Screening Devices and the ASSURE Program

A screening device is not a small laboratory in a box. Its task is narrower: to rapidly separate samples that meet a predefined criterion from those that require further testing.

This distinction is critical in the modern diamond trade, especially for melee goods and large batches, where it is not economically realistic to subject every stone immediately to a complete laboratory workflow.

A property tester and a DVI are not the same

The handheld thermal or electrical tester from Chapter 67 measures a limited physical property. A Diamond Verification Instrument (DVI) may use a more complex combination of optical, spectroscopic, or imaging signals and algorithmic interpretation.

But the name DVI still does not mean that the device definitively determines every aspect of a sample. Its actual scope is defined by:

• the sizes it accepts;
• the shapes it accepts;
• loose or mounted stones;
• an individual stone or a parcel;
• natural diamonds, laboratory-grown diamonds, and/or simulants;
• automated or operator interpretation;
• the validated hardware and software version.

PASS, REFER, and DETECT must be read literally

Different devices use different labels. Their typical meanings may be:

• PASS — the sample falls within the criteria that the device classifies as a diamond/natural-diamond pass within its own scope;
• REFER — the result is insufficient to complete the case, and the sample requires further testing;
• DETECT — the device has classified a specific target category within its declared scope;
• OUT-OF-RANGE — the sample is outside the validated range;
• TEST AGAIN / unreadable — the measurement was unacceptable.

The most important rule is:

REFER is not the same as laboratory-grown.

The referral population may contain natural diamonds, laboratory-grown diamonds, simulants, or a technical problem—depending on the device.

[VISUAL 73.1: Screening decision tree—PASS / REFER / DETECT / OUT-OF-RANGE]

Loose, mounted, and melee require different validation

A device tested on loose round stones is not thereby automatically validated for mounted jewelry. A setting can obscure part of a stone, change the optical geometry, or limit manipulation.

Melee introduces another problem: a large number of stones, small mass, and the need for throughput. A system with excellent analysis of a single stone can be entirely impractical for ten thousand melee stones per day.

Automatic feed, interpretation, and dispense are not synonyms

Automation may encompass different functions:

• automatic stone feeding;
• automatic measurement;
• automatic interpretation;
• physical separation of results into different output bins.

A device may have automatic feed yet require an expert to interpret the image. Another may automatically interpret an individual stone without automatic feed. These functions must be stated separately.

What ASSURE is

The Natural Diamond Council’s ASSURE Program does not certify diamonds. The program organizes standardized performance testing of diamond verification instruments, and the official ASSURE methodology states that instruments are tested with a verified ASSURE sample at an independent third-party laboratory. Results are published for each specific device model.

As of the factual cutoff of August 8, 2026, ASSURE 2.0, launched in 2022, is active, and testing is still ongoing. The Directory is updated as new results become available. Results from the ASSURE 1.0 archive must not be mixed with ASSURE 2.0 results.

The ASSURE Sample is not the market

The ASSURE Sample intentionally includes natural diamonds, laboratory-grown diamonds, and, when relevant, simulants, including challenging specimens.

It is a challenge set. It is not a representative estimate of the share of natural and laboratory-grown goods in a particular trade, country, or market channel.

Performance on the ASSURE sample therefore answers the question:

“How did this instrument perform on this defined test set?”

It does not directly answer the question:

“What is the probability that a stone from my store is laboratory-grown?”

Three key ASSURE 2.0 metrics

ASSURE 2.0 specifically publishes three metrics that must not be combined into a single marketing number.

Diamond False Positive Rate

The proportion of laboratory-grown diamonds and simulants that the device incorrectly classifies as a diamond/natural-diamond pass under the ASSURE definition. The optimal value is 0%.

Diamond Referral Rate

The proportion of actual diamonds that the device sends to referral instead of correctly passing them. The optimal value is 0%.

Diamond Accuracy

The proportion of actual diamonds correctly classified as diamond. The optimal value is 100%.

A low false-positive rate can coexist with a high referral rate. It is therefore not enough to advertise only one metric.

[VISUAL 73.2: The three ASSURE metrics—false positive / referral / accuracy]

Zero in a test is not a mathematical guarantee

If a particular model has a 0% Diamond False Positive Rate in ASSURE testing, this means that no false positive was recorded in that test sample and protocol. It is not proof that the device will never make an error on future goods.

The following remain relevant:

• the size and composition of the test set;
• uncertainty/repeatability;
• new growth products and treatments;
• the firmware and software version;
• the condition and cleaning of the instrument;
• the user and environmental conditions.

The novice operator is part of the ASSURE 2.0 test design

ASSURE 2.0 tests devices with a novice operator following the manufacturer’s standard training. This is an important design decision because it assesses performance closer to that of a real first-time user, not only a specialist familiar with every visual artifact of the system.

For devices that require operator interpretation, this difference may be particularly important.

Throughput is not accuracy

“Thousands of stones per hour” describes processing capacity under specific conditions. It says nothing about the false-positive rate, referral rate, or how the device performs with a different size or with mounted goods.

Procurement therefore considers the following separately:

• accuracy metrics;
• referral burden;
• actual throughput;
• acceptable sizes and shapes;
• mounted capability;
• operator requirement;
• service, calibration, and software updates;
• total cost of ownership.

Model, version, sample, and date define a result’s identity

Performance must not be transferred from one device to an entire brand. A professional record reads approximately:

model + hardware/software version + ASSURE test ID + sample category + date.

This is particularly important because a manufacturer may change the algorithm or optics without changing the basic commercial name.

ASSURE is not a ranking list

ASSURE does not exist to declare a universally “best device.” Different business workflows have different requirements.

A small jewelry store may prefer a slower device for individual mounted stones. A large sorting operation may require automatic melee feed. A laboratory may use multiple methods and accept a higher referral rate if it can readily resolve a referral using another instrument.

[VISUAL 73.3: Device-selection matrix—sample type / throughput / referral / operator / cost]

Referral governance must exist in advance

The worst screening workflow is one that has a device but no rule governing what to do with a referral result.

The SOP must define:

1. whether the measurement is repeated and under what conditions;
2. when the sample is cleaned or repositioned;
3. which other, orthogonal method follows;
4. when the sample is sent to an external laboratory;
5. how the sample’s identity is maintained;
6. who is authorized to close the case.

Retesting must not become “hunting for a PASS.”

Factual cutoff: August 8, 2026

The ASSURE Directory 2.0 is a living directory. The book therefore deliberately does not embed a large static table of all devices and prices in the main text. Such content belongs in an updatable web supplement.

The stable message remains:

an independent test result must be read according to the specific model, version, sample, protocol, and date.

[VISUAL 73.4: How to read an ASSURE record—model → sample → metrics → scope → date → referral workflow]

Chapter summary

• A DVI is a screening system, not automatically a complete laboratory.
• A handheld property tester and a DVI are not the same category of instrument.
• PASS, REFER, and DETECT apply only within the declared scope.
• REFER is not a synonym for laboratory-grown diamond.
• Loose, mounted, and melee require separately validated workflows.
• Automatic feed, interpretation, and dispense are different functions.
• ASSURE 2.0 independently tests devices using a standardized challenge sample.
• The ASSURE Sample is not a model of market prevalence.
• False Positive Rate, Referral Rate, and Accuracy must be read separately.
• A test result of 0% false positives is not a universal guarantee of future infallibility.
• Throughput is not accuracy.
• Model, version, sample, test protocol, and date are part of the identity of a performance claim.