The question I ask in a laboratory assessment is deliberately awkward. Pick a result from eighteen months ago and show me, from records alone, who did the work, on what equipment, with which reagent lots, against which method version, and what state that equipment was in on the day.
Well-run labs produce it in twenty minutes. Most take considerably longer, and a meaningful proportion cannot produce it at all — usually because a balance calibration record is missing, a reagent lot was never recorded, or the method has been revised twice with no version control.
That single test is the whole of Good Laboratory Practice compressed. Reconstructability.
Raw data is what you wrote down first
The most common finding I raise, across pharmaceutical, agricultural and industrial labs, is transcription. An analyst records a reading on a notepad, transfers it to the worksheet, and the notepad goes in the bin.
Under GLP the original observation is the raw data, regardless of what it was written on. Transferring and discarding destroys it. If the transcription is wrong, and roughly one in a few hundred will be, there is no way to detect it.
The remedy is unglamorous: record directly onto the controlled worksheet or into the instrument system, in indelible ink, single-line corrections with initial, date and reason. No overwriting, no correction fluid, no back-dating, no pencil. Every lab knows this and a surprising number still keep a scrap-paper habit alive because it is faster at the bench.
Equipment records are where studies die
An instrument used in a study needs a documented history: calibration against traceable standards, maintenance, and performance verification at defined intervals. What frequently exists instead is a service sticker from a contractor and nothing else.
Two specific issues recur. Balances are verified daily with check weights in good labs and monthly in poor ones — and a balance that drifts affects every gravimetric result since the last verification, which means a failed check triggers a retrospective impact assessment, not just a recalibration.
The second is temperature. Fridges, freezers, incubators, stability chambers. Continuous monitoring with alarms is cheap now, and the labs still relying on a daily manual reading have no idea what happened over a long weekend when the compressor tripped.
The Study Director role is real, not nominal
In GLP the Study Director is the single point of study control. In practice I frequently find the role assigned to a senior scientist who signed the plan, was not present for the conduct, and signs the report on trust.
That is not the role. The Study Director approves the plan, must be aware of deviations as they occur, must assess their impact, and personally signs the statement of GLP compliance. If deviations are being recorded and only surfacing at report stage, the control has failed even if every individual record is perfect.
The Quality Assurance function must be independent of the study conduct — genuinely independent, reporting to management rather than to the lab manager whose work is being inspected. Small facilities struggle with this and the honest solution is an external QA arrangement rather than a polite fiction.
Deviations are evidence of control, not failure
A study report with zero deviations makes me more suspicious, not less. Real work produces departures — a sample arrived warm, an incubation ran forty minutes long, a reagent lot changed mid-study.
What matters is that each was recorded when it happened, assessed for impact by the Study Director, and reported. A lab with an open deviation culture is a lab where I trust the data. A lab with an immaculate record has either extraordinary discipline or a habit of not writing things down, and the second is far more common.
Electronic records
The moment a chromatography data system enters the picture, audit trails matter. Audit trail review has to be an actual scheduled activity, someone has to look at who reprocessed what and why, and integration parameter changes need justification. Shared logins are indefensible and still widespread.
Where to start if your system is weak
Run the reconstruction test yourself on three historical results before anyone external does. Whatever you cannot produce is your gap list, in priority order, and it will be shorter and more specific than any consultant's report.