A water sample is a single moment: a jar filled at a tap or a pond edge on one morning, sent to a lab, returned as a number a week or two later. The water an auditor actually cares about is everything that happened before and after that moment — where the sample came from, whether it represents the water actually used, and what the farm did once the number came back. A lab report by itself answers none of those questions.
This is the gap that trips up otherwise careful operations. The testing itself is rarely skipped; testing is the easy part, and most farms already budget for it. What gets lost is the record around the test — the source, the date, the method, and the response — which is exactly the part an auditor or buyer is trying to read.
The standard sets the requirement; this page describes the record. Which water has to be tested, how often, against what microbial or chemical limits, and what response a failing result requires are all set by your certification scheme, your buyer, or your regulator, and they differ by commodity and by jurisdiction. Your certifier and the written standard are the authority. What follows is the shape of a record that can answer for itself regardless of which scheme is asking.
The sample is a moment; the record has to survive the season
A water test dated in May is being asked, months later, to say something true about the water that touched a crop in August. It can only do that if the record around it is specific enough to establish that the sample was representative — same source, same point of use, same conditions that plausibly apply across the intervening months. A jar of water with a lab number attached, and nothing else, cannot make that case. The record has to do the work the sample alone cannot: tie a moment to a season.
Name the source the way an auditor will ask about it
“We tested our water” is not a source. A well, a pond, a shared irrigation canal, and a municipal connection carry different risks and are usually tested and evaluated differently, even on the same farm. The record should identify the specific source by name or number, the point at which the sample was drawn — at the wellhead, at the point of use in the field, at the wash line — and which fields, lots, or wash operations that source actually supplies. A farm with two wells and one pond has, in effect, three water systems, and a reviewer will ask which test covers which one. If the answer requires memory rather than a record, the record has failed at its one job.
This is the same discipline behind spray and input records: an input applied without a specific field, rate, and date attached is not useful no matter how carefully it was measured. Water is an input like any other. It just arrives through a pipe instead of a sprayer.
Date, method, and result belong in the same line
The sample date, the date it was tested, the method or parameters used, and the numeric result should live together as one entry, not scattered across a lab PDF and a separate notebook. When these are split, the reconstruction happens at the worst possible time — the week before an audit, or the day a buyer asks — and reconstruction under deadline is how records quietly become approximate.
A record that reads clearly is one where a stranger, a year from now, could look at a single line and know: this source, on this date, tested by this method, came back at this result. Nothing about that requires software. It requires that the four facts be written down together, at the time, rather than assembled later from whatever documents still exist.
An out-of-range result is not a failure of the record — a missing one is
Every water testing program eventually produces a result outside the acceptable range. That is not evidence the farm is doing something wrong; it is evidence the testing is real. What an auditor is actually assessing is what happened next: was the result noticed, was a decision made, and is that decision written down.
The response might be a resample, a change of source, a treatment step, or a decision about which product that water touched. Whichever it is, the record of the response belongs beside the result that triggered it — not in a separate corrective-action binder that has to be cross-referenced by date. A test result with no linked response reads exactly like a test result nobody looked at, whether or not that is true. The same principle runs through organic certification records and it is worth reading alongside this one, because both live or die on whether an exception gets written down.
Testing frequency and range are the scheme’s call, not yours
It is tempting to settle on a testing cadence that feels reasonable and treat it as sufficient. The cadence, the acceptable range, and which uses of water require testing at all are set by the standard you are held to — a buyer’s program, a government rule, or a third-party certification — and they are not interchangeable. A frequency that satisfies one scheme can fall short of another, and the acceptable ranges for irrigation water, wash water, and drinking water are frequently different from each other on the same farm. Confirm the actual requirement with your certifier rather than carrying a number forward from last season or from a different operation.
The record has to reach forward, into the harvest lot
A water test that cannot be connected to the crop it touched is answering a question nobody asked. The full chain an auditor traces runs from the source, through the test, to the field or wash step it covers, to the harvest lot that field produced — the same join at the center of traceability more generally, and the reason water testing records belong to the same discipline as food safety audits as a whole rather than sitting off to the side as a lab-compliance chore — see preparing for a food safety audit for how that chain gets tested end to end.
This is where Farm40 can help with the mechanical half of the problem: its harvest and traceability exports join a lot back to the field and the inputs recorded against it, so a water test attached to a field shows up in the same packet as the harvest it covered. The limit is the same one that runs through every export — the join depends on the lot code and the field being recorded consistently in the first place. A test entered without the field it covers will not appear in that packet no matter how carefully the harvest itself was logged, because the software has no way to infer a connection nobody wrote down.
None of this is complicated to do by hand. A source, a date, a method, a result, and — when needed — a response, kept together and tied to the field it covers, is the whole record. What is hard is doing it every time a sample is drawn rather than only when someone remembers to.