• Resolution (counts / digits): the smallest step the display can show. Real resolution is limited by the noise floor and settling, not only by ADC bit depth.
• Noise (short-term scatter): how much the reading wanders over seconds/minutes. Increasing integration time (NPLC/aperture) reduces noise but does not fix nonlinearity, thermal EMF, or long-term drift.
• Accuracy (uncertainty vs true value): the bounded error after calibration. It is an error budget that includes gain/offset, linearity, temperature effects, drift/aging, and switching/terminal artifacts.

• Accuracy spec format matters: “ppm of reading + ppm of range (or counts)” reveals whether error grows with measured value (gain/linearity) or is dominated by fixed offsets and switching/terminals.
• Temperature coefficients: separate the reference/divider tempco (systematic) from noise (random). A stable temperature does not guarantee low error if thermal gradients exist near terminals.
• Linearity and range switching: high digit count is meaningless if divider/shunt linearity and relay behavior introduce range-dependent steps or hysteresis.
• Time vs confidence: NPLC/aperture improves mains rejection and noise, but measurement throughput is limited by switching + settling + integration time (not “ADC speed”).

• Any chain of dissimilar metals (binding post → screw → lug → solder → copper) forms thermocouple junctions. When a temperature gradient exists, µV-level EMF appears and looks like a real signal.
• The most damaging gradients are dynamic: airflow, hand warmth, nearby heat sources, or uneven internal heating. A stable ambient temperature does not guarantee a stable terminal gradient.
• Best practice is to minimize junction count, use low-thermal materials in the critical path, and keep terminal-related heat flow symmetrical so gradients cancel rather than accumulate.

• MOV/TVS/PTC/series resistors/fuses protect against overloads, but may introduce finite leakage, parasitic capacitance, and temperature-dependent behavior that shifts the effective input conditions.
• Leakage that is irrelevant at low impedance can dominate high-impedance measurements (and can create slow “creep” effects after range switching).
• The terminal protection layout must keep high-leakage parts away from ultra-high-impedance nodes, and avoid thermally coupling hot components into terminal junctions.

• Flux residue, fingerprints, and humidity can create surface conduction that behaves like a hidden parallel resistor. The symptom is often slow drift or reading instability rather than obvious noise.
• Guard rings and driven-guard techniques reduce effective leakage by holding nearby surfaces at a similar potential, preventing current from flowing into the sensitive input node.
• Physical separation, clean dielectric surfaces, and controlled creepage/clearance near terminals are “free digits” because they reduce uncorrectable errors before any calibration is applied.

• Compensation is only meaningful if the sensor placement tracks the temperature of the junctions creating EMF. A sensor on the chassis can miss the terminal gradient entirely.
• The target is repeatability: make the thermal environment predictable (mechanical fixation, insulation where needed, reduced airflow sensitivity), then apply compensation to reduce residual drift.
• Over-aggressive compensation can inject noise if the sensed temperature is influenced by airflow or intermittent contact; the measurement must be thermally “quiet” first.

• Contact resistance stability: not “low once,” but stable across switching cycles, vibration, and temperature. Unstable contacts show up as random last-digit steps that correlate with switching history.
• Thermal EMF (low-thermal behavior): dissimilar-metal junctions inside relays and interconnects create µV-level EMF when thermal gradients change after switching.
• Insulation leakage: relay leakage and surface leakage form hidden parallel paths, especially damaging for high-impedance modes and after high-voltage channels.
• Lifetime and drift: the metric that matters is the stability of path behavior after many cycles, not just the rated number of operations.

• Minimize thermocouple junctions: reduce dissimilar-metal transitions in the sensitive path; keep materials and connector interfaces consistent where possible.
• Thermal symmetry: route HI/LO and adjacent sensitive nodes with mirrored geometry so thermal gradients tend to cancel instead of accumulate into a net EMF.
• Heat-source isolation: keep regulators, protection hot-spots, and high-dissipation shunts physically and thermally separated from the relay matrix region.
• Guard-aware spacing: reserve space for guard rings/driven guard around high-impedance nodes so leakage paths are controlled rather than accidental.

• Channel-to-channel memory: switching from a high-voltage/high-charge channel to a low-level channel can cause transient bias through parasitic capacitance and surface charge, requiring a defined settling window.
• Settling time is sequence-dependent: it depends on the previous channel, the new range, input source impedance, and the selected bandwidth/integration settings.
• Path self-test is mandatory for repeatability: include loopbacks or known references to detect stuck relays, elevated leakage, contact instability, and range-path discontinuities early.

• Linearity first: voltage coefficient effects, resistor network gradients, and stress distribution can create range-dependent deviation that cannot be “averaged away.”
• Segmentation reduces stress: spreading voltage and power across sections helps keep each element in a more controlled operating region, improving predictability.
• Thermal coupling matters: self-heating in the divider and nearby components can shift ratios through local temperature gradients, not only datasheet tempco numbers.

• Different DCV ranges select different divider sections and switching paths, changing what the DUT “sees” at the input. High source impedance DUTs are most sensitive to this.
• Protection networks can present voltage-dependent leakage and capacitance, which alters effective input conditions even if the nominal “10 MΩ” looks unchanged in a simplified spec line.
• Filtering and buffering nodes define where the impedance boundary is measured; moving that boundary changes both loading and settling behavior after switching.

• Bandwidth limiting suppresses wideband noise and reduces sensitivity to interference, helping the integrating/ΣΔ conversion produce stable readings at practical NPLC/aperture settings.
• The tradeoff is settling: after range or channel changes, RC nodes and filter states must settle before the measurement uncertainty reaches its intended bound.
• Overload recovery and input step response must be controlled; otherwise “fast update rate” simply reports unsettled data more quickly.

• Settling is the sum of: switching transient → divider/buffer recovery → filter/RC charge movement → conversion window (NPLC/aperture) + any digital filtering latency.
• A meaningful spec is not “reads per second,” but “time to reach a defined error threshold” after a step or range change.
• Verification method: apply a known DC step and record reading vs time; define the settle criterion in ppm or counts rather than subjective stability.

• Burden is the total series drop inside the meter, not only the shunt. It includes shunt resistance, fuse and protection elements, relay contacts, and interconnect resistance.
• A practical view is: burden = current × series resistance. When current is high, even small added resistance can create a non-negligible voltage drop and disturb the DUT.
• A stable specification is “time to a trustworthy reading” after switching or overload—not just a fast update rate. Self-heating and protection recovery can shift the last digits for seconds to minutes.

• Range-shunt selection is driven by both target burden and resolution. Small shunts reduce burden at high current, while larger effective shunts (or higher-gain sense paths) improve low-current resolution.
• Switching path repeatability matters as much as shunt value. Relay contact changes and path-dependent offsets can introduce range-to-range discontinuities that are difficult to calibrate away if they vary with temperature or usage.
• Self-heating is an accuracy term: shunt power dissipation creates temperature rise and resistance change. The resulting drift can look like “slow settling” after a current step or after spending time in a high-current range.

• Fuses and protection elements enforce safe energy limits, but they can add series resistance, temperature rise, and contact variability—each contributing to burden and drift.
• After high-current stress or overload events, internal temperature gradients can temporarily increase offset and change the effective burden. A precision workflow therefore includes a recovery window before trusting final digits.
• The best measurement practice is to define an allowed burden ceiling for the DUT and choose the meter range and method accordingly, rather than forcing a single range for all conditions.

• 2-wire (2W) includes lead and contact resistance in the measured result. If lead/contact resistance is a meaningful fraction of the target resistance (or its uncertainty budget), 2W is not sufficient.
• 4-wire (4W Kelvin) separates force and sense paths so the measured voltage is taken directly at the DUT terminals, largely removing lead and contact drops from the computed resistance.
• The practical decision is based on error budget, not a single resistance threshold: use 4W when clamp quality, lead length, or contact variability can move the last digits.

• A stable current source (or controlled stimulus) sets the measurement scale; the voltage across the DUT is then digitized by the same precision front-end used for DCV.
• Current reversal / chopping is used to cancel fixed offsets such as thermal EMF and amplifier offset by measuring with both polarities and combining results.
• Synchronous sampling and integration help reject mains interference while preserving a predictable settling and uncertainty model.

• Low resistance: the DUT voltage can be in the µV–mV range, so thermal EMF from junctions and gradients can look like real signal. Kelvin connections and polarity reversal reduce this error source.
• High resistance: tiny leakage currents through relay insulation, PCB surfaces, humidity, or contamination can act as a hidden parallel resistance. Guarding and clean insulation geometry keep leakage from dominating.
• Measurement stability often depends on environment control: airflow and touch change thermal gradients; humidity changes surface leakage behavior.

• NPLC — integration window length. Integer NPLC naturally suppresses 50/60 Hz because a full-cycle average cancels the sine.
• Aperture — how long the input is effectively observed/averaged for a single reading (integration or filter equivalent window).
• Reading rate — how often results are reported. Faster reporting does not guarantee the data has been averaged enough to be stable.

• The converter integrates the input over a defined window and then de-integrates (run-down) against a reference. The result is proportional to the time-averaged input.
• When the integration window equals an integer number of power-line cycles, the average of 50/60 Hz interference tends toward zero, producing strong mains rejection.
• Increasing NPLC reduces noise and improves repeatability, but it also increases measurement latency and step settling time.

• The modulator oversamples and shapes quantization noise toward high frequency. The decimation filter removes that out-of-band noise and produces a lower-rate, higher-resolution output.
• Higher OSR and stronger filtering usually reduce noise, but the filter has a longer effective window and more latency, reducing the output update rate.
• “More stable” typically means a longer effective aperture (filter window). This is why a meter can be stable yet slow.

• Need mains suppression: use integer NPLC (integrating) or a ΣΔ mode with strong 50/60 Hz attenuation.
• Need more stability: increase the averaging window (NPLC↑ / OSR↑ / stronger filter) and allow more settling time.
• Need more speed: shorten the window (NPLC↓ / OSR↓ / lighter filter) and accept higher jitter in the last digits.

• Tempco (temperature coefficient) sets how the scale moves with ambient and internal heating.
• Aging sets long-term drift. A very low-noise reading is not “precision” if the scale slowly walks away.
• Thermal environment can defeat a good reference: local gradients and airflow change nearby junction temperatures and offsets.

• Offset — estimated from internal short/zero conditions and applied as a subtraction.
• Gain — estimated by injecting known reference points and scaling the measurement chain.
• Part of temperature drift — corrected when temperature is measured and the system remains thermally repeatable.

• Nonlinearity — not captured by a single gain/offset update when the transfer curve shape changes with range or network state.
• Thermal gradients — create offsets (thermal EMF) that vary with airflow, touch, load history, and local heating.
• Relay thermal EMF / contact variability — can shift after switching or with temperature; self-cal captures “now” but cannot guarantee the next state matches.

• At power-up / after warm-up: establish coefficients once the internal temperature is stable.
• On temperature change: trigger when the internal sensor indicates a meaningful shift in thermal state.
• On a timer: re-normalize for long runs, balancing drift risk against downtime for calibration cycles.
• Before critical measurements: allow a controlled interruption to protect the uncertainty budget.

• Short-term: reading noise, residual mains pickup, quantization/filter residue, and immediate post-switch transients.
• Long-term: reference drift/aging, tempco-related scale changes, and slow thermal equilibrium shifts.
• Modelable terms: offset/gain-like behavior that can be corrected when the internal state is repeatable.
• State-dependent terms: thermal gradients (thermal EMF), relay contact variability, and leakage that depends on humidity and contamination.

• NPLC / stronger filtering: reduces mains residue and random noise (tradeoff: slower response and lower throughput).
• More repeats / averaging: reduces random noise but does not remove drift, thermal EMF, leakage, or settling bias.
• Range selection: changes divider/stimulus/sense paths and can shift which term dominates (linearity, noise, or settling).
• Thermal discipline: reduces thermal gradients and slow drift (tradeoff: time and environmental control).

• Estimate random terms from repeated readings under fixed settings (standard deviation and stability vs time).
• Estimate state-dependent terms with controlled A/B tests: change channel order, change range, change NPLC, and observe whether the mean shifts.
• Treat a measurement as “done” when it reaches the target uncertainty threshold, not when the display updates quickly.

• Switch time: relay movement and contact stabilization, plus switching transients.
• Analog settle: RC nodes, input protection, and bandwidth-limit networks reaching a new equilibrium.
• Window completion: NPLC integration or filter equivalent aperture required for stable digits.

• Charge injection / residue: switching from high level to low level can leave a temporary offset on sensitive nodes.
• Ohms stimulus residue: excitation and switching can leave transient charge and require extra recovery time.
• Leakage state: high-impedance measurements can be distorted by surface condition and previous channel history.

• Voltage: high → low to avoid lifting low-level channels with residue.
• Resistance: low → high to protect the most leakage-sensitive channels from prior stress.
• Group “sensitive” channels and give them longer settle + larger NPLC as a controlled policy.

T_per_read ≈ T_switch + T_settle + T_window(NPLC/filter) + T_compute + T_interface
Throughput ≈ 1 / T_per_read
        

Reporting faster than the system settles simply produces more numbers faster, not more trustworthy data.