Controls the provider and the lab each supply

The symptom is a tidy report in which every sample worked and none of the controls you care about are mentioned. Either the controls were never supplied, or they failed and left without a sentence. Both versions make the report hard to interpret. The repair starts by naming two owners. The laboratory owns experimental controls the provider cannot invent. The provider owns process controls, and they should name the metric. Commissioning the study as a whole is set out in commissioning a sequencing or proteomics study. Writing the lines so a quotation can answer them is how to write a laboratory sourcing enquiry. A scope can be discussed through laboratory procurement or academic research, and sent by quotation request or contact. A sourcing conversation does not mean the desk runs the assay.

Two owners, two kinds of failure

An experimental control asks whether the biology is the biology you think it is. The right tissue, the treatment that really happened, a biological replicate from an independent animal or culture, a vehicle control, a knockout or a known positive sample you prepared: these exist only if you made them. A provider who never saw the animals cannot reconstruct them from a FASTQ header.

A process control asks whether the provider's workflow behaved. A library quantification above a threshold they publish in the statement of work, a spike-in they added at a stated amount, an internal standard in a mass-spectrometry batch, a blank extraction carried through their chemistry: these are theirs to define and to report. You can require that they name them. You cannot honestly require a metric they have not stated, and then punish them for missing your private number.

The symptom table at the end separates "we cannot tell if the biology is real" from "we cannot tell if the process ran". The next check follows the owner.

What you must send

Send the design, not a pile of tubes with hopeful caps. Each tube's label carries the group, the biological replicate identity, and the sample type. If the contrast is treated versus vehicle in mouse liver, the vehicle animals have to be in the box, and the tissue has to be liver. A neighbouring brain piece is not a near enough control. It is a different experiment.

Biological replicates stay independent. Four extractions of one culture are technical replicates. Say which kind you sent, so nobody upgrades them in a caption. If a spike-in or a control RNA is part of your experimental design rather than their process, you supply it and you say so. If it is theirs, they supply it and they say so. The enquiry should not leave the word control floating between you.

Include the fail rule you will live with. Example: if a biological group loses a replicate before sequencing, you will not interpret that contrast, and you will not quietly proceed with two versus four. Write the rule while the design still feels abstract. It is much harder to write after a pretty plot arrives with a hole in it.

Method collections such as Addgene's protocols and protocols.io show controls inside bench workflows. Use them to remember which controls are experimental. Do not paste a protocol into the statement of work in place of your own groups.

What the provider must name

Ask for one paragraph in the quotation. Which process controls will they run? What metric is a pass? What happens on a fail? A library QC metric might be a molarity or a size profile they define. A proteomics batch might include a quality injection or a spiked peptide they define. The exact threshold is theirs to justify, then yours to accept before the samples ship. Once accepted, it is the rule. Moving it after you see which samples fail is a different, worse practice.

They also report the metric per sample, not only a batch average that hides one empty library. A report that says "QC passed" without the metric is the symptom you started with. Send it back for the numbers that were agreed.

A worked failure on an expression study

You sent eight RNA samples: four treated cultures and four vehicle cultures, each from a separate flask. The provider's written process control is a library metric they named, plus a spike-in they add to every library and score. Your experimental controls are the four vehicle flasks and the independence of the flasks. Nobody is allowed to invent a fifth flask.

The report arrives with seven samples and a figure that still says four versus four. The missing sample is one vehicle, and its library metric is absent rather than marked fail. This is the quiet drop. Stop. Ask for the metric on all eight. If the missing library failed their threshold, the pre-written rule applies: you do not interpret the contrast as four versus four, and you do not let the figure keep the old caption. You either repeat the library under a new line or you reinterpret a study that is no longer balanced, with the loss shown.

Suppose instead the libraries all pass and your notebook shows that two "treated" flasks never received the treatment. That failure is yours. The provider's process control cannot see it. You do not ask them to relabel the FASTQ files. You correct the design, and you do not publish the contrast those labels implied.

Suppose the spike-in they own fails in one batch and passes in another. The batches are not comparable under their own rule. Stop the cross-batch sentence. Reinterpret inside the batch that passed, or repeat. Do not scale the failed batch by a factor you invent so the plot looks smooth.

What "stop and reinterpret" looks like

Stop means the affected samples leave the claimed result. Reinterpret means you write what study you still have. A balanced eight may have become a repeat-the-library problem, or a smaller study with a named hole, or no study until new cultures grow. Quietly dropping the sample is the forbidden branch: the file disappears, the caption stays, and the control that would have changed the sentence is gone. The statement of work should say the forbidden branch is forbidden.

Record the failure next to the metric. Future you will need to know whether the hole was biological or procedural, because those holes are repaired differently. One needs new samples. The other needs a new library or a new injection under the provider's process.

Failure modes beyond the missing row

The first is a shared, blurry control. You each thought the other added the blank. The blank was never run. Name it on one side in the enquiry.

The second is a metric invented after the failure. The provider lowers a threshold so a weak library passes, or you decide a replicate "does not count" because it weakens the story. Both moves are quiet drops in slower clothes. The threshold and the replicate rule had to exist earlier.

The third is a process pass used as a biological pass. Libraries can be excellent and the treatment can still be wrong. Read the owner column before you celebrate the QC table.

Safety and research limits

Controls do not set biosafety level. The institution does, for both your laboratory and the provider's handling of material you send. The CDC BMBL is a public biosafety reference, not a permit for the parcel. Research controls are not diagnostic positive and negative controls unless you are in a regulated diagnostic system, which this note does not describe. Do not borrow clinical words for a culture experiment.

Ordinary trouble before the parcel leaves

Heat and a delayed courier can degrade the RNA that was your only treated replicate. That is an experimental-control failure in transit, and the pre-written rule still applies: you do not pretend the replicate arrived. A power cut in your freezer is the same class of problem. The provider's library metric will eventually notice a ruined sample, which is useful, and it is late. Your own check before shipment is part of supplying the control. Do not add a guessed transit time. Do specify the temperature class and the receiving rule, and do not ship a design you no longer hold.

What the enquiry states

State the experimental controls you will supply, including replicate type and the fail rule. Ask the provider to name process controls, the metric, and their fail rule. State that a failed control stops the original claim and is reinterpreted in writing, and that samples are not dropped quietly. Place this beside the sample count and the file types, through laboratory procurement or academic research. Use contact or a quotation request. A scope can be discussed. When the report arrives, find every control in it before you read the conclusion.