Identity checking is the one control performed on every single container. It is also the one whose rejection threshold repays the most attention.
The laboratory check gives the value that counts. A spectroscopic identification returns the answer in seconds, without consuming the sample, sometimes without opening the container, and it records the evidence.
Three questions not to confuse
- Is it the right material? That is identification: a library, a threshold, a binary decision.
- Is it compliant? That is a content or a property measured against its specification. It then takes a model, not just a library.
- Will it behave in your process? A material compliant with its certificate can still behave badly in production. That question is settled in feasibility, on the batches that caused trouble. Feasibility.
The question that decides how well it holds in routine is asked early: at what value do you refuse a container, and on what data was that threshold established?
Depending on your sector: excipient drums in pharma, plant butters and oils in cosmetics, polymers to sort in chemicals, grain at the tipping pit in agrifood.
The texts by sector: pharma (USP <197>, Ph. Eur. 2.2.40, ICH Q2(R2)), cosmetics (ISO 22716), food and feed (ISO 12099, 852/2004). The detail.
Show us your materials and your containers: forty-five minutes is enough to say what can be identified. Book a call.
The texts by sector, and what Annex 8 requires
Identity checking works the same way everywhere: a library, a threshold, a binary decision. What changes from one sector to the next is the text that says what this control must demonstrate, and to whom.
- Pharma and drug substance — the spectroscopic identification test is USP <197> and, for near infrared, Ph. Eur. 2.2.40; validating the procedure falls under USP <1225> and ICH Q2(R2), whose section 3.1.2.1 deals specifically with identification tests.
- Cosmetics — incoming material control falls under cosmetic good manufacturing practice, ISO 22716, within the frame of Regulation (EC) No 1223/2009.
- Food and animal feed — applying near infrared falls under ISO 12099, official control under Regulation (EC) No 152/2009, and hazard control under your food safety management plan pursuant to Regulation (EC) No 852/2004.
“Is this container really what the label says? Can I check a hundred per cent of the containers? How long does my material stay on hold?” Annex 8 of the EU good manufacturing practice guidelines puts it this way: the identity of a complete batch of starting material can normally only be ensured if samples are taken from all the containers and an identity test performed on each. A derogation exists — sampling only a proportion of the containers — but it rests on a validated procedure showing that no single container is incorrectly labelled, which is rarely simpler than measuring them all. So these three questions come up on every delivery.
Annex 8, Sampling of Starting and Packaging Materials, §§ 2 and 3, EudraLex Volume 4.
What is actually measured
A resemblance, not an identity
The instrument compares the acquired spectrum with the entries of a library and returns a match index. It is a distance, turned into a percentage or a score, to which someone has to attach a decision.
The library is the real product
An instrument only identifies what it contains. The libraries supplied cover the common families. As soon as you move away from the most widespread materials, the missing entries have to be built: acquisition of reference spectra, qualification, versioning. It is method work, not a purchase.
Two close materials stay close
Some pairs (neighbouring fluorinated polymers, different grades of the same excipient) give very similar spectra. The ability to tell them apart is a property of the pair, not a performance of the instrument. It is verified experimentally, on your materials, before purchase.
The rejection threshold is almost never supplied with the instrument. Documentation shows match indices with a colour code, red, yellow, green, and bounds given as a guide. It is a reading aid, not an acceptance criterion.
A defensible acceptance criterion is established on true negatives. The close materials you really hold in stock, the same materials from other suppliers, the neighbouring grades that could be delivered by mistake. You measure the distribution of indices for the true positives and for the true negatives. Then you place the threshold where the separation is demonstrated. Then you document it, version it with the library, and review it when the library changes.
It is the work that sells least, because it does not come in a box.
Which technology for which case
| Technology | What it follows | Suitable matrices | Calibration effort | Installation | Known limit |
|---|---|---|---|---|---|
| NIR spectroscopy | Overall signature of the matrix | Powders, granules, excipients, material in a thin bag | Library + classification (model calibrated on samples) | Handheld in contact, or a fixed station in the goods-in area | Weakly discriminating between very close compounds. Sensitive to moisture and particle size |
| RAMAN spectroscopy | Molecular signature, chemical bonds | Solids, liquids, polymers, active ingredients | Library + match index | Handheld in contact, or through a transparent container | Fluorescence of some matrices. Heating of dark materials |
| TERAHERTZ | Solid-phase signature, without opening the packaging | Packed solids, non-metallic packaging | Library + classification | Contact-free, no radiation protection | Published results obtained in the laboratory on pure substances. Strong spectral distortion through packaging |
| MIR spectroscopy | Composition of a liquid, content of a constituent | Liquids, solvents, organic media | None: decomposition onto pure components | Short optical path cell | Liquids only. Water absorbs strongly |
The choice between NIR and RAMAN is rarely made on raw performance. It is made on the matrix: RAMAN is more selective on bonds, NIR more tolerant of heterogeneous materials, and on the container.
Checking without opening: what each container allows
| Container | What it is realistic to expect |
|---|---|
| Thin, colourless polymer bag | Can be crossed in NIR in many cases, with a correction. To be validated on your real film, not on a generic one. |
| Clear glass vial or ampoule | Natural ground for RAMAN: glass contributes little to the signal. It is the strong argument for checking without opening in a sterile area. |
| Cardboard or opaque plastic packaging | The domain of TERAHERTZ, which goes through most non-metallic materials. The counterpart is significant spectral distortion, to be handled in signal processing. |
| Metal drum or container | None of the four technologies goes through metal. It has to be opened. The question becomes that of the sampling device, not of the sensor. |
What is worth preparing on your side
- The real list of materials to check, with their suppliers and their grades. It is what says how many library entries are to be built, and it is the first item of effort on the project, well ahead of the hardware. What a calibration costs.
- Samples of true negatives: neighbouring materials, other grades, other suppliers. They are what turns a threshold into a justified figure rather than a habit.
- Several batches of each material, and where the product is of natural origin, batches from different seasons. An entry that covers the real variability is the one that keeps accepting conforming material.
- The real container, with its film, its label and its dust. A trial run on material decanted into a laboratory bottle describes the laboratory bottle.
- The data access procedure: who triggers a measurement, who sees the result, how it is recorded, what follows a rejection. It is what turns measurements into decisions.
Conditions for success and limits
Identity and conformity are two different tests
A high match index says “this spectrum resembles that of lactose”. It does not say that the material conforms to its specification, nor that it is pure, nor that it is the right grade. Identification replaces an identity test. It does not replace the quality tests. Confusing the two is the most expensive error in this field, because it goes unnoticed until the day it no longer does.
Dark materials call for reduced power
In RAMAN, a dark material absorbs the excitation light instead of scattering it. The surface heats, can melt or burn, and leave residues on the optics. Instruments provide a reduced-power mode, and their documentation explicitly requires it. It is an integrity instruction, for the sample as for the instrument. The right question to a supplier is therefore “how does your instrument handle black materials”. The answer must be a mechanism, not an adjective.
Ambient light is part of the protocol
On a handheld instrument used in a goods-in area, the building lighting enters the analysed zone. Fluorescent tubes are the most troublesome, because their emission lines overlap the signal. The remedy is mechanical: shrouds, dedicated sample holders, and a procedure that requires their use. A trial that succeeds on a bench and fails on a loading dock is almost always this.
Reading the standards cited on a data sheet
Some product sheets cite 21 CFR standards that relate to laser safety, not to data integrity. The useful question is “which compliance matrix can you show me, and on which software version”. What a declaration of conformity covers, and what it never covers.
What it changes, in practice
It is the application whose return on investment is easiest to establish, because the cost items are visible.
The receiving sequence changes shape. Identification is done at the dock, through the packaging in many cases, without sampling in a sampling room and without waiting for a laboratory slot. What is gained is not a faster analysis — the laboratory is already fast — but a step removed from the critical path, and quarantine space given back.
The QC laboratory recovers capacity. Identity often accounts for a large share of the test volume; that capacity shifts to the analyses that really call for its expertise.
And testing becomes exhaustive: one hundred per cent of containers. It is the rule set by Annex 8 — “The identity of a complete batch of starting materials can normally only be ensured if individual samples are taken from all the containers and an identity test performed on each sample” — and it changes the nature of the check against a substitution risk.
The same paragraph allows an exception, and that is the one to work up. Annex 8 permits not sampling every container where a validated procedure justifies it, and its § 3 states the conditions: supplier reliability, quality assurance system, manufacturing conditions. Knowing which side of that line each of your materials sits on is the first deliverable of an identity project, and it is built before the choice of instrument.
The counterpart is clear. Most of the effort lies in the library and in the thresholds, not in the instrument. An identity project that budgets the instrument and not the method building ends with an instrument in a cupboard.
Frequently asked questions
Can we really check through the bag?
Often, and it is demonstrated on your film and your material. Three factors decide: the thickness of the film, its pigment and additive load, and the real contact between film and powder. A bag held loosely leaves a layer of air that dilutes the signal. It is a short trial, best done before the instrument is chosen.
How many spectra does a library entry take?
Enough to cover the variability the material will really present: several batches, several containers, and the measurement conditions of the floor. An entry built on three spectra of one bottle performs beautifully in a demonstration. Covering the real range from the start is what keeps it accepting conforming material two years later.
What happens when a material is absent from the library?
The instrument leaves it unidentified, which is the safe behaviour. Reference spectra are then acquired on samples whose identity is established by another route, sometimes on a bench instrument to obtain a spectrum of sufficient quality before carrying it into the handheld library. It is a routine operation, and it is exactly where an integrator earns their place.
How is the rejection threshold justified in an audit?
With data rather than with a factory setting. The expected file holds the list of true negatives tested, the distribution of indices obtained for positives and negatives, the threshold retained and its margin, the associated library version, and the review rule when anything changes. It is a method validation deliverable, and it is prepared alongside the library.
Does the instrument meet data integrity requirements?
The instrument can. A system reaches compliance through the operator’s procedural controls as well: account management, training, procedures, administration. Manufacturers state this themselves in their declarations, and it is to their credit. The work between the two is what a validation plan covers. 21 CFR Part 11 and validated systems.
Handheld or fixed station?
The handheld goes to the material, which removes decanting and sample transport, decisive in a goods-in area and on heavy containers. The fixed station offers better measurement conditions, controlled shrouding and simpler integration with the information system. Many sites end up with both, and the split is decided on the flows.
The libraries we supply, and what they run on
Identifying is not assaying: a library is judged on its ability to separate close products, not on a prediction error. We supply identification libraries in RAMAN, on the handheld Raman from Enwave.
A supplied library is a starting point rather than a finished deliverable: it is checked against your references and completed with your materials before it can be relied on, particularly on the pairs you know are hard to separate. The same split as for a quantitative calibration.
Send us your list of materials. You will know which ones identify cleanly and which ones call for library work.
Forty-five minutes is enough to pick out the straightforward materials, the close pairs worth testing early, the containers that call for another route, and to size the library work. You will leave the call with a clear view of all four.