Why are parallel conductors carrying unequal current?

Why do parallel feeder conductors carry unequal current?

Unequal sharing comes from the entire path, including both connections and the installed cable arrangement. A poor joint usually diverts current into its mates, while unequal magnetic coupling can change AC sharing even when isolated DC resistances match. Confirm the current difference before choosing a connection repair or a route correction. [S1]

Intermediate joints and bus connections also contribute. Current flows through every connected path; the hottest cable body need not contain the defect.

What makes apparently identical cables electrically different?

Matching cable markings do not prove matching in-service impedance. Oxidation, strand damage, poor lug seating, an extra splice, or asymmetric bus connections change the path. At AC, spacing, phase order, raceway properties, and mutual coupling also matter. See the parallel-conductor installation guide for sizing and matching rules. [S4] [S10]

Keep three findings separate: unequal A1/A2 currents within phase A, unequal total A/B/C currents, and unequal surface temperatures. Moving single-phase loads addresses phase loading; it does not automatically correct unequal sharing within one phase. Establish which finding exists before specifying work.

How do you calculate current sharing?

For independent paths at one frequency, divide each path’s admittance by total admittance and multiply by total current. Admittance is the reciprocal of impedance; AC quantities include phase angle. The method is I_k = I_total × (1/Z_k) / Σ(1/Z_j). Significant mutual coupling requires a coupled impedance model within the scope of IEC 60287-1-3:2023, edition 2.0. [S1]

For two resistive paths, each current equals total current multiplied by the other path’s resistance divided by their resistance sum. Anvilfield reports sharing by averaging simultaneous branch RMS currents, finding the largest absolute departure, and dividing by that average: U = 100 × max(|I_k − I_avg|)/I_avg. This reporting index is separate from code acceptance or motor voltage-unbalance criteria.

Worked example: one connection adds resistance

In Anvilfield’s 400 A resistance-only calculation, parallel paths of 1.00 and 1.50 mΩ carry 240 and 160 A, producing a 20% maximum departure from their 200 A average. Its high-to-low spread divided by the mean is 40%. Name the denominator with every percentage.

Anvilfield estimates use the table’s cable and joint resistances, hold total current constant, and neglect reactance. Resistance division gives each branch current; multiplying resistance by current squared gives the losses, rounded to two decimals.

In the same Anvilfield calculation, a 0.52 mΩ joint carrying 160 A dissipates 13.31 W, while a 0.02 mΩ joint carrying 240 A dissipates 1.15 W, so the lower-current path dissipates more heat at its joint under the stated inputs. The voltage-drop guide explains whole-run voltage loss.

What changes with three paths or an open path?

These Anvilfield estimates hold total current at 600 A and neglect reactance. Include the installed open branch as zero when calculating the mean; deleting it falsely reports perfect balance between the surviving pair. Actual loads can change total current after a fault, so record feeder current independently. For AC currents with different angles or harmonics, RMS magnitudes cannot simply be added; use synchronized waveforms or phasors.

How much current imbalance is acceptable?

Compare the highest branch current with that branch’s permitted loading, verify the installation, and investigate repeatable departures from baseline that exceed measurement uncertainty. No universal 5% or 10% tolerance applies to every parallel feeder; any project tolerance needs a defined measurement method and branch-loading check. A total-phase breaker does not separately supervise each parallel branch.

With an Anvilfield estimate input of 220 A permitted per branch, the worked 240 A branch exceeds its limit by 20 A although total current remains below the two-branch sum of 440 A. At a stable 60% maximum share, the branch constraint corresponds to 366.67 A total, calculated by dividing 220 A by 0.60. This screens branch loading; a damaged connection still requires correction.

Establish the branch limit using conductor ampacity and derating, terminal ratings, continuous-load treatment, ambient conditions, grouping, and equipment protection. Ampacity is a thermal current limit, while the NEC load calculation establishes demand. Neither calculation proves equal sharing. [S12]

Fluke’s 2005 thresholds set corrective-action priorities, rather than safe operating temperatures. Compare the actual component temperature with its manufacturer limits and its prior trend even below either trigger. NETA currently lists MTS-2023 and ATS-2025; use the edition specified for the job. [S7]

How do you measure parallel conductor currents correctly?

Map A1/A2, B1/B2, C1/C2 and parallel neutrals at both ends. Record voltage, frequency, operating mode, phase-total current, and the acquisition interval. Prefer synchronized capture of every branch within the phase being assessed; separate wireless timestamps alone do not establish simultaneous sampling.

Use true-RMS instruments with suitable range, bandwidth, crest-factor capability, installation category, and sensor geometry. Enclose one branch per sensor and record channel identity. A clamp around multiple phases measures their combined magnetic effect, not individual current sharing. Record the RMS averaging window as well as the displayed time. [S2]

Fluke’s January 2022 1773/1775/1777 Product Specifications lists four current inputs. One unit therefore cannot capture six branch currents at once. Capture the same-phase pair and a phase-total reference together, then repeat by phase; use synchronized additional equipment for all six. Confirm that the selected recorder mode retains the independent channel measurements. [S16]

OSHA 1910.333(a)(1) requires de-energization before exposure to live parts unless the employer demonstrates increased hazards or infeasibility; testing that requires power is a stated example. Paragraph (c)(2) limits energized work to qualified persons using suitable protective practices. Plan access with the arc-flash study and labels; install sensors during isolation whenever safe placement requires moving conductors. [S13]

How large can meter error be?

At 60 Hz sinusoidal current, centered jaws, and 18°C to 28°C meter temperature, the September 2015 Fluke 376 FC manual’s jaw specification gives worst-case base bounds of ±4.50 A at 200 A and ±4.62 A at 206 A, so a 6 A difference alone does not establish unequal current beyond those instrument limits. [S2]

These are specification bounds, rather than statistical confidence intervals; systematic errors need not cancel. At 240 A and 160 A the corresponding jaw bounds are ±5.30 A and ±3.70 A, well below the 80 A difference. Check closed jaws, centering, range, calibration, and nearby magnetic fields; a discrepancy that follows a swapped sensor indicates measurement bias.

The manual adds 0.1 times specified accuracy per degree Celsius outside that band and an additional 2% for crest factor above 2. Flexible-probe resolution changes from 0.1 A through 999.9 A to 1 A above that through 2500 A; position error is additional. Recalculate uncertainty for the actual sensor, waveform, range, and enclosure temperature. [S2]

Which installation details change path impedance?

Trace the physical route against the one-line diagram, including intermediate joints and taps. NFPA lists NEC 2026 as its current publication, but the locally adopted edition governs the installation. Mike Holt’s 2023 article and 2026 textbook identify parallel conductors at 310.10(G), previously 310.10(H)(2) in Connecticut’s NEC 2017 interpretation. [S19] [S12] [S20] [S10]

The named 2023/2026 interpretations match length, material, area, insulation, and termination method within each paralleled phase or corresponding neutral. This does not require phase and neutral conductors to have identical sizes. Where separate raceways are used, compare conductor count and electrical characteristics as well as conductor matching; confirm the actual phase grouping at enclosure entries. [S12] [S20]

Connecticut’s I-09-18 interpretation requires the tapped load to connect across all conductors of the phase in that installation. Trace taps before treating source-end and load-end currents as a simple divider. [S10]

Equal resistance can still produce unequal AC current

Anvilfield estimates use the independent-path complex divider and the table’s 60 Hz inputs. The vector sum is 400 A; RMS magnitudes add to 404.96 A. IEC 60287-1-3:2023 addresses current sharing and circulating losses in parallel single-core cables across different physical layouts, so equal DC resistance alone cannot establish equal AC current sharing. [S1]

Use route photographs, landing drawings, and a coupled model when mutual terms matter. The IET study, Steady and transient characteristics analysis for parallel cables, treats a coupled 220 kV installation. Do not correct geometry by arbitrarily coiling spare cable or changing phase formation. [S14]

Is the problem termination resistance or AC impedance?

A local hot spot together with unusually high isolated joint resistance supports a connection diagnosis. Persistent sharing differences with matched DC resistance direct attention to cable arrangement, bus impedance, or hidden joints. Compare source joints, receiving joints, cable bodies, and every same-phase mate before choosing the next test.

Before isolated resistance testing, OSHA 1910.333(b)(2) requires disconnection from all energy sources, control of hazardous stored energy, lockout/tagout, and qualified-person verification, including induced voltage and backfeed. Follow the electrical isolation procedure. Kelvin current leads inject current; separate potential leads sense voltage inside the injection points. [S13] [S17]

Four-wire sensing removes lead-drop error from the sensed voltage; it does not remove a real parallel bypass. Across the still-connected 1.00 and 1.50 mΩ example paths, an ordinary end-to-end measurement returns their 0.60 mΩ equivalent. Document disconnected bypasses and the exact sense boundaries before assigning the result to one cable or joint.

Megger’s July 2, 2021 article describes a supported DLRO100 current-clamp method that accounts for parallel-path current. Retain protective grounds and use that instrument’s supported selective method, or isolate the branch under the approved procedure. An ordinary four-wire meter does not supply that compensation merely because it has Kelvin leads. [S18]

Record actual test current, polarity or supported reversal/offset mode, settling, resolution, calibration, repeatability, and conductor temperature. Thermal EMFs can distort micro-ohm readings; use the meter’s documented compensation method. Choose an instrument-supported current that yields a stable reading without damaging or heating the connection. Select test current from the meter and equipment instructions; NI’s 1 mA DMM example addresses another application. [S6] [S23]

Compare bulk copper resistance at a common temperature. Megger’s 2005 guide multiplies measured resistance by the ratio of 234.5 plus reference temperature to 234.5 plus measured temperature: R_ref = R_meas × (234.5 + T_ref)/(234.5 + T_meas), with temperatures in Celsius. Anvilfield estimates that uniform copper measuring 1.00 mΩ at 20°C measures about 1.16 mΩ at 60°C without deterioration. [S23]

Correct bulk cable resistance using conductor temperature; an IR surface reading does not establish core temperature. Compare joints with like construction and equipment baselines, since mixed-metal contacts and films behave differently. Cooling or disturbing a joint can change its outage reading; retain the loaded evidence.

Use coupled-path analysis when mutual inductance, sheath bonding, or differing phase angles materially affect sharing. Dividing a common voltage-drop magnitude by each branch’s RMS current does not establish independent impedance elements in a coupled network. Include the relevant return paths and geometry. [S1]

How do you compare temperatures at unequal currents?

Pair images with current readings from the same stable operating window. Measure ambient air separately and record target surface, view angle, distance, emissivity, reflected temperature, IR-window transmission, and air movement. The infrared inspection guide covers the broader camera workflow; here the comparison must also preserve each branch’s current. [S5]

Use temperature rise above ambient rather than absolute Celsius temperature. Multiply that rise by the square of reference current divided by actual current: ΔT_ref = (T_surface − T_ambient) × (I_ref/I_actual)². This Anvilfield estimate assumes unchanged resistance and heat transfer; convection, radiation, thermal history, and changing resistance can defeat the approximation.

Worked thermal comparison at a common 200 A reference

At a common 200 A reference, Anvilfield estimates rises of 8.33°C and 50.00°C from the table’s current and temperature inputs using the squared-current factors. The sixfold normalized-rise ratio directs follow-up toward the lower-current joint and its thermal surroundings.

Use normalization to rank follow-up tests; use measured temperatures and equipment limits for operating decisions. The normalized 50°C rise is calculated, and the raw 20°C surface spread occurs at unequal currents. Compare those actual surfaces at comparable loading before applying a similar-load severity cutoff. Small actual currents amplify normalization uncertainty, making a repeat at comparable operating load more useful.

Fluke’s 2005 guide recommends ideally at least 40% of maximum load; its live article uses 40% of typical load. ASNT’s October 2026 note labels its NFPA 40%-of-design-load attribution historical. Follow the site’s governing load requirement and record its denominator and actual branch currents. [S3] [S5] [S8]

A camera cannot see a termination through an ordinary steel cover. Provide a suitable IR window or authorized access, then establish stability with repeated images and current records. Changing emissivity cannot correct an obstructed view, and a low-load image cannot establish the temperature at the next production peak. [S5]

Which pattern points to which fault?

Each pattern needs a discriminating test. Poor connections, unfavorable coupling, and instrument bias can coexist. Address visible damage, unstable operation, excessive branch loading, and abnormal heating before optimizing the sharing percentage.

Anvilfield’s matrix combines circuit analysis with the measurement, thermal, and installation guidance. Confirm the leading cause with the discriminating test before specifying repair. [S1] [S5] [S6]

Why can a nearly open connection look cool?

A near-zero-current branch deserves attention even when it looks cool. As a joint approaches an open circuit, falling branch current can reduce its local heating while transferring load to the remaining paths. Recheck identification and sensor range, then establish continuity under isolation; inspect the mates that carried the additional load.

What changes with harmonics, neutrals, DC, and shielded cables?

Steady DC sharing is governed by resistance, including temperature and connection condition. Use the instrument’s DC capability and zero procedure: the Fluke manual’s flexible AC probe does not measure steady DC. [S2]

With nonlinear AC loads, retain true-RMS values and waveform or harmonic files. Frequency-dependent impedance and sensor bandwidth affect both the physical sharing and its measurement. A 60 Hz accuracy example cannot establish accuracy for every harmonic-rich waveform; check the instrument conditions against the captured current.

Measure each parallel neutral and compare corresponding paths separately from phase totals. Triplen harmonics from nonlinear single-phase loads can add in the neutral even when fundamental phase currents balance. Separate that spectrum problem from unequal sharing between the parallel neutrals. The harmonics and power-quality guide addresses nonlinear-load investigation; motor voltage-unbalance limits do not provide a parallel-neutral acceptance criterion.

Shielded medium- and high-voltage cables add sheath bonding and circulating-current losses to the analysis. IEC 60287-1-3:2023 calculates phase currents for any sheath-bonding arrangement but assumes both-end bonding for its sheath-loss calculation. Record the bonding arrangement and model the relevant conductors and sheaths rather than transferring an unshielded feeder calculation unchanged. [S1] [S14]

What repair addresses the cause?

Isolate the entire circuit and verify absence of voltage before disconnecting or tightening a parallel branch. Both ends connect to common nodes, so opening one end leaves the other energized; the surviving paths also take more load. [S13]

Inspect strands, lug compatibility, conductor count, strip length, crimp tooling, plating, and thermal damage. Match replacement conductor and connection characteristics to the approved installation, and inspect the mates that carried extra current. Use the splices and terminations guide to structure the compatibility record.

Why blanket retorquing is the wrong repair specification

Schneider Electric’s FlexSeT instruction bulletin JYT1078000 treats its bus joints as maintenance-free after pre-energizing checkout, prohibits sanding away plating, and directs replacement of damaged bars or lugs. A generic instruction to clean and retorque every connection conflicts with that equipment-specific procedure. [S15]

Separate installation torque, repair assembly torque, and routine maintenance. Record the actual device, conductor, fastener, units, instruction revision, and required torque; use a calibrated tool when that procedure calls for one. The termination torque QA guide supports the record. No single torque value applies to all parallel feeders.

Route length, geometry, or bus landing corrections require an engineered arrangement. Do not add arbitrary resistance, swap phases, or move a single conductor into a separate steel raceway to improve a meter reading. A matched DC-resistance result alone cannot authorize an AC geometry change.

What proves the repair restored current sharing?

Verify current sharing at the first suitable loaded operating period after repair, then repeat the thermal comparison after temperatures stabilize. Define that verification point and its owner before the outage. When returning load is too small to reproduce the finding, include a later loaded visit in the work order.

Repeat applicable isolated tests, followed by synchronized current and both-end temperature measurements on every path, including unrepaired mates. Close the finding when the defect is corrected, the installation complies, each branch meets its loading limit, and sharing meets the documented design or project criterion. Explain any repeatable residual difference with measurement uncertainty and the installed impedance model.

Compare more than one normal load level when the operating cycle permits. A stable geometry difference usually preserves a repeatable sharing pattern; a changing joint condition can make the split drift as it warms. Record that trend alongside the final percentage.

Post-repair balance record: Anvilfield estimates

Restoring the modeled A2 joint makes both paths equal and gives 200 A per branch. The after-repair temperatures are separate estimate inputs; collect actual loaded temperatures and retain raw readings beside the calculations.

Join observations by circuit, path, target, and acquisition window. Repeated images and currents establish stability; mass, ventilation, ambient conditions, and preceding load determine settling time. Record the actual warmup interval.

Grace’s technical article reports maximum thermography intervals of 12 months for Conditions 1/2 and 6 months for Condition 3 under NFPA 70B:2023. NFPA now lists 2026; establish the routine interval from the edition governing the site. Schedule fault follow-up at the first suitable loaded condition. [S11] [S9]

What should diagnosis and repair cost?

Anvilfield estimates $1,750 for the defined diagnostic visit and $1,650 for one planned connection repair plus loaded verification, or $3,400 combined. The scope is one accessible low-voltage feeder with two paths per phase, one coordinated outage, no cable replacement or engineering redesign, and a return loaded check.

Anvilfield estimates four field hours for the two-worker diagnostic visit, three for the planned repair visit, and two for separate loaded verification, plus three reporting hours across the job. Field hours include setup and logging; they are not hours of interrupted service. Switching, access, repair extent, and the operating window set outage duration.

Method: Anvilfield estimates dated November 5, 2026 multiply stated worker-hours by assumed billing rates and add the listed allowances. Each additional two-worker hour adds $250 before materials, equipment, and operating losses. Price utility charges, temporary power, production losses, excess travel, permits, taxes, and after-hours premiums separately when required.

Public-project bid scope should separate diagnosis, connection repair, and loaded verification so an outage does not consume the testing allowance. Identify who supplies switching coordination, operating load, and access to both ends. Past electrical bid winners and award amounts in PorchRocket Bid Control give estimators a procurement reference when packaging feeder diagnosis, outage repairs, and loaded verification as separate bid items.

Field checklist and evidence package

Use this sequence alongside the employer’s switching procedure. Keep logs, images, drawings, instrument details, governing editions, loading limits, repair instructions, and unresolved findings in one asset file.

1. Identify the feeder and every same-phase branch at both ends; resolve missing or contradictory IDs before comparing data.
2. Record the one-line, system voltage/frequency, operating mode, applicable code edition, permitted branch loading, and equipment instructions.
3. Confirm the authorized access method and electrical work procedure; arrange isolation for any intrusive test or repair.
4. Verify sensor category, range, bandwidth, crest-factor capability, calibration, placement, and channel identity.
5. Capture simultaneous branch currents and a feeder reference; preserve the time window and changing-load information.
6. Measure both-end surfaces and cable-body patterns with ambient air and optical/environmental settings; retain visible and thermal image pairs.
7. Calculate sharing separately within each phase, retain any open branch as a zero-current path, and compare every branch with its permitted loading; apply the actual uncertainty before declaring a small imbalance.
8. Trace lengths, joints, taps, raceway grouping, cable formation, and bus landing positions against the design.
9. Isolate suspected paths under the work procedure; document bypass removal or the supported selective method, four-wire sense points, test current/mode, temperature, and repeatability before interpreting resistance.
10. Select the cause-specific repair and device instructions; replace damaged components where required and inspect the loaded mates.
11. Repeat electrical and loaded thermal checks under comparable conditions; arrange a revisit if operating load cannot support the comparison.
12. Record actual results, remaining limitations, closure decision, owner-assigned next inspection date, and linked evidence files.

Sources checked November 5, 2026. Update log: November 5, 2026 editorial review clarified isolation, sensor acquisition, thermal priorities, neutral harmonics, and repair acceptance. Byline: Anvilfield.

How to cite this page: Anvilfield, Why are parallel conductors carrying unequal current?, 2026, https://anvilfield.com/field-guides/electrical/parallel-conductor-current-imbalance/

Related tools

FAQ

People also ask

• What does it mean to parallel conductors?
• What does an infrared inspection find?
• What is an electrical load calculation?
• Does the NEC require a torque wrench for terminations?
• What is conductor ampacity?

Sources checked

• S1: IEC, IEC 60287-1-3:2023 CMV, edition 2.0, published May 22, 2023. Checked November 5, 2026. Publisher scope, May 22, 2023 publication date and edition 2.0 read with curl; full standard clauses not used.
• S2: Fluke, 374 FC/375 FC/376 FC Users Manual, September 2015, pp. 24-27. Checked November 5, 2026. Complete manual retrieved with curl and parsed; jaw/flexible-probe accuracy, frequency, resolution, crest factor, temperature coefficient and position allowances checked.
• S3: Fluke, Industrial maintenance thermal imaging application guide, copyright 2005, document 8037ENEN. Checked November 5, 2026. Complete 2005 guide read with curl; historical 15°C peer and 40°C ambient priorities and 40%-of-maximum-load guidance verified.
• S4: Fluke, Solid ground: Increasing parallel conductor ampacity, discussion of NEC 2011. Checked November 5, 2026. Manufacturer article retrieved with curl; historical NEC 2011 discussion, rather than current code text.
• S5: Fluke, Using thermal imaging cameras for electrical inspections, undated live application article. Checked November 5, 2026. Manufacturer article read with curl; paired load/thermal evidence, optical access and post-repair imaging checked.
• S6: NI, Part III: How To Reduce Errors When Switching Low Resistances, updated September 23, 2026. Checked November 5, 2026. Manufacturer technical note read with curl; September 23, 2026 update, Kelvin method and thermal-offset compensation checked.
• S7: InterNational Electrical Testing Association, Standards Development, live edition schedule. Checked November 5, 2026. Publisher schedule read with curl; MTS-2023 and ATS-2025 editions verified, without assigning unseen table contents to either.
• S8: ASNT, Infrared Thermography in Electrical Maintenance, Materials Evaluation, October 1, 2026. Checked November 5, 2026. Public introduction and historical-load editor’s note read with curl; October 1, 2026 publication date verified.
• S9: NFPA, NFPA 70B Standard Development, current edition 2026. Checked November 5, 2026. Official page retrieved with curl; embedded metadata identifies 2026. Final clause text was not available and is not used for current interval requirements.
• S10: Connecticut Office of the State Building Inspector, interpretation I-09-18 Meter Disconnect, November 20, 2018, applying NEC 2017. Checked November 5, 2026. Complete agency interpretation read with curl; NEC 2017 section numbering, matching purpose and all-path tap connection verified for that installation.
• S11: Grace Technologies, How Often Does NFPA 70B Require Electrical Maintenance?, live October 14 article checked November 5, 2026. Checked November 5, 2026. Manufacturer article read with curl; 12/12/6-month values explicitly attributed to NFPA 70B:2023, not final 2026 clauses.
• S12: Mike Holt Enterprises, Conductors, based on the 2023 NEC, undated page. Checked November 5, 2026. Code educator’s original article read with curl; NEC 2023 interpretation of matching and ampacity, rather than primary NFPA clause text.
• S13: OSHA, 29 CFR 1910.333, Selection and use of work practices, current web text. Checked November 5, 2026. Curl access denied; official OSHA text reopened through web retrieval. Paragraphs (a), (b)(2) and (c)(2) checked.
• S14: IET, Steady and transient characteristics analysis for parallel cables, ACDC 2020 conference, added to IEEE Xplore September 15, 2021, DOI 10.1049/icp.2020.0029. Checked November 5, 2026. Curl returned an access page; publisher abstract reopened through web search. Arrangement/coupling, 220 kV validation and conference/DOI metadata checked.
• S15: Schneider Electric, FlexSeT Switchboards instruction bulletin JYT1078000, Bus Bar Joints, Lug Terminations, and Insulating Materials, live section without visible revision date. Checked November 5, 2026. Manufacturer bulletin section retrieved and read with curl; bus-joint retightening prohibition, plating protection and damaged-part replacement verified.
• S16: Fluke, 1773/1775/1777 3 Phase Power Quality Analyzer Product Specifications, January 2022, Current Inputs. Checked November 5, 2026. Manufacturer Product Specifications, January 2022, retrieved with curl and parsed; four current inputs verified in Current Inputs section.
• S17: Seaward, How do we measure low resistance?, undated manufacturer FAQ, checked 2026-11-05. Checked November 5, 2026. Manufacturer FAQ read with curl; separate current/potential leads and sense boundaries verified.
• S18: Megger, Testing parallel resistances, July 2, 2021. Checked November 5, 2026. Curl blocked; complete manufacturer article reopened through web search. Ordinary parallel-network errors and supported DLRO100 current-clamp method verified.
• S19: NFPA, NFPA 70 (NEC) Code Development, current edition 2026. Checked November 5, 2026. Official page retrieved with curl; embedded metadata identifies NEC 2026. Publication status is distinct from local adoption.
• S20: Mike Holt Enterprises, Understanding the 2026 National Electrical Code, Volume 1, first printing 1.26, pp. 365-367 and 370. Checked November 5, 2026. Complete publisher-hosted sample PDF read with curl; first printing 1.26 and parallel-conductor interpretation on page 366 checked.
• S23: Megger, A Guide to Low Resistance Testing, document MEG-429_V04/MIL/2.5M/7.2005, p. 14. Checked November 5, 2026. Complete manufacturer-authored guide retrieved with curl from distributor host; copper correction on page 14 and document identifier checked.