TURNAROUNDQC
← Back to Turnaround QC
Last updated 12 August 2026
Seven things that cost money on a turnaround, and are not obvious until the second time they happen to you. Each one produces a record that looks fine.
Fitting a blind is a permitted job with a card behind it. Pulling it is a milestone in a queue at four in the morning. One of those gets written down reliably; the other gets remembered.
A blind goes in because a permit needs it. It is planned, it is on the isolation certificate, and nothing downstream of it opens until somebody has signed that it is fitted. The record makes itself, because the work in front of it is blocked until the record exists.
Removal runs the other way. De-blinding happens during box-up, at night, alongside forty other things, and on most jobs it is one line on a start-up checklist covering the whole unit. Nothing is blocked by it. Nobody is waiting on the signature. The crew pulling the blind is frequently not the crew that fitted it, and by then the man who fitted it is on days off. The asymmetry is structural, not carelessness, which is why exhorting people to be careful does not fix it.
What you can actually count is the tag. A numbered tag wired to the flange when the blind is fitted comes back off with the blind and returns to the board. A spreadsheet row marked removed is a sentence somebody typed. A tag in your hand is an object that was physically at that flange. Those are not the same class of evidence and they should not be treated as one.
The practice worth arguing about: “the blind list is the permit’s problem.” The permit closes when the isolation is lifted. The record of what was lifted is the turnaround’s problem, and start-up is normally the first time anybody compares the two.
One signature covering a joint two crews touched eight days apart. The gap it hides is not the torque figure. It is everything that happened between breaking the joint and making it up.
Start with the number itself. The same stud, the same wrench and the same figure produce very different clamp loads depending on what is on the threads and under the nut face, because most of the torque is spent on friction and only what survives becomes tension. Dry and as-received against properly lubricated is a spread wide enough to matter on any joint you would bother recording. So a torque figure with no lubricant written beside it does not describe the joint — it describes a wrench setting.
The larger gap is the one either side of the number. A joint is broken on day three by a crew getting access, blinded on day four by somebody else, left open through a week of weather with the studs in a bucket, and made up on day eleven by a night shift who never saw it open. A single line reading made up and torqued spans all of that and reports none of it: not whether the faces were inspected, not what gasket went in, not whether those are the original studs.
And the joint that leaks on start-up is usually not a joint that was torqued badly. It is a joint nobody recorded breaking. Scaffolders, riggers and blind crews break joints for access, those joints never reach the register, so they never reach the make-up list, are never tested, and announce themselves when the unit comes up.
The practice worth arguing about: “it was torqued, I watched them.” Attendance is not a record, and the man who watched is three states away by the time it weeps.
A hold point and a witness point are two different obligations. Treating them as one is invisible until a reviewer lines the dates up and finds the signature dated after the insulation.
A hold point stops the work. Nothing proceeds past it until the nominated party attends, or releases it in writing. A witness point does not stop the work. The party is given notice, may attend, and if nobody comes within the notice period the work proceeds and no obligation has been breached — provided you can show the notice was given.
That proviso is the whole mechanism. It means notification is a separate field from attendance, and a witness line carrying only a signature proves attendance and nothing about notice. Worse, a witness line with no signature and no notice recorded is indistinguishable from an item that was simply skipped. The honest case and the bad case produce the same blank.
The failure has a signature of its own. A final visual signed on a date later than the insulation contractor’s date on the same line. The work may well have been perfect. The record now says the examination was performed through cladding, and no amount of explaining afterwards puts that back.
The owner’s reviewer does not re-inspect anything. They read down the ITP column and ask four questions: was this an H or a W; if H, is there an attendance or a written release; if W, is there a notification time; and do the dates run in an order the plant could physically have been in. Most rejected packages fail on the fourth.
The practice worth arguing about: “we told them on the radio.” A verbal notice that was never logged is the same as no notice at all, and it is the contractor who needs it to exist, not the owner.
The certificate is in a folder and the reading is on a sheet. Nothing joins them until somebody compares two dates — which is the cheapest check an auditor has, and the first one they run.
An expired gauge does not invalidate one test. It invalidates every test it recorded back to the last date it was known good, which is normally its previous calibration. One sticker can therefore take out a fortnight of tests across several systems at once, and it is found at exactly the point where those systems are drained, insulated, painted and boxed up.
Two gauges of different ranges is the field control, and the reason is simple: a gauge that has quietly drifted reads exactly like a gauge that has not. There is no way to tell from the needle. A second gauge of a different range and preferably a different make does not share the failure, so disagreement between them is information you can act on before the pump comes off.
The fix is where the calibration data lives. The gauge serial number and the calibration due date belong on the test sheet, beside the reading, written in the field at the time. A certificate in a folder is evidence about a gauge. A serial number on the sheet is the only thing that connects that gauge to this test. Without it the folder proves nothing at all about the test in question, however thick the folder is.
The same applies to everything else on the test that produces a number somebody will rely on later: the chart recorder, the dead weight tester, the temperature instrument if the test is temperature-corrected. Each one is a separate serial number and a separate due date, and each is capable of invalidating the test on its own.
The practice worth arguing about: “it is in the folder.” The folder proves a gauge was calibrated. It does not prove that gauge took this reading.
Every weld examination there is will pass a perfectly sound weld in entirely the wrong material. NDE looks for discontinuities. Nothing in it asks what the metal is.
Radiography, ultrasonics, penetrant and magnetic particle are all looking for flaws. A sound weld joining a carbon steel elbow into a chrome line has no flaws. It passes, correctly, on every method, and the report is honest. The examination regime was never designed to answer the question and does not pretend to.
Receiving inspection is where most people believe this is caught, and it half is. Reading the MTR is reading a document about a heat of steel. It does not establish that the item on the rack came out of that heat, that the stencil on it is right, that it did not lose its marking to a wire brush, or that a fitter under pressure at two in the morning took the nearest elbow off the wrong pallet. Reviewing the paperwork is not the same activity as reading the item. PMI reads the item.
Do the weld as well as the base metal. Wrong filler is at least as common as wrong fitting and considerably harder to find afterwards, because the wrong deposit is sitting between two components that are both entirely correct and both stencilled to prove it.
The cost curve is the argument. Caught at receiving, it is a fitting. Caught after the weld, it is a cut-out, a re-weld, fresh NDE and possibly heat treatment again, at the point in the schedule where there is no float left. Caught in service, it is a component thinning or cracking at a rate the run either side of it does not — in a location nobody is watching closely, because the monitoring plan was written for the material that was supposed to be there.
The practice worth arguing about: “it came out of the alloy cage.” Cages are restocked by people in a hurry, at night, from pallets that arrived mixed. The label on the cage is a claim about the cage.
Nothing leaks and nothing fails. The tower simply does not make spec, and the reason is sitting behind a manway that will not be open again for four years.
The liquid on a tray is shallow — a modest depth over the weir, no more. Tilt the tray and the liquid runs to the low side: the crest over the weir goes uneven, the high side runs thin, and vapour does what vapour does and goes up through the path of least resistance rather than through the liquid. Contact is lost across part of the deck. The tray still works. It just does not work as well as the design assumed, and neither do the ones above it.
You buy that back every day, in reflux and reboiler duty, or you sell a poorer cut. Either way it is a running cost with no event attached to it, which is why it is normally diagnosed months later by somebody modelling the tower and finding fewer stages than the drawing says there are.
The tolerance is the tower vendor’s number, it is small, and it is measured against a datum rather than judged by eye. It also does not survive the rest of the job by itself: trays move when the clamps are pulled up, and they move when people stand on them to reach the one above.
“Internals installed” is not a record. A record is tray by tray: panel identification, level readings at the points the vendor nominates, taken after final tightening and not when the deck was set; hardware issued against hardware fitted; valve or cap count; weir height and downcomer clearance measured rather than assumed; clamps checked. Then the box-up sign-off, by two named people, saying what was in the vessel and what came out of it.
That last one is the one that matters at three in the morning on day nineteen. A torch, a bolt bag, a length of scaffold board. Vessel entry closes with a count, and the count is a record like any other.
The practice worth arguing about: “we levelled it when we set it.” The reading that counts is the one taken after the last man is out and nothing further will be tightened or stood on.
“Extra work” is one bucket holding four different things. They have different owners, different remedies and very different arguments at the end of the job.
Four categories, and the boundaries are not subtle. Planned is what was on the approved worklist at scope freeze. Discovery is what was found because a piece of equipment was opened and looked at — expected, budgeted for, the reason the outage exists. Emergent is work the execution created: damage, rework, a weld cut out, a failed test. Owner-added is what the owner put in after freeze.
Held as one bucket, growth tells you only that it is growing. Split, the same growth is diagnostic, because each category indicts something different. Discovery climbing says the inspection assumptions going in were wrong — an argument about the pre-turnaround inspection scope, and not about the crews. Emergent climbing says execution is manufacturing its own work, which is supervision, sequence or quality. Owner-added climbing says scope control has failed, and that is a commercial conversation which is cheap in week one and ruinous in week four.
The categorisation has to be made when the job is raised, and this is the part that gets skipped. Nobody re-categorises three hundred work orders honestly at the end of a job. The only person who reliably knows why a job exists is the person who raised it, and they know it for about an hour.
The turnaround manager is asked for this number every morning before shift change, and the answer drives the schedule forecast, the decision to call extra crew, and the date given to operations. If the answer is a feeling, so is the date.
These are field observations, not code. Where a number matters — a test pressure, a torque figure, a tray tolerance, a preheat — take it from the edition of the code your job is bought to, the job spec, the WPS in your hand and the vendor manual for the item in front of you. The equipment’s own documentation outranks anything general, this page included.
Every note above turns on the same thing: a record made at the time, in the field, by the person who did the work, carrying the two or three extra fields that make it checkable a year later by somebody who was not there. That is what this app is built to record.
Field notes — turnaroundqc.com/field-notes