# Source-parameter audit: pulses to plasticity
20 October 2026 • retrospective computational analysis • no external review

## What is new
The supporting information to Rivera-Sierra & Bisquert (2025), DOI
10.1063/5.0289570, was retrieved from Figshare collection
10.60893/figshare.aml.c.8063077. Table S1 supplies four voltage-specific
calibrations; they were not supplied in the prior NetPhysics scalar check.
The deposited MATLAB script still references missing fitted .mat files and
nonportable local paths. We did not execute it or silently substitute a new fit.

The table's gL/gH values are in microSiemens, voltage parameters in volts.
Its K=64.1 ms V^gamma, gamma=1.85, and t0=0.10 ms give
xeq(u)=1/[1+exp(-(u-VB)/Vm)] and
 tau(u)=K [dxeq(u)/du]^gamma+t0.
Separate calibrations at 2, 3, 4 and 5 V are explicitly required by the source.
There is no Table S1 calibration for 1 V; its measurements remain in the extraction
but no 1-V prediction is invented.

## Defined calculation
Use ideal 2-ms pulses at the named voltage and 2-ms zero-voltage rests,
starting at the zero-voltage equilibrium. Assume internal voltage u equals
applied voltage, and omit capacitive current. In this approximation
I=[gL+(gH-gL)x]u. This is not a full circuit integration.
The full circuit uses V=Rs I+u and includes Cm du/dt. A short Rs Cm alone
is insufficient to justify u approximately V: Rs G must also be small.

Let a=exp(-ton/tau_on), b=exp(-toff/tau_off).
aF=a(1-b)/(1-ab); an=aF+(a-aF)(ab)^(n-1), where n=1 is the first pulse.
xn=x_on+(x_off-x_on)an. This is the source recurrence with pulse counting
shifted by one relative to the paper's a0. Its fixed point is not a new NetPhysics
law. State gain xF/x_off is not generally current gain when gL is nonzero.

The analytic result agrees with independent numerical integration of the same
scalar ODE to below 2e-12 in state, across all four Table S1 cases. This verifies
implementation of the stated equations; it does not verify their physical truth.

## Comparison with deposited Figure 4(c) observations
The bundled Fig_4c.xlsx has 1,330 valid rows across five voltage traces.
Original time and voltage values are retained; source mA is converted to microA.
Each trace has 11 high-voltage intervals. The declared descriptive extraction
uses voltage >90% of nominal amplitude, then the last high sample of each
interval. Source row numbers are preserved. All valid rows remain available.

Predicted ideal end of pulse 11 / last sampled high current (microA):
2 V: 11.349226 / 13.014281
3 V: 30.064915 / 26.757527
4 V: 47.777491 / 43.738939
5 V: 64.852446 / 62.564805

Relative differences are -12.79%, +12.36%, +9.23% and +3.66% respectively.
The plotted comparison also shows all 11 pulse endpoints, including early-pulse
mismatch. A late endpoint alone can hide transient disagreement.
These are post-inspection descriptive choices, not a preregistered test or an
untouched holdout. Last sampled high points occur before ideal end times and
are not exact end-of-pulse measurements. Instrument timing, uncertainties,
finite rise time, capacitive current and internal voltage remain unresolved.
The four voltage traces are not four independent laboratories, and successive
pulses are correlated. No uncertainty interval, fit quality pass, or replication
success is inferred. Last-five medians are descriptive only.

## Figure 3: unresolved time and parameter discrepancy
The published caption specifies tau_off=0.97 ms and tau_on=0.98 s, with pulse/rest
durations 0.1/0.05 s and 0.05/0.2 s, while the panels label the time axis in ms.
It also identifies the model as Figure 1's. Figure 1's inverse-rate expression,
with tau_M=tau_0=1 ms and positive exponential terms, forces tau(u)<=1 ms;
therefore 0.98 s cannot follow from that parameter set.

output/figure3-unit-audit.csv reports both the literal Figure 3 caption
calculation and the alternative obtained from Figure 1's functions at Figure 3's
voltages and durations. Neither is labeled the corrected figure. The difference
is an unresolved source specification issue requiring author clarification.
Cached notebook outputs are not a substitute for an unambiguous execution order.
No claim that the complete paper has failed replication is made.

## Next requirements
Clarify Figure 3 units/parameter overrides, obtain the missing fitted objects or
reproduce their fitting pipeline, resolve the measured voltage/current observation
model, and freeze one numbered figure with a quantitative acceptance rule.
Only then assess suitability for a replication manuscript. Author clarification,
independent code execution and peer review are separate activities, none completed
by this author-run audit. ReScience's own-work/close-collaborator restriction and
negative-replication author-contact requirement must be respected.