Ask a farm what records it keeps and you will usually get a list of documents. That is the wrong unit. A record is not a document; it is an answer, held in reserve, to a question someone has not asked yet. The buyer who wants to know what was sprayed on the lot they are about to sign for. The certifier who wants to know whether the field was managed the way the plan says. The auditor who wants to know what you did the week before a positive test came back. Three readers, three questions, and — this is the part that trips people up — often the very same piece of paper.
This page is about the whole file, not any single record in it. It ends where most farms actually live — not short of records, but unable to find the one that matters.
The label, the certifier, and the regulator are the authority. This page describes the process of keeping records. It does not tell you which records you are required to keep, or for how long — those are set by your buyers, your certifier, your regulator, and, for anything applied to a crop or an animal, the product label. They vary by program and jurisdiction and they change. A record proves what you did; it does not decide what you were required to do. When this page gives a figure, it is a visibly hypothetical illustration, never guidance.
The same record answers to three different readers
A single input application record can be read three ways. The buyer reads it to confirm the produce is safe to put their name on and to check that no interval was violated before harvest. The certifier reads it to confirm the input was permitted under the program you are enrolled in and that it matches what your plan says you would do. The auditor — who arrives only when something has gone wrong — reads it to reconstruct a sequence of events after the fact, backwards from a symptom.
These readers want different things from the same fields: the buyer wants the pre-harvest interval honored, the certifier wants the active ingredient on the permitted list, the auditor wants the lot number that ties the application to a specific batch. A record built to please only one of them fails the other two at the worst possible moment. This is why the discipline that governs a spray and input record is not “write down what you sprayed” but “write down everything each of three readers will one day need, while you still know it all.”
A mixed farm carries more categories than it thinks
A farm that grows crops and keeps livestock is really running two record systems that occasionally touch. On the crop side there are plans, plantings, scouting observations, input applications, and harvest lots — the chain that crop management is built to keep intact. On the livestock side there are groups, health and treatment events, breeding, weights, and feed, which is the province of livestock management. Each of those is a category with its own fields and its own reader.
Then there are the categories that sit across both. Traceability — a lot traced one step back to what went into it and one step forward to who received it — is not a crop or a livestock record; it is a spine both hang off. Financial records tie every expense back to the field or group that incurred it, which is what makes farm finances more than bookkeeping. Labor records answer for who did the work and what training they held. And if you are certified or audited, an organic system plan or a food safety program overlays commitments you have to produce evidence for.
The number of categories is not the problem. The problem is that they are supposed to connect — the treatment record has to reach the sale, the input record has to reach the harvest lot, the lot has to reach the customer. A farm with tidy records in eleven binders that do not reference each other has done the recording and skipped the part that makes records worth keeping.
The longest retention period governs the whole file
Different records have to be kept for different lengths of time, set by different parties — a certifier names one period, a buyer’s contract another, a regulator a third for a particular class of record. These do not negotiate with each other and do not expire together. The only safe reading is that the longest applicable period governs the record it touches, and that it is easier to hold the whole file to the strictest standard than to track a dozen clocks and gamble on which record you are allowed to discard.
There is a cruel logic to which record you will actually reach for. It is almost never today’s. An audit samples the past; a recall traces backwards from a product already in the market; a residue investigation works back from a test result toward a treatment that happened long before anyone suspected a problem. The record you need is, reliably, the oldest one you are still obliged to hold — the one whose details nobody remembers and whose context has evaporated, and which therefore has to be complete on its own terms.
A record that describes is not a record that constrains
Most farm records describe. A planting record says what went into the ground; a labor log says who worked and for how long. They are testimony: made once, filed, and consulted only when a question happens to arise. Their whole job is to be accurate and findable.
A minority of records do something harder. They do not describe the past; they constrain the future. A treatment record that starts a withdrawal window makes an animal ineligible to sell until a date arrives. A re-entry interval makes a field unsafe to walk into until time passes. These records carry a consequence forward, and the consequence has to be honored whether or not anyone thinks to look the record up. This is the entire subject of withdrawal and residue — the sharpest case, where a record has to be able to stop a sale that has not happened yet.
The two kinds fail differently. A descriptive record fails quietly — you go to find it and it is not there. A constraining record fails loudly — the window was open, nobody checked, the animal shipped, and the failure surfaces as a residue violation weeks later. Treating a constraining record like a descriptive one, a thing you consult when curious rather than a thing that acts on you, is how the loud failures happen.
Latency between doing and writing is the root cause
Nearly every recordkeeping failure traces back to a single gap: the distance in time between doing the thing and writing it down. Close that gap and most of the other problems dissolve; leave it open and no amount of downstream diligence repairs it.
A record made at the moment of the work is made by the person who did it, who knows the dose and can read the lot number off the container still in their hand. The same record made that evening at the desk is made by someone reconstructing — and reconstruction quietly rounds. The exact rate becomes the usual rate; the precise time becomes “that afternoon”; a record made on Sunday for the whole week is a well-intentioned fiction. Every hour of latency trades a fact for an approximation, and an approximation is worse than a blank, because it carries the authority of the written word without the accuracy.
So the real test of any system — paper, spreadsheet, or software — is not how it looks in a review. It is whether an entry can be completed with cold hands, standing up, in bad light, while the work is still going on. A system that can only be filled in later will be, and later is where records go to become approximate.
“Defensible” means it holds up without you in the room
People reach for the word “defensible” without saying what it means. Concretely, a defensible record answers the question it was made for to a skeptical stranger, months or years later, without your presence, your memory, or your benefit of the doubt. If producing it requires you to explain, to fill a gap from recollection, or to say “what that column really means is…,” it is not yet defensible.
That standard has parts you can check. A defensible record is contemporaneous — made at the time, not reconstructed. It is complete on its own terms, so it does not lean on a second document or a person’s memory. It is attributable — it says who did the work and who recorded it. It is specific: “group B-12, until March 4” rather than “the back pasture, sometime after calving.” And it is durable, stored so it survives the wiped whiteboard and the truck that got traded in. A record can be honest and still fail this test, which is why “but I really did it” is not a defense. Preparing for a food safety audit is largely the work of turning honest-but-thin records into defensible ones before anyone asks.
Most farms have a retrieval problem, not a recording problem
Here is the admission that reframes everything above. Walk onto a farm that just failed an audit or fumbled a recall and you will rarely find a farm that did not record. You will find a farm that recorded and cannot retrieve. The note exists — on a whiteboard that got wiped, in a notebook riding around in the wrong truck, in someone’s head, in a spreadsheet three versions behind. The information was captured; it simply cannot be produced in the minutes available when a buyer or an auditor is standing there waiting.
And from their side of the table, a record you cannot produce is indistinguishable from one that never existed. The buyer does not get the reassurance; the certifier cannot verify the claim; the auditor writes it up as missing. All the discipline of contemporaneous, complete, attributable recording is undone by the last step nobody practices: getting the oldest relevant record into a waiting person’s hands. So the honest advice to most farms is not “record more.” It is “put every record in one place, connected and searchable, so retrieval stops being an archaeology project.”
Retrieval is also the one thing a farm software decision should actually turn on. Not features, not dashboards — retrieval. The question to ask any system, digital or paper, is whether it can hand you the oldest record you are still obliged to keep, joined to the records around it, before the person asking loses patience. If a spreadsheet does that, keep it. If it does not, that is the problem worth spending money to solve, and the only one.
Where a tool fits, and where it stops
This is the one place software belongs in this page, because retrieval is the problem software is genuinely built for: holding many records in one place and joining one to the next so an old entry is not stranded. Farm40 builds a traceability packet that walks a harvest lot back to the inputs applied to it — through the planting they share — and forward to the customers it shipped to, so a buyer’s question or a mock recall can be answered from a single trail instead of five binders. The limitation belongs in the same breath: that packet is only ever as complete as the data you entered. If a planting was never linked, or an input went unlogged, the trail has a hole exactly where you were relying on it, and no amount of software fills a gap in the record itself. A tool can make good records easy to retrieve. It cannot make records you never wrote.
Which returns to where the page began. Keep the categories your buyers and certifiers name, hold the file to the longest clock, and write at the moment of the work. But above all build for retrieval, because the day it matters, the record you need will be the oldest one you have, and the only thing that counts is whether you can put your hand on it.