PROPOSED NEXT EXPERIMENT - NOT IMPLEMENTED OR EXECUTED
October 6, 2026. Prepared after the frozen iteration05 study.

Priority: test sensitivity to structural-mode truncation and the hypothetical
contact law before spending more effort tuning the frequency estimator.

Reason: iteration05 establishes only conditional benefits. A better estimator
cannot establish physical relevance if those benefits disappear when omitted
beam modes or another passive contact model are included. This proposal is a
new diagnostic experiment, not an outcome-selected repair to iteration05.

1. Freeze six declared conditions before execution:
   I4-00, I4-01, I4-02, I4-F-error, I4-H-error and I4-H-zero.
   These cover reference, the two frequency-only gains, the two original
   combined failures, and the positive H counterfactual. They do not estimate
   a success probability and do not replace the full fifteen-case evidence.

2. Keep two protocols and their commands fixed:
   existing pre-probe adaptive pair (9.9 mJ main plus .1 mJ charged probe) and
   unprobed fixed pair (10 mJ main), each with 20 g total moving mass. Import
   existing measured/estimated timestamps, delay commands and gap errors;
   do not rerun or improve the estimator while testing structural truncation.

3. Structural truncation branch:
   derive mass-normalized cantilever modes analytically for 3, 5 and 7 modes,
   keeping physical beam dimensions, material parameters, first-mode damping,
   contact positions and current passive contact law unchanged. Run paired
   protocols in each condition. Use timesteps meeting the highest-mode
   resolution rule and refine each by a factor of two; compare at a common
   sufficiently fine timestep where feasible. Retain every omitted-mode
   energy contribution and the existing per-striker/total energy ledgers.

   Predeclare a diagnostic failure if the 5-to-7-mode change exceeds 2% of a
   protocol's target energy, or changes a comparison sign beyond the timestep
   screen. This is a working convergence screen, not a certified error bound.
   A failure requires reporting inconclusive truncation, not picking whichever
   mode count produces the preferred result.

4. Only after the truncation branch, test contact-law uncertainty separately:
   compare the current force k*d + a*d*max(d_dot,0) with a passive Hertz-type
   form kH*d^(3/2) + aH*d^(3/2)*max(d_dot,0), for positive indentation only.
   Its stored elastic energy must be (2/5)*kH*d^(5/2); contact-loss rate must
   be aH*d^(3/2)*max(d_dot,0)^2. Preserve nonnegative dissipation and explicit
   per-source work. Match force and damping at fixed declared reference
   indentations (.1 mm and .5 mm) using kH=k/sqrt(d_ref), aH=a/sqrt(d_ref).
   These are hypothetical law comparisons, not calibrated rock/tip properties.
   Matching force at one indentation does not match tangent stiffness or prove
   either model correct. Do not fit the references to improve outcomes.

5. Output the absolute target energies, common-budget deltas, ratios, energy
   in added modes and numerical/contact-domain violations for every run.
   Preserve changes of sign and negative outcomes. Record all source/input
   hashes before and after. Use an explicit bounded execution deadline and
   retain partial/failed diagnostics if it expires; never report completion
   or convergence from unfinished runs.

6. Decision after both branches:
   if advantages are materially sensitive to mode count or contact law, the
   next evidence need is contact/impedance measurement and specimen calibration.
   If they are stable within declared screens, independently test the sensing
   rule on a fresh fixed uncertainty matrix. Neither outcome authorizes a
   fracture claim or substitutes for experiment on a physical specimen.

No iteration06 source, simulation, hardware action or publication is performed
by this plan. Root owns the next implementation and publication decision.
