How to Build a Documentation File for a Recurring Site

Why this matters

Most site records get written in the truck at the end of the day, out of memory, and they are thin in exactly the places that matter: the dip switch positions nobody wrote down, the wire colors at a terminal strip that has since been disturbed, the model string on a nameplate that is now behind a reinstalled panel. Those facts were available for about twenty minutes and then closed.

The reason site files come out thin is not laziness. It is that the capture was attempted at the wrong moment. Almost everything worth recording is only recordable while the equipment is in one particular state, and the states arrive in a fixed order that the job dictates and you do not control. Work with that order and the file costs you well under an hour. Work against it and the file costs you a second trip.

The three windows, and why the order is not yours to choose

Read that table as a constraint, not a checklist. There is no field in row two you can capture in row one, and every field in row one is gone the moment you open a door. The order below follows the windows because the equipment gives you no other order.

Window one: what exists only while it is running as you found it

Before a tool comes out of the bag, before a cover comes off, before you cycle anything:

1. Photograph the equipment in place, wide enough to show what it connects to. A photograph of a unit with no surroundings tells the next tech nothing about clearance, drainage, or how the piping or ductwork leaves it.
2. Record every switch, valve, damper, and setpoint in the position you found it. Not the position it should be in. The as-found position is frequently the diagnosis, and the moment you correct one you have destroyed the evidence that somebody else set it wrong.
3. Watch it run and write the sequence you observe, with the input that triggered each step. Call for operation, what energized, in what order, how long between steps. Do not copy the published sequence into the file. Write only what you watched, and note the document you would have expected to match.
4. Photograph every label, tag, and directory already on site, including the ones you think are wrong. Especially the ones you think are wrong. A panel directory that misidentifies a circuit is a hazard the next tech needs warned about, and your photograph is the proof that it read that way before you touched it.

Window one is cheap and it is the one most often skipped, because nothing has broken yet and it feels like delay. It is the only window that cannot be reopened by coming back.

Window two: what exists only while it is open and isolated

Lead with isolation, not with the camera. A pump set, a compressor, a packaged unit and a boiler all carry two independent hazards and they cite differently.

For the electrical side, open the disconnecting means, apply your own lock and tag, and prove dead using the live-dead-live sequence, checking your meter against a known live source immediately before and immediately after the test (NFPA 70E-2021, 120.5). The general-industry duty to de-energize and lock or tag circuits before working on or near exposed energized parts is 29 CFR 1910.333(b)(2); on construction work the counterpart is 29 CFR 1926.417.

For the mechanical side, isolate and relieve stored energy before opening anything that holds it - pressure in a vessel or piping loop, a charged accumulator, a spring-loaded damper or check, a suspended load. That is 29 CFR 1910.147 territory, and the reason it is a separate citation is that 1910.147 expressly sends electrical exposure elsewhere. On a pump set you will use both, in the same visit, on the same skid.

With the equipment open and proven dead, capture in this order:

1. The nameplate you could not see from outside, square-on and legible, plus the ratings you did not use as well as the one you did.
2. The terminal strip, in one uncropped photograph, before you land a single lead. Then a second photograph zoomed enough to read designations. If you disturb landings, photograph again after. Terminal capture is the single highest-value item in this window because it is the field the next tech most often needs and the one most likely to have been changed since the drawing.
3. Configuration switches, jumpers, and rotary settings, in the positions found. Photograph and transcribe. A photograph of a bank of eight small switches is easy to misread at a glance later; the transcription is what gets used.
4. Board or module revision markings, and any diagram adhered inside the compartment, with its revision. Record whether the adhered diagram matches the installed board. This is the field that later saves a whole diagnosis.
5. Component identifiers on anything that is a wear part - motor, contactor, capacitor, control transformer, seal kit, valve cartridge. The next visit starts with parts availability, and a legible identifier is worth more than a category name.

Window three: what exists only after it is back together

1. Confirm the sequence still runs as you recorded it in window one, and record any difference. If it now runs differently and you did not intend that, you changed something.
2. Record the operating values you would want as a baseline - the readings your trade takes to call a system healthy, on this machine, on this day, with ambient conditions noted alongside them. A baseline without its conditions is not comparable to anything.
3. Correct the site's own labels. If the panel directory was wrong, fix it. NFPA 70 (the National Electrical Code), Article 408, requires a panelboard circuit directory that legibly identifies the purpose of each circuit, so correcting one is not optional courtesy. Photograph the corrected directory into the file so a future disagreement has a date on it.

What to do when window two never opens

On occupied or process-critical equipment you will sometimes finish a visit without ever isolating - the tenant cannot lose the system, or the shutdown window belongs to someone else.

That does not mean you skip the file. It means you write the gap in as an explicit field: what was not captured, why, and what would have to be true to capture it. "Terminal landings not photographed - unit could not be taken down during occupied hours, next opportunity is the scheduled shutdown" is a usable line. A silent gap reads to the next tech as if the field did not exist, and they will plan a visit on the assumption they already have it.

This is also the only condition under which the window order changes: with no window two, your window three baseline is the only verification the file will ever carry, so take more readings than you otherwise would.

Worked example: the filled-in file for a pump set

A packaged pump set with its own control panel in the basement of a small multi-family building. Second visit in eight months, so it met the trigger for a file.

Total time on site was 2.5 hours, of which 0.6 hour was capture, split 0.2 hour in window one, 0.3 hour in window two, and 0.1 hour in window three (0.2 plus 0.3 plus 0.1 equals 0.6). What went into the file:

• Identity: pump set nameplate transcribed, plus the separate motor nameplate found only after the guard came off. The two disagreed on full-load current, which is normal - the set plate carries the assembly rating and the motor plate carries the motor's - and the file records both with a line naming which one the overload sizing follows.
• Documentation state: a control schematic adhered inside the panel door, undated, showing a two-pump alternation scheme. The installed controller carried a revision marking two steps later than the schematic's part number, and its alternation was lead-lag on hours rather than on cycles. Logged as a mismatch, with the observed behavior in window one as the evidence.
• Deviations: one pump's suction isolation valve found roughly one third closed, in a position that was clearly deliberate rather than drifted. Photographed and left as found on this visit, with a note, because changing it would have changed the system's behavior before anyone had characterized it.
• Isolation and access: panel and circuit identified, both mechanical isolation valves located, the note that the piping loop holds pressure after the pumps stop and must be relieved before any seal work, and the building access rule (key held by the on-site manager, no basement access after hours).
• Baseline: discharge and suction readings at both pumps with the building's occupancy noted, taken in window three.

The failure mode this file was built against, and what it actually cost the previous tech. The prior visit produced a work order that said "replaced seal, tested, OK" and nothing else. When this visit needed the terminal landings, the panel had to be isolated and opened again purely to read them - 0.4 hour and a second lock-and-tag cycle for information that had been sitting open, in good light, in front of somebody eight months earlier. That is the whole economics of window two: 0.4 hour to reopen against a capture that would have cost a fraction of it while the panel was already open, plus the risk that lives in every additional isolation cycle.

Note what is not claimed here. The 0.6 hour of capture is unbilled effort the shop absorbs, and the 0.4 hour is technician time on a later ticket. Both are labor hours, so they subtract cleanly, but neither is margin until you decide what the recovered time gets used for.

How to verify the file will survive a cold read

Hand it to someone who has never been to the site and ask them to answer three questions from the file alone, with no calls:

• Where do I isolate, and what stays stored after I do? If the file names a disconnect but not the pressure that remains in the loop, it will get somebody hurt on a job it was written to help.
• Which document on that site should I distrust, and how do I know? A file that lists documents without verdicts has recorded inventory, not knowledge.
• What did the equipment look like before anyone touched it? If every photograph shows an open panel, window one was skipped, and the as-found record does not exist.

If any answer requires a phone call, the file is not finished. Fix it from the photographs you still have, and if the photographs do not carry it either, write the gap in as a gap rather than leaving a blank that reads as an absence of a problem.

References

• 29 CFR 1910.333(b)(2), general industry, and 29 CFR 1926.417, construction, on de-energizing and lockout or tagging of circuits before work on or near exposed energized parts
• 29 CFR 1910.147, control of hazardous energy, for isolation and relief of stored mechanical, hydraulic and pneumatic energy
• NFPA 70E-2021, 120.5, for the live-dead-live proving sequence
• NFPA 70 (National Electrical Code), Article 408, on panelboard circuit directories
• See related: The Site Documentation File SOP; The As-Found Drawing and Why You Make One