Field Notes

Field notes

Last updated 3 August 2026

Four things that cost electrical contractors time and sign-offs, and none of them are about whether the work was done right. They are about what got written down, and when.

The megger reading that fails a good cable in August

Insulation resistance drops as the cable warms. Without a temperature on the sheet the reading cannot be corrected, and an uncorrected summer reading fails cable that is fine.

Insulation resistance is strongly temperature dependent, and the direction surprises people the first time: hotter insulation reads lower. The rule of thumb passed around is that resistance roughly halves for every 10 °C rise. It is only a rule of thumb — the real relationship depends on the insulation system — but it is close enough to explain the whole problem.

Work it through. A cable that reads comfortably at 20 °C in the spring is pulled through a duct bank in August, and the conductor sits at 50 °C. Three doublings down, the same healthy cable now reads roughly an eighth of what it did. If the spec puts the acceptance criterion at a fixed number and the tech writes down what the instrument said, that cable just failed.

Now the expensive part starts. Somebody raises an NCR. Somebody argues about whether to re-pull. And the cable was never bad.

Winter runs the same trick the other way, which is worse, because nobody investigates a pass. A marginal cable tested at 5 °C reads high, sails through, and the problem is found later by something more expensive than a megger.

What actually fixes it

Correct the reading to the base temperature your acceptance criterion is written at, and record everything the correction needs:

The temperature, measured, not guessed. And say what you measured — ambient air is not conductor temperature. A cable that has been energized, or lying in the sun on a tray, is hotter than the air around it.
The base temperature the criterion assumes. A number with no base temperature behind it cannot be compared to a corrected reading.
The correction factor and where it came from. IEEE 43 publishes factors for rotating machine windings referred to 40 °C. For cable, the source is your test standard or the cable manufacturer.
Test voltage and duration. The reading climbs while the test runs, so a number with no duration beside it is not repeatable.

The other half of a low reading is often not the cable at all. A damp, dusty or salted termination leaks across its surface, and the instrument cannot tell that from leakage through the insulation. Wipe the terminations, let them dry, and use the guard terminal if the instrument has one. A good clean-up has rescued more “failed” cable than any amount of arguing about the criterion.

The practice worth arguing about: “It read over a gig, it’s fine.” A single number with no temperature, no test voltage, no duration and no phase combination against it is not a test result. It is a number somebody remembers seeing. It cannot be corrected, cannot be trended against next year’s reading, and cannot be defended when a reviewer asks which of the six readings it was.

The ground grid you tested after a week of rain

Soil resistivity swings hard with moisture and temperature. A grid measured on wet ground reads optimistically and will not repeat in August or in a hard freeze.

Ground resistance is not a property of your grid alone. It is a property of your grid and the dirt around it, and that dirt changes. Moisture is the big one: as soil dries, resistivity climbs steeply. Freezing does the same thing and more — frozen ground is a dramatically worse conductor than the same soil at 5 °C.

So a grid measured on a Tuesday after four days of rain gives a number that is real, defensible, correctly measured, and not the number the grid will show in August. If your spec sets a limit the grid has to hold year round, a wet-season measurement has not demonstrated that. It has demonstrated the best case.

This matters most on the jobs where it is easiest to get wrong: duct bank and ground grid work happens early, often in whatever weather the schedule landed in, and the test gets done when the crew is there rather than when conditions are representative.

What to put on the sheet

Soil condition and recent weather. Days since measurable rain. It costs one line and it is the difference between a defensible record and an argument.
Which method, and the geometry. Fall-of-potential per IEEE 81 needs the current probe far enough out to reach flat ground, and on a large grid “far enough” is much further than people expect — often further than the site allows, which is exactly when a shortcut gets taken quietly.
What you measured, precisely. A clamp-on reading on a grid with many parallel paths is measuring something, but it is not the grid-to-earth resistance you are being asked for.
The limit you were working to, and where it came from. Not the 25 ohm figure people quote out of the NEC — that is about supplementing a single made electrode, not a system performance limit. The spec quick reference goes into that one.

If the spec requires performance in the dry season and the schedule will not wait, say so in writing at the time and agree a retest. That conversation is cheap in March and expensive at energization.

What a client audit actually looks at

Reviewers do not recheck your engineering. They pull a few IDs at random and follow each one end to end, and they widen the sample the moment something does not line up.

The fear is that a reviewer will re-derive the work — recompute the grid, re-read the relay curves, second-guess the method. That is almost never what happens. They do not have the time, and it is not their job.

What they do is sample and trace. They pick a handful of cable numbers off the schedule, more or less at random, and follow each one all the way through the package: the pull record, both termination records, the torque, the insulation resistance result, the loop check that depends on those terminations. Five cables, maybe ten.

If all of them trace cleanly, the sample stands in for the rest and the review moves on. If one breaks — a missing sheet, a date out of order, an ID that does not match — they widen the sample. And once they are widening the sample, they have stopped assuming your records are right and started looking for the next problem. That shift is what actually costs you the sign-off, not the original discrepancy.

The rest of the pass

Calibration against report dates. Take each instrument certificate in the package and check the reports that instrument produced fall inside its calibration period. This is mechanical, fast, and finds things.
Signatures. Was the person who signed qualified for it, and does anything else in the package put them somewhere else that day.
Open items. Every NCR closed with evidence of what was actually done, not just a date. Every punch item marked complete carrying a sign-off.
Revisions. Did you test to the current revision of the spec, the drawing and the relay settings — or to the one that was current when the work started.
Counts. Cables into the panel on the drawing against rows on the schedule. Nobody has to read a single test result to find that one.

Audit yourself the same way, a month out. Pick five cable numbers at random — genuinely at random, not the five you know are tidy — and trace each one end to end. If you cannot do it in ten minutes with the package in front of you, neither can the reviewer, and they will not extend you the benefit of the doubt you are extending yourself.

Why the records disagree, and what it costs

Nothing forces separate records to agree until the package is assembled, which is the worst possible moment to find out that they do not.

I have not seen a turnover package rejected because a test result was out of tolerance and properly documented. That is a known condition with an NCR against it, and it gets dispositioned. What gets packages sent back is two sheets that cannot both be true.

The reason is structural rather than anybody's fault. These records are created by different people, at different times, in different places, and often in different systems — the test sheet on paper in the field, the NCR in a spreadsheet, the settings on the engineer's desk, the schedule in the drawing office. Nothing forces them to agree while the work is happening. The first thing that ever compares them is the act of assembling the package, and by then the sheets are signed and the crew has demobilized.

What the reviewer sees	What actually happened
A loop check dated before the termination it depends on	The termination was remade to close an NCR. The NCR closed in one place, the loop sheet lives in another, and nobody re-ran the check.
Two cables that are obviously the same cable	`C-2104` and `C2104`. The ID format was never agreed at kickoff, so two crews each picked a reasonable one.
Relay as-left values that are wrong for the settings on file	The settings were reissued mid-job. The test tech was working from the revision that was current when they mobilized.
Three reports from an instrument that was out of calibration	The due date passed during the job. Nothing flagged it, and the certificate only met the reports when the package was assembled.
A schedule that does not reconcile to the drawing	The drawing went to a new revision. The schedule was built from the old one and never rebased.

Every one of these is cheap to fix on the day and expensive to fix at handover, and the gap between those two costs is entirely down to how long it took anyone to notice.

What actually helps

Agree the identifier format at kickoff and hold it. Cable numbers, equipment tags, loop numbers, instrument IDs. This is the cheapest item on this page and the one that pays back the most.
Put the expiry dates somewhere that is checked, not somewhere that is filed. Calibration, qualifications, permits, one-call tickets.
Make closing an NCR name the record that has to be redone. The rework is not the closure. The retest is.
Trace a sample monthly, not at the end. Five IDs, end to end. It takes a morning and it finds these while the crew is still on site.

Where we stop. Sparky QC finds the subset of these that is objectively true from the records — expired dates, open NCRs and RFIs, rows missing their document. It does not catch a date out of sequence, it does not reconcile a count against a drawing, and it does not know that C-2104 and C2104 were meant to be the same cable. Those are still a person tracing a sample. How Sparky QC works sets out exactly which is which.

About these notes

These are practices, not requirements. Your specification and the engineer of record govern, and where anything here disagrees with them, they win. The spec quick reference covers which document asks for which record; this page is about what goes wrong between them.

Corrections and disagreements are both welcome, and get published. Write to info@sparkyqc.com.

See also the blank form pack — the sheets most of this ends up on, free and ungated.

Field notes — sparkyqc.com/field-notes