The Site Documentation File SOP
Purpose
Published documentation describes equipment as the factory shipped it. It does not describe the building it went into, the parts somebody swapped in year six, or which of the three drawings taped inside the panel door actually matches the board in front of you. That gap is where return visits get expensive: the second tech re-learns the site from scratch, and often re-learns it wrong, because the paper on site is confidently out of date.
A site documentation file is the shop's own record of what the published documents get wrong at one address. It is not a repair history and not a job record. It answers one question for the next person: what do I need to know before I trust anything printed here?
Scope
Applies to any service address that meets the file trigger in Step 1, for every trade the shop performs at that address.
Does not apply to one-off retail calls at addresses the shop does not expect to see again, or to equipment covered under another contractor's service agreement (note who holds it and stop there - documenting somebody else's scope invites you to be blamed for its condition).
Does not replace the job record. The repair narrative, the photos taken for a damage claim, and the invoice detail stay on the work order. The site file holds only what stays true after the ticket closes. If a fact would be identical on the next visit, it belongs in the site file; if it describes what happened on this visit, it belongs on the ticket.
Roles and responsibilities
| Role | Owns |
|---|---|
| Technician on site | Captures the identity block, deviations, and isolation points before leaving the property, not from memory that evening |
| Lead technician | Reviews every file created or changed in the prior week against the three audit questions in the verification section |
| Dispatcher or office | Attaches the current file to the work order before the tech leaves the shop, and flags a dispatch to a triggered site that has no file |
| Owner | Sets the trigger threshold and the retirement rule, and pulls four files a quarter for the cold-read audit |
Procedure
1. Decide whether the site earns a file
The unit of analysis is the service address, not the customer. A property management company with eleven buildings gets eleven files, because the equipment, the disconnect locations, and the access rules differ per building.
Open a file when either of these is true - this is an OR, and one alone is enough:
- The address has produced two or more visits in a rolling 12 months, counting any trade.
- A single visit involved equipment the shop does not stock parts for on the truck, which means the next visit will start with a parts search that documentation shortens.
Below both triggers, do not open a file. A library of near-empty files trains techs to skip the step, and a file nobody opens is worse than no file because it implies coverage that is not there.
2. Open the file with the identity block
Photograph the nameplate square-on and legible, then transcribe into the file: equipment type, manufacturer designation as printed, model, serial, date code or revision marking, and the listing mark. Transcribe rather than relying only on the photo, because a photo is not searchable and the next tech is searching for a model string on a phone in a mechanical room.
Where one nameplate carries several ratings, record all of them plus which one you used and for what. A tech who sees a single number in the file will assume it governs everything.
3. Record the documentation state
This is the section that makes the file different from a maintenance record. For each document you used, log four things: which document, which revision or date, where you obtained it, and whether it matched the installation.
That last field is the whole point. "Wiring diagram, revision B, from the label inside the control compartment, does NOT match - installed board is revision D and the low-voltage terminal designations changed" is worth more than the diagram itself.
4. Record every deviation between paper and installation
A deviation is any place where the drawing, the schedule, or the manual says one thing and the installation does another. Record what the document shows, what is actually there, and how you determined it. Date every deviation line.
Never edit over an existing deviation line. Add a dated line beneath it. A crossed-out original tells the next tech that somebody changed something between visits, which is a diagnostic clue on its own; a clean overwrite hides it.
5. Record isolation and access before you record anything else about the equipment
Do this on the first pass, while you are still tracing the installation, not on the way out.
For electrical work: record the exact location and identification of the disconnecting means, the panel and circuit number, and whether the panel directory was correct. On the job itself, open the disconnect, apply your own lock and tag, and prove dead with the live-dead-live sequence, testing your meter on a known live source before and after the check (NFPA 70E-2021, 120.5). The general-industry requirement that circuits be de-energized and locked or tagged before employees work on or near exposed energized parts is 29 CFR 1910.333(b)(2); on construction work the counterpart for lockout and tagging of circuits is 29 CFR 1926.417.
For mechanical and stored energy - a compressor, an accumulator, a spring-loaded damper, a pressurized vessel - the governing standard is 29 CFR 1910.147, and the file should record where each energy-isolating device is and what stored energy has to be relieved or restrained before service.
For gas and fuel: record the shutoff location and type. If anyone on site ever smells gas, everyone leaves the building immediately, nobody touches a switch or a light, nobody uses a phone inside, and the call to the utility is made from outside.
Also record access: lockbox, key holder, elevator or roof-hatch procedure, hours the space is available, and any escort requirement. Access failures cost whole trips.
6. Record the sequence of operation as observed, where it differs from published
If the published sequence says the unit proves airflow before ignition and the installed control does something else because of a field-added interlock, that difference is the single most valuable line in the file. Write what you watched happen, in order, with the input that caused each step. Do not write the published sequence into the file; write only where reality departed from it, and name the document you are departing from.
7. Update on the same visit that finds the change
The update trigger is event-driven, not scheduled: any visit that finds a new deviation, a newer document revision, or a changed isolation point updates the file before the tech leaves the property. No quarterly review cycle. Scheduled reviews of documentation get skipped when the schedule is tight, and the tight-schedule visits are exactly the ones that discovered something.
The step size is one dated line per finding. Not a rewrite, not a summary.
8. Retire the file when the equipment goes
When equipment is replaced, do not delete the old file. Mark it superseded with the date and open a new one. The old file is how you answer a question about the previous system two years later, and it is where the reason for the replacement lives.
Worked example: what one file was worth on the second call
A rooftop unit at a small commercial site, first visited in March.
The March tech spent 0.75 hour beyond the repair building the file. In it: the nameplate block, a note that the control transformer had been replaced with a different VA rating than the one the drawing showed, and a note that the wiring diagram adhered inside the control compartment was revision B while the installed board carried revision D, with the low-voltage terminal designations differing between the two.
A different tech took the July no-heat call and read the file before rolling. He landed his meter using the revision D designations, found an open limit circuit in 1.5 hours, and left.
The shop's median time to close that same fault type, across 9 comparable tickets at sites with no file, was 3.25 hours. Against that comparator the July call ran 1.75 hours shorter (3.25 minus 1.5). The comparator is uncorrected for equipment age and site access, both of which vary across those 9 tickets, so treat 1.75 hours as the order of the saving rather than a precise figure.
Set that against the 0.75 hour invested in March and the file was 1.0 hour ahead after a single return visit (1.75 avoided minus 0.75 invested). Only that one return is measured. If the two later returns saved at the same rate, and two of the three returns each cost 0.25 hour of file updates, the arithmetic runs 5.25 hours avoided against 1.25 hours invested, about 4.2 hours avoided per hour invested - but that projection assumes every call turns on documentation, which is generous.
Both sides of that comparison are technician labor hours on site, so they compare cleanly. What the shop keeps from them is not automatic: on flat-rate pricing the avoided hours become capacity for another call, while on time and materials they become a smaller invoice, which buys the renewal rather than the revenue.
The failure mode, concretely. Had the July tech used the door diagram, he would have read continuity across a pair that revision D repurposed, concluded the limit was closed, and gone looking at the board. The likely outcome is a control board sold and installed on a unit whose actual fault is an open limit - a callback, a returned part, and a customer who now believes the shop guesses. That is not a hypothetical failure of documentation; it is the ordinary result of trusting the most convenient document on site.
How to verify a file is real
A file that exists is not the same as a file that works. Test it cold, four files a quarter, using a tech who has never been to that address:
- Can you find the disconnecting means and the shutoff from the file alone, without calling anyone? If not, the file records equipment and not the site.
- Does the file tell you which document to distrust, and why? A file listing a manual revision but not whether it matched has recorded a fact, not a warning.
- Is the most recent deviation line dated, and is the line above it still visible? If the history has been overwritten, the file can no longer show that something changed between visits.
Score each file pass or fail on all three, not on a scale. A file that fails one of the three is opened and fixed that week by the tech who last touched it, because the failure is a habit, not a document defect, and the person who formed the habit needs to close it.
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 mechanical isolation and stored energy
- NFPA 70E-2021, 120.5, on the process for establishing an electrically safe work condition, including proving the tester before and after
- See related: The As-Found Drawing and Why You Make One; How to Tell Which Manual Revision You Are Holding