
  results/iteration06/I6_*.json             one completed integration row each
  results/iteration06/THREE_MODE_VALIDATION.json
  results/ITERATION06_summary.json          final complete OR partial summary
  results/iteration06/TEST-report.json and TEST-log.txt
  results/iteration06/AUDIT.json
  results/iteration06/FIELD-INTERPRETATION.json

Each snapshot explicitly reports planned/completed/running/failed/not-run counts.
A completed row with failed numerical constraints is separately counted; it is
not hidden and does not equal passed convergence. Source/input SHA-256 digests
are captured before and after. Each completed detail is hashed. The 20 minute
deadline or a cancellation preserves all completed files and marks unrun rows;
no incomplete study is reported as complete. Snapshot timestamps describe saved
progress, not a promise of a live website connection.

REPRODUCTION

From custom-poc with Python and NumPy plus the unchanged iteration04/05 source
and data artifacts:
  python efficiency/iteration06.py
  python -m unittest discover -s efficiency -p test_iteration06.py -v

Only the main-event initial mechanical budget is controlled here. Electrical,
probe transfer, sensors, launching, positioning, handling and reset costs are
unknown. No software row demonstrates rock/diamond fracture, a controlled split,
high-power acoustic capability, a complete product, or real throughput.
This script does not control hardware, access accounts, or publish externally.

OBSERVED NUMERICAL OUTCOMES, OCTOBER 6, 2026

The bounded study completed all 72 planned integration rows in 469.416 seconds,
with zero failed rows, zero not-run rows and zero failed numerical constraints.
All 14 final tests passed with zero skips. All six source hashes, 47 inherited
input hashes and 72 completed-row hashes passed the post-run integrity check.
The final PROGRESS snapshot and final summary have the same SHA-256 digest.

All six fixed 5-to-7-mode screens passed. Fine-step target-energy changes across
the two protocols range from 0.0428% to 0.0617%, below the declared 2% screen.
No declared comparison changes its resolved sign between five and seven modes.
The maximum relative target-energy timestep change is 3.24e-6; the maximum
relative energy-balance residual across all matrix rows is 2.18e-6.

At SEVEN modes and the FINE timestep, adaptive minus full-budget unprobed fixed:

  I4-00       -0.8923% relative; -0.030374 mJ target energy.
  I4-01       +6.4599% relative; +0.204335 mJ target energy.
  I4-02      +10.7743% relative; +0.328692 mJ target energy.
  I4-F-error  -6.8420% relative; -0.233060 mJ target energy.
  I4-H-error  -8.6365% relative; -0.288572 mJ target energy.
  I4-H-zero  +10.2612% relative; +0.303088 mJ target energy.

This retains the conditional result: pre-probe retiming gains energy in the
declared mistuned/counterfactual cases and loses in the reference and original
combined F/H cases. Adding these modes does not turn it into a universal
advantage. This screening result is confined to the uncalibrated beam/contact
theory and frozen commands. Contact-law uncertainty and physical specimen
calibration remain separate, untested requirements.

The metadata companion verifies all 36 unprobed fixed rows use the correct
nominal fixed delay plus perturbation and their own 1 m/s initial gap. The
inherited adaptive frozen_command/frozen_estimator context fields were retained
without rewriting original reports; they must not label the baseline's input.

Saved detail contains inputs, per-mode/total peaks, actual contact-event times,
and complete final source/structural energy and work ledgers. Full state/force
time-series waveforms were not retained. A reconstruction or animation must
be labeled as a new simulation, not displayed as an original recorded waveform.
