What to Record So the Next Water Problem Is Solvable
Why this matters
Almost every water investigation dead-ends in the same place, and it is not where people expect. It is not "what is in the water." A test kit and a lab answer that in days. It is "when did this start, and what else happened around then," and that question is answered by a record that most sites do not keep at all.
A water file at a typical commercial site is a stack of values: readings, lab reports, dose logs, all accurate, all dated. It can tell you everything about the water and nothing about the system, and the second half is where the cause lives. Water does not change on its own. Something changed the water, or changed what the water touches, and that something has a date and a name.
The two questions a file gets asked
Over a system's life, a water record is opened for exactly two reasons.
The first is routine: is this parameter where it should be right now. A values file answers this perfectly, which is why nobody notices the gap for years.
The second only arrives when something is wrong: this parameter is out of band, so when did it leave, and what else happened in that window. A values file answers the first half of that with the resolution of its own sampling interval, and cannot answer the second half at all, because it contains no entries about anything except water.
The consequence is specific and it repeats. An investigation that cannot name a candidate event ends in a change to the treatment program, because that is the only lever the file describes. Sometimes that is right. Often the water was fine and something upstream of it moved, and the treatment change is a cost added on top of an unfixed cause.
The change log, and what actually goes on it
The missing half of the file is a dated log of changes to the system and to its inputs. Twelve entry types cover nearly everything, and the ones people leave off are marked because those are the ones that solve cases:
- Water source or supply change, including a utility switching sources seasonally, which happens without any notice to the building
- Softener or filter service: media change, regeneration setting change, salt type change, and any bypass opened during the work
- Any valve position change, especially a bypass, with the date it was opened and the date it was closed as two separate entries
- Equipment added to or removed from the loop
- Treatment product change, including a substitution described as equivalent, because equivalent means equivalent in function and not in composition
- Dose rate, setpoint or controller change
- Any drain-down and refill, with the volume
- Repairs that introduced new metal, such as a copper repair on a steel system
- Chemical cleaning, with the product class and the date
- Load and occupancy changes, and seasonal changeover dates
- Work by anyone else on the system, including other trades and the customer's own staff
- The makeup or supply meter reading at every visit, which turns the log into a rate record as well as an event record
The pattern in that list is worth naming. More than half of those entries are made by people who are not the treatment vendor, which is exactly why a vendor's service log is not a change log. A treatment vendor records the treatment vendor's actions faithfully and has no visibility into a plumber's bypass, a controls contractor's setpoint edit, or the day the building switched a wing back on.
Where it lives and who is allowed to write in it
One file, with the values and the changes in the same place, ordered by date. Two files defeat the purpose, because the entire value of the log is reading it alongside the value series.
Entries are one line each, dated, with the name of whoever made the change or observed it. Nobody needs to judge relevance at the time of writing, and asking them to is how entries get omitted: the person who opens a bypass for forty minutes does not think of it as a system change. So the rule is stated as an action type rather than as a judgment. If you opened, closed, added, removed, replaced or adjusted something, it gets a line.
Two things do not go in. Narrative impressions with no measurement behind them, because they read like data later and cannot be compared to anything. And a second copy of any value that lives elsewhere, because two copies of a series diverge and the divergence is discovered at the worst moment.
There is also a limit on how you reconstruct a log that was never kept. Do not move a valve to find out what it does on a live system. A bypass or isolation valve changed on an occupied building can remove protection from a connection, dead-head a pump, drop a zone, or send treated water somewhere it has never been. Reconstruct from drawings, photographs, and the people who did the work, and where a position genuinely has to be verified, that is a planned outage with the building notified rather than a handle turned during a walkthrough.
The gate
One question, run against a file:
Given an out-of-band reading, can this file name the window in which the parameter left its band, and can it name what else happened inside that window?
Below are two files with identical value histories, and they resolve oppositely.
File A: values only
Quarterly conductivity readings on a makeup line, eight readings across two years:
410, 415, 405, 420, 690, 720, 705, 760 microsiemens per centimeter.
The first four scatter within about 15 of each other around 410, with no trend. Then there is a step. The last four scatter within about 70 of each other around 720, a noticeably wider spread than the first group, which is itself a finding: whatever is controlling this parameter got looser at the same time it got higher. The second group averages about 1.7 times the first.
What the file can say: the parameter stepped up somewhere between reading four and reading five. That is a 90-day window, because the sampling interval is quarterly, and no amount of care in taking those readings makes the window smaller.
What the file cannot say: anything at all about what happened in those 90 days. The investigation now starts with the water, because the water is the only thing described. The likely next step is a treatment adjustment sized to the new condition, which treats a symptom whose cause is still running, and the second-order effect is that the new dose becomes the new normal and the step is never revisited.
File B: the same eight readings, plus a change log
Same values, same step, same 90-day window. The log has four entries inside that window:
- Softener resin replaced; service bypass opened for the duration of the work.
- A rooftop unit tied into the loop.
- Treatment product changed to a different supplier's equivalent.
- Cooling season startup.
The file has not identified the cause. It has done something more useful: it has replaced an undated 90-day window with four dated candidates, and three of them can be tested in an afternoon.
Entry three is checked first because it is the cheapest: compare the two products' composition and the dose rate, and if the new product carries a different salt load it would raise conductivity without anything being wrong. It did not, in this case. Entry two is checked next: a new unit adds volume and surface but does not add dissolved solids to the makeup line, and the reading in question is on the makeup line rather than in the system, so it cannot be the cause of a step there. Entry four is a seasonal pattern, and the previous year's readings do not show a seasonal step, so it does not fit.
Entry one is left, and it is testable directly: a hardness reading downstream of the softener against one upstream. The softener bypass had been opened during the resin work and never closed, so unsoftened supply had been feeding the system for the entire period covered by readings five through eight. The wider scatter in that second group fits the same explanation, because the raw supply varies more than the softened output did.
The fix is a valve handle. The treatment program never needed changing.
Why the log beat the values, stated plainly
The values told you a parameter moved. They could not tell you why, and they could not have, no matter how many of them there were. Increasing sampling frequency from quarterly to monthly would have narrowed the window from 90 days to about 30 and left the second question exactly as unanswerable, at three times the sampling effort.
That trade is worth sitting with, because shops reflexively reach for more readings when a file fails them. More readings buy resolution on the timing question only. The change log is the only thing that buys candidates on the cause question, and it costs a line of text per event rather than a visit per interval.
How to verify a file is actually answerable
Do not audit the file by reading it. Audit it by asking it a question it has not been asked.
Pick any parameter with a series in the file and pick the largest single move in that series. Then try to name what else happened in the interval containing that move, using only the file. If you cannot, the file has the gap, and you have just found it on a day when nothing is broken, which is the only comfortable day to find it.
Two more checks worth running once a year. Count the change entries against the number of service events you know occurred, including other trades: a log with only your own visits in it is a vendor log wearing a change log's name. And look specifically for paired entries, a bypass opened with no matching closed, a system drained with no matching refill volume, a temporary connection made with no removal. An unpaired entry is either a record-keeping lapse or a condition that is still live in the building, and there is no way to tell which from the file, which is precisely why it deserves a walk. A temporary hose connection left open is the same finding a cross-connection survey is looking for, reached from the paperwork side.
References
- Treatment supplier documentation for product composition, target ranges and dose records
- The water utility for source changes and supply quality history, which a building will not otherwise learn about
- See related: How to Read a Water Test Report; The Heat Exchanger That Failed Twice in One Year; How a Cross-Connection Turns Chemistry Into a Health Problem