When the Drawing and the Installation Disagree

Why this matters

Sooner or later the print in your hand describes something that is not in the box in front of you. The reflex answer, "the print is out of date," is right most of the time, and that is exactly what makes it dangerous: a rule that is right most of the time gets applied to the case where it is wrong, which is the case where a safety has been defeated. Deciding which one governs is a separate question from deciding which one is true, it has a rule, and the rule takes about thirty seconds to run per deviation.

First, exclude the cheapest explanation

Before either of them is wrong, check that they are about each other.

• Model and serial transcribed off the nameplate, compared against the document's applicability block, including any serial break.
• The right section of the unit. A family print covering three sections is easy to read one section off.
• The right unit on a multi-unit site. Two identical rooftops with a print pulled from the wrong door pocket disagree with each other for the most boring possible reason.
• Option package resolved. A "when equipped" note explains most apparent disagreements in a family print, and explains them completely.

Misidentification is the most common cause of a print-versus-panel disagreement, and it is the only cause that costs nothing to eliminate. Run it first every time.

The gate

Per deviation, not per document: does this deviation touch a protective function, a listed configuration, or a code-required function? If any one of the three is true, the document governs and the field has to change. If none is true, and the installation as it stands is safe and code-compliant, the field governs and the record has to change.

The Boolean is OR across those three conditions, because each one alone is sufficient. And note it runs per deviation. A panel with several deviations gets several runs of the gate and can legitimately land on both answers in the same enclosure.

What steps up when the gate lands on "document governs" is not just the paperwork. The response changes from a note you write to a correction you make before the equipment goes back in service, and if you cannot make the correction on this visit, the equipment stays down and the customer gets told why in one sentence.

The three conditions, spelled out.

• Protective function. Anything whose job is to interrupt, relieve, prove, or limit. Limits, safeties, interlocks, flow and float and pressure switches, relief devices, freeze and overflow protection.
• Listed configuration. Where equipment is listed or labeled, NFPA 70 (NEC), Article 110 requires it be installed in accordance with the instructions included in that listing or labeling, so a field arrangement that departs from those instructions is not merely undocumented, it is outside the terms the listing was granted on.
• Code-required function. Grounding and bonding, conductor identification, overcurrent protection, clearances, venting, backflow protection. The code requirement stands whether or not the manufacturer's drawing mentions it.

Precedence when documents conflict with each other

The gate settles document against installation. This settles document against document, in descending authority:

A manufacturer instruction that is more restrictive than code still applies to that listed equipment. Rank 1 above rank 2 does not mean the loosest requirement wins, it means code sets a floor no document can lower.

Outcome A: the document is superseded

Air handling unit, family print, revision block legible, serial inside the printed range. The print shows two separate relays performing a fan interlock. The unit has one combined board doing the same job, and the print shows nothing like it.

Deviation classified. It touches a control function, not a protective one. Nothing about it is code-governed. The remaining question is the listed configuration, which is settled by evidence on the part itself: the board carries its own manufacturer label and a build date later than the unit's nameplate build date, the factory harness lands on it with factory terminals, and there is not a splice in the section. That is a documented service replacement, not a field improvisation.

Gate result: none of the three conditions is met, the installation is safe and compliant, so the field governs and the record changes. The print gets marked at that location, dated, initialed, with one line naming what is actually installed. Nothing gets "restored" to match the drawing. Rewiring a current arrangement backwards into an obsolete one because a piece of paper shows it that way is a real failure mode and it is done in good faith every time it happens.

Outcome B: the installation is wrong

Same visit, same panel. The print shows the equipment grounding conductor landed on the marked grounding terminal in the enclosure. In the panel it is under a sheet metal screw shared with a mounting bracket.

Run the gate. Protective function, arguably. Listed configuration, arguably. Code-required function, unambiguously: equipment grounding and the terminations that carry it are governed by NFPA 70 (NEC), Article 250, and a shared sheet metal screw is not a terminal identified for that purpose.

One condition is enough. The document governs, the field changes, and it changes on this visit because it is a five-minute correction. The conductor moves to the marked grounding terminal, torqued to the value marked at the terminal or given in the equipment instructions, and it gets photographed before and after.

Same gate, same panel, same half hour, opposite answers. That is the point of running it per deviation instead of forming an opinion about the print as a whole.

The proportion you should expect, and the one that should worry you

Across that panel, five deviations were logged. Four resolved to outcome A, stale documentation to be recorded. One resolved to outcome B, a correction. That ratio is ordinary and healthy: documentation drifts constantly, installations are wrong occasionally.

What is not ordinary is a panel where several deviations all land on outcome B. Multiple code or protective deviations in one enclosure is not a set of independent mistakes, it is a signal about who worked in that panel and how, and it changes your scope: you stop treating deviations as items and start treating the enclosure as unverified in its entirety.

When you cannot classify a deviation at all

Sometimes you find a conductor, a jumper, or a device whose purpose you genuinely cannot determine, and the gate has nothing to bite on.

Treat it as outcome B until you can classify it. Not because it is probably wrong, but because the two errors are not symmetric. Calling a stale drawing a defect costs you a phone call and some time. Calling a defeated safety a stale drawing costs whatever that safety existed to prevent. When the cost of the two mistakes differs by that much, the tie does not go to convenience.

Concretely: the unknown item gets photographed, drawn on your as-found, described in writing without a conclusion attached ("18 AWG blue, non-factory jacket, from TB1-5 to an unmarked terminal on the board, purpose not determined"), and the equipment does not go back in service on the assumption that it is harmless. If the customer needs it running before you can resolve it, that is a conversation with a named risk in it, not a default.

What goes in the record either way

Both outcomes produce a written result, and they are different documents.

Outcome A produces a markup on the drawing and a line in the job record: what the drawing shows, what is installed, the evidence that the installed arrangement is legitimate, dated and initialed. Outcome B produces a correction record: the condition found, the code or protective function it touched, what was done, and a photograph of before and after.

The distinction matters months later. A marked print says "this was checked and the drawing is behind." A silent print says nothing at all, and the next tech pays for the check a second time.

Checking that you ran the gate honestly

Two quick self-tests, both aimed at the way this rule actually gets abused.

Take any deviation you classified as outcome A and ask what evidence made the installed arrangement legitimate rather than merely present. If your answer is that it looked factory, that is a description of appearance, not evidence. Build date, factory terminations, absence of splices, a matching service part label: those are evidence.

Then take any deviation you classified as outcome B and ask which of the three conditions it met, by name. If you cannot name one, you may be correcting a drawing rather than a defect, and rewiring a working unit to match an obsolete print is how a service call creates the callback it was sent to prevent.

References

• NFPA 70 (National Electrical Code), Article 110, requiring listed or labeled equipment to be installed in accordance with the instructions included in the listing or labeling
• NFPA 70 (National Electrical Code), Article 250, equipment grounding and the terminations that carry it
• Manufacturer documentation for model applicability and serial breaks, which is what settles ranks 3 through 5 in the precedence table
• See related: How to Verify Documentation Against the Equipment in Front of You; The Drawing That Was Right and the Installation That Was Not