An IV curve or run record does not match the expected model

Check that the reference equation matches the model

The constant-resistance overdamped RSJ formula requires the corresponding resistance law, zero capacitance, no thermal-noise drive and appropriate averaging. RTYPE=1 retains a gap-transition branch even if R0 equals RN; that equality alone does not establish the simple analytical model. Changing the number of bias samples cannot remove a physical-model difference.

An initial teaching comparison used RTYPE=1 with R0 = RN and failed at a bias sampled by the finer grid. That was an invalid constant-resistance reference assumption, not evidence of a general solver defect. Correcting the assumption exposed the separate RTYPE=0 gap-dependence comparison below; the original failed runs remain retained.

Recognise the unresolved gap comparison — KB-IV-001

On the exercised Windows JoSIM 2.9 candidate, the RTYPE=0, CAP=0, IC=100 µA and RN=R0=10 Ω input with VG=2.8 mV completes its 40-interval IV run but differs from the constant-resistance analytical reference by up to approximately 0.301 mV in the checked high-bias region. It fails the same stated acceptance check used for the numerical control.

At 370 µA, separately authored transient runs with 50 fs and 12.5 fs steps remain approximately 6.70% and 5.48% away from that reference. Smaller steps alone therefore did not resolve the observed difference. The otherwise identical artificial VG=100 mV controls agree within approximately 0.316% and 0.0265%, respectively. These observations identify a relevant gap-region dependency; they do not by themselves settle intended semantics or a root-cause repair.

Source review finds gap-region selection in JJ::getValue and RTYPE-dependent stamping paths, which motivates implementation review. No product correction is claimed. Keep iv_gap_check.cir as a diagnostic and retain the comparison as unresolved. The high-gap control is not a physical workaround for a real junction; do not substitute its gap value into a calibrated model.

Read IV-specific metadata — KB-IV-002

The native IV run record correctly identifies model, bias range, point count and voltage/current units in iv_characterization. However, its general effective_settings fields in this checked run report a 2.5 ps step and 100 ps stop, while IV::setup_iv uses 50 fs and 1 ns for the internal helper. Treat those general timing fields as enclosing-job metadata, not as a record of the actual IV helper timing.

The distinction is source-backed and observed in the current native record. The helper controls are not exposed as additional .IV arguments in this guide. For an explicit time-step or settling study, write a transient circuit with the intended source/model and retain its actual time grid; changing an unrelated general setting or increasing N is insufficient proof of helper refinement.

Use a repeatable diagnosis

1. Preserve the complete model, .IV statement, current raw CSV, process/task verdict and exact runtime identity. Check that the x coordinate is voltage and the y values are current.
2. Confirm 4N+1 rows and bias order. Repeated extrema and a nonzero final interval are expected for this implementation.
3. State the physical assumptions behind any reference equation and the numerical acceptance criterion. Distinguish a failed assumption from an unresolved model/implementation comparison.
4. Use explicit transient experiments for time-step and averaging checks. Retain both matching controls and failing diagnostics; do not label an artificial control as a repaired real-device result.
5. Before sharing, review logs and run records for source paths and private design/licence-status information. No real customer data is needed to reproduce these teaching cases.