The record is the result: data integrity and traceable documentation in a small laboratory

A measurement that nobody recorded did not happen. That sounds like a slogan until you try to settle a disagreement six months after the fact, with two people who remember the day differently and a spreadsheet that has been saved over four times. Then it becomes an operating rule. The thing a laboratory actually produces is not a number on a screen for a few seconds. It is a record: a durable, readable, attributable account of what was done, by whom, to what, when, and with what result.

Everybody working in a small laboratory already knows this in outline. The trouble is that record keeping degrades quietly. No single shortcut is serious. A time written in at the end of the afternoon rather than at the moment. An identifier shortened because everyone knows which one is meant. A file called final-2. A shared login because it is quicker. Each one is defensible on the day. Together they turn a record into a story, and a story cannot be checked by anyone who was not there.

A record is not the paperwork that follows the work. It is the only part of the work that still exists tomorrow.

This guide is about the ordinary end of that problem: what a record has to contain before someone outside the room can rely on it, how to correct one without destroying what it said before, how a single identity is carried from the moment something arrives to the line that reports it, and where record sets usually break. It is written for readers in small settings, where one or two people do everything and nobody has a quality department to fall back on. The standards it draws on are written for regulated laboratories, but the reasoning is not regulatory. It is what makes a record worth keeping at all.

What a record is for

Records answer three different audiences, and it helps to keep them apart because they ask different questions.

The first audience is your future self. Six weeks on, something looks odd, and you need to know exactly what was different about that day: which reference material, which instrument, which settings, who was standing there. This is the audience that is served by detail, and the one most often let down by a record that captures only the conclusion.

The second is a colleague who has to repeat the work. They need enough to do it the same way without asking you, which means the record has to contain the things that felt too obvious to write down: the order of steps, the waiting times, the make and setting of the equipment, what was done when something did not behave.

The third is someone sceptical: an auditor, a customer, an insurer, a court. They are not interested in whether you remember it correctly. They are interested in whether the record could have been changed after the fact without leaving a mark, and whether what it says can be corroborated by something outside itself. This third audience is the one that sets the standard, because a record built for a sceptic satisfies the other two automatically.

Raw data, and the copies that follow it

The single most useful distinction in this whole subject is between raw data and everything derived from it. Raw data is the first durable record a measurement leaves behind: the instrument's own output file, the printed trace, the reading written down at the bench as it was read, the photograph taken at the time. Everything after that is a copy or a calculation, and each step away can only lose information or introduce it wrongly.

Two consequences follow, and they are worth stating plainly.

The summary is not the record. A table pasted into a report is a derived document. If the underlying file has been deleted, the table cannot be checked against anything, and its correctness now rests entirely on the honesty and attention of whoever built it. That may be fine between colleagues. It is not evidence.

Transcription is a step that can fail. Every time a number is copied by hand from a screen to a book, or from a book to a spreadsheet, there is a chance of a transposed pair of digits, a dropped decimal point or a value written on the wrong line. A record that keeps the original output alongside the transcription lets a reviewer catch that in seconds. A record that keeps only the transcription makes the error permanent and invisible.

The practical rule is unglamorous: keep the first thing, not only the tidy thing. Export the instrument's own file and store it under the identifier of the work it belongs to. Photograph the printed trace before it fades. Keep the bench book even after the results have been typed up. Storage is the cheapest part of the laboratory.

The nine attributes: ALCOA and the four that were added

The initialism ALCOA came out of regulated laboratory practice in the 1990s as a memory aid for what an entry has to be. Four further attributes were added later, and the extended set is usually written ALCOA+. It is worth knowing not because anyone will test you on the letters, but because it is a complete checklist for a single entry, and most record problems are one of these nine failing.

The first five letters are ALCOA. Complete, consistent, enduring and available are the four that were added later, and in small settings they are the four that fail most often, because they are about the system around the entry rather than the entry itself.

Read the middle column as a set of questions to ask about your own book. Most people find at least two they cannot answer, and the two are usually original and complete: the first output has been thrown away, and the attempts that failed were never written down.

Contemporaneous: why the clock matters more than the memory

Of the nine attributes, the one people concede most easily is the one that costs the most. Writing up at the end of the day feels harmless, because the memory is fresh and the account will be more coherent than a set of scrawls. That coherence is exactly the problem. A record written after the fact is a reconstruction, and a reconstruction quietly smooths over the things it cannot recall: the order in which two steps were actually done, the eleven minutes of waiting that should have been four, the second attempt that replaced the first.

A contemporaneous record has a texture that a reconstructed one does not. It contains times that are not round numbers. It contains crossings-out. It contains lines that stop mid-thought because something needed attention. Reviewers learn to read that texture, and a suspiciously neat book, in a suspiciously even hand, in a single shade of ink, with all times ending in zero or five, is the first thing that makes an experienced reader slow down.

The practical version of this rule in a small setting is not "write everything down as it happens", which nobody manages. It is:

• Record the time at the point of the action, not the point of writing. If they differ, say so, and say why.
• Write on the record itself, not on a scrap to be copied over. If a scrap is unavoidable, keep it and attach it. The scrap is the original.
• Enter the awkward things first. Failed attempts, interruptions and anything out of the ordinary are the entries a reconstruction loses, and the ones a reviewer looks for.
• Never leave blank lines to fill in later. A gap in a bound book invites a later entry that cannot be dated, and reviewers treat it that way.

Corrections, and how to make one that survives review

Records are made by people, so they contain mistakes, and the handling of a mistake says more about a laboratory than the mistake does. There is one correct way to correct a written entry, and it has been the same for decades:

• One line through the error, thin enough that the original stays readable.
• The new value written beside it, not over it.
• Initials and the date of the correction, not of the original entry.
• A short reason, in the margin or a note: transcription error, wrong line, instrument reset, misread scale.

What that produces is a record with two truths in it: what was first written, and what it was changed to. Both are information. Correction fluid, an overwritten figure, a torn-out page or a rewritten sheet destroys the first truth and, with it, the credibility of the second. A reviewer who finds a page that has been recopied "for neatness" has no way to tell a tidy-up from an edit, and is obliged to treat it as the worse of the two.

The reason for the reason is worth spelling out. Three corrections on one page with no explanation reads as an unreliable record. Three corrections that each say transposed digits, checked against printout reads as an attentive one. The same facts, opposite conclusions, and the only difference is four words.

One identity, from arrival to report

The thing that turns a pile of correct entries into a traceable record set is a single identity carried the whole way through. Something arrives. It is given an identifier. That identifier appears on the receiving entry, on the storage location, on every record of work done with it, on the instrument output, and on the line in the report that refers to it. Where that thread runs unbroken, the report is evidence about a real item. Where it is cut, the report is a statement about something unspecified.

Threads break in a small number of predictable ways, and it is worth being blunt about them.

• Renaming at a boundary. A receiving reference becomes an internal number, which becomes a short form on the bench, which becomes a column heading in a spreadsheet. Each step is sensible; nobody wrote down the mapping; the chain is now only in one person's head.
• Sub-division without a new identity. One arrival is split into three parts and all three keep the same reference. A result that differs between them can no longer be attributed to a part.
• Re-use of an identifier. The numbering restarts each year, or a short form collides with an older one. Two different things now share a name in the archive.
• Silent gaps. Between arrival and first use, nothing is recorded. The record cannot say where it was or who had it, and that period is exactly where a sceptical reader will push.

The fixes are equally ordinary. Assign the identifier at the moment of receipt, before anything else happens to it. Write the mapping down the first time a name changes, in a register that anyone can find. Give a new identity to anything divided. Never re-use a retired one, even years later; the cost of a longer number is nothing next to the cost of an ambiguous archive.

Custody: who held it, when, and the gaps between

Custody records answer a narrower question than people assume. They do not prove that nothing happened to an item. They record who was accountable for it during each period, and therefore make the unaccounted periods visible. That is their entire value: a custody record with no gaps tells you where to ask, and a custody record with a gap tells you the honest limit of what anyone can claim.

A usable custody entry has four parts, and it is the fourth that is usually missing: who released it, who received it, when, and in what condition. Condition at hand-over is what separates a transport problem from a handling problem later, and without it every subsequent dispute becomes a matter of opinion.

In a two-person laboratory this sounds like bureaucracy for its own sake, and for routine internal movements it often is. The proportionate version is to record custody at the points where accountability actually changes hands: arrival from outside, transfer to another site or person, return from anywhere it left, and final disposal. Between those points, a storage location and a date are enough.

Equipment: logs, calibration, and the check that is not calibration

Every instrument that contributes to a result carries two record obligations. The first is a use log: date, who used it, what for, the settings that were not the defaults, and anything unusual about how it behaved. The second is a status history: calibration, servicing, repairs and the checks that fall between them.

Two words get used interchangeably here and should not be. Calibration establishes the relationship between what an instrument indicates and a known reference, and is normally done by a competent external party against a traceable standard, on a schedule, with a certificate at the end of it. Verification is the much cheaper routine check that the instrument is still behaving as the last calibration said it would: a known mass, a known reference, a duplicate reading, done by you, often daily.

Verification does not extend a calibration. It only tells you the day something drifted rather than the month. That distinction matters most after the fact: if a verification on the 14th passed and the one on the 15th failed, every result between the 14th and the discovery is in a defined window. If nobody verified anything between two annual calibrations, the window is a year, and a year of results is now questionable. Verification records are therefore not housekeeping. They are what limits the damage.

Electronic records: audit trails, clocks and shared logins

Most small laboratories are half on paper and half in files, which is the hardest combination to keep straight. Three points cover most of what goes wrong.

An audit trail is only an audit trail if it cannot be switched off. Instrument software often logs who changed what and when, and just as often that logging is disabled by default, or can be turned off by any user, or is kept in the same file that it is meant to protect. Find out which of those is true of your own software before you rely on it, because a trail that the person of interest could have cleared is not evidence about that person.

A shared login destroys attributability at a stroke. One account used by three people produces records that cannot be attributed to anyone, which fails the first attribute outright and cannot be repaired afterwards. This is the single most common serious finding in small laboratories, and it is nearly always adopted for convenience rather than any real constraint.

Clocks drift and nobody notices. A benchtop machine with a wrong date stamps every file it produces with a time that contradicts the paper record. That contradiction, found two years later by a stranger, looks exactly like backdating. Check the clock on every device that writes a timestamp, write the check down, and if you correct a clock, record the offset so earlier files can still be placed in order.

The same logic applies to file naming. Final, final-2 and final-real are a version history with the dates taken out. A file name carrying the identifier and an ISO-style date is readable by anyone, sorts correctly on its own, and survives being moved somewhere else.

Which document answers which question

A working record set is not one big book. It is a small number of documents that each answer a different question, and most confusion in small settings comes from expecting one of them to do another's job.

The right-hand column is the one to read twice. Nearly every overstated claim in a small laboratory comes from using one of these documents to answer the question in its own third column.

Procedures need versions, and versions need dates

A written procedure is a record like any other, and it has one requirement people forget: a reader must be able to tell which version was in force on a given day. Without that, a bench record saying "performed per procedure" points at nothing in particular, and if the procedure has since been improved, the record now claims compliance with a document that did not exist at the time.

The minimum is a version number and an effective date on every page, one controlled copy that is known to be current, and superseded versions kept rather than destroyed. Keeping the old ones feels like clutter until the first time someone asks what the instruction actually said in March, at which point it is the only thing that can answer.

The other half is honesty about drift. Procedures get quietly adapted at the bench: an extra rinse, a longer wait, a different order because it works better. None of that is wrong, and all of it is invisible to a reader who has only the document. Either the procedure is updated to match the practice, or the practice is recorded as a deviation each time. What cannot hold is a document that everybody knows is not what happens.

Deviations and results that fall outside expectation

The entries nobody wants to write are the ones that matter most, and the reflex to leave them out is the single behaviour that turns an honest record set into a misleading one.

A deviation is any departure from the plan: a step out of order, a wait that ran long, a substitution, an interruption. Recording one should be routine and unembarrassing. The note needs four things: what happened, when it was noticed, what was decided, and who decided. A laboratory whose records contain no deviations at all is not a laboratory where nothing goes wrong. It is one where the going-wrong is not written down, and an experienced reviewer reads an unbroken run of perfect records as a warning rather than a reassurance.

A result outside expectation needs more care, because there is a strong pull towards discarding it. The discipline is to investigate before deciding: first look for a cause in the record itself, in the equipment history and in the handling, and only then decide whether the result stands. Whatever the outcome, the original result, the investigation and the decision all stay in the record. Repeating until a result is acceptable, and keeping only the acceptable one, is not a method. It is a way of guaranteeing the answer you wanted, and a record set that shows only final values with no repeats is where that practice becomes visible.

Retention, archiving and the backup nobody has tested

Retention is the question of how long a record has to survive, and it is decided by whichever is longest of: a contractual obligation, a legal one in your jurisdiction, the working life of whatever the record is about, and your own need to defend the work. Write the answer down as a schedule, per record type, with the person responsible named. A retention rule that exists only as an assumption produces two failures at once: things thrown away that were needed, and everything else kept forever because nobody dares decide.

Archiving is the practical half. Paper wants a dry, secure, defined place, with an index somebody other than the author can use. Electronic records want three things people routinely get wrong:

• A copy in more than one place, at least one of which is not in the building and not on the same account.
• A format that will open later. A proprietary file that needs a licence and a specific version of one program is a hostage. Keep an open-format export alongside the original, not instead of it.
• A restore that has actually been tried. An untested backup is a belief about a backup. Restore one folder, on a different machine, once a quarter, and write down that you did. The failures show up on the day you need it, and never before.

The same applies to the medium. Thermal printouts fade to blank within a few years, which is why a photocopy or a scan made at the time, marked as a verified copy with the date and the initials of whoever made it, is worth more than the original it came from.

Who is allowed to sign, and what a signature means

A signature on a record is a claim, and it should be a specific one. There are only three in ordinary use, and they should never blur together:

• Performed by. I did this work and this entry is my account of it.
• Checked by. I have read this record against the source and it is complete, legible, dated and free of unexplained corrections.
• Approved by. I take responsibility for the conclusion drawn from it.

A check is not a rubber stamp; it is a defined reading, and it should have a defined list. Is every page dated and initialled? Does every correction carry a reason? Does the identifier on the record match the one on the item and on the output? Is any repeat or failure accounted for? Five questions, two minutes, and they catch most of what a later reviewer would find.

Who may sign each of those is a training question. The record that says somebody was competent to perform a task is part of the same set: what they were trained on, by whom, when, and how competence was demonstrated. In a small setting this can be one page per person, updated when something changes. Without it, a signature is just a name.

Eight ways record sets actually fail

These are not exotic. They are the recurring findings, in roughly the order they turn up.

1. One login, several people. Nothing in the electronic record can be attributed to anyone, and no later process can repair it.
2. The original was thrown away. Numbers were typed up and the instrument file deleted, so nothing can be checked against anything.
3. Only the good runs are written down. The record describes a laboratory in which nothing ever fails, which no reader believes.
4. Corrections without reasons. A page of crossings-out that may be careless transcription or may be something else, and the record cannot tell you which.
5. The identifier changes at a boundary and the mapping was never written down, so the report and the item are joined only by memory.
6. Undated procedures. "Performed per procedure" on a day when nobody can say which version that was.
7. A clock nobody checked. File timestamps that contradict the paper record and look, years later, exactly like backdating.
8. A backup nobody restored. Discovered on the one day it was needed, which is also the only day it cannot be fixed.

A ten-point look at your own records

Take one completed piece of work from three months ago, pull every record connected to it, and answer these in order. Anything you cannot answer from the paperwork alone is a gap, whatever you personally remember.

1. Can you name the person who made each entry, from the record and not from memory?
2. Is every page dated, and do the dates and times run in a sensible order?
3. Does the first, original output still exist, and can you put your hand on it now?
4. Does one identifier run unbroken from the receiving entry to the final report?
5. Does every correction have a visible original, initials, a date and a reason?
6. Is there any record of something not going to plan, and if not, is that plausible?
7. For every instrument involved, can you show its status on that day?
8. Can you produce the version of the written procedure that was in force then?
9. Did a second person read the record, and is it clear what they checked?
10. Could a stranger, with the file alone and no access to you, reach the same conclusion?

The tenth question is the whole guide in one line. Everything above is a way of making the answer to it yes: not because the work was good, though it may have been, but because the record of it can be read, checked and relied on by somebody who was never in the room.

Good work with a poor record is indistinguishable, later, from poor work. That is not a criticism of the work. It is a description of what a record is.