Certificate · Module 4 of 6

This is the module the Certificate is really about. It is the hardest to fake, the most valuable to have externally validated, and the one an employer will ask about. The free Foundation's testing lesson ends by admitting that density cannot rule out a tungsten core. Everything below is the answer to that sentence.

It also makes two corrections to things the trade teaches almost universally. Both are more valuable than the claims they replace.

Correction one: fire assay is not "the referee method", universally

The standard framing is that cupellation is definitive and everything else is judged against it. The primary literature does not support that at high fineness.

• ISO 11426:2021 — Jewellery and precious metals — Determination of gold — Cupellation method (fire assay) — scopes the method at 100 to 999.5 parts per thousand. Above 999.5‰ the standard itself directs you to spectroscopy.
• Metalor Technologies, presented at the LBMA Assaying & Refining Seminar, 7–8 March 2011: on Au 999.99‰ proof samples, uncertainty bias was 13 ppm for cupellation against 2 ppm for ICP-OES.

Within its range, cupellation is superb — published repeatability of ±0.04 to ±0.09 parts per thousand. Say it accurately: the long-established reference method for gold in the 100–999.5‰ range, codified in ISO 11426. Above that, instrumental methods are more accurate and the standard says so.

This is the first time in the programme I will tell you that something you will read everywhere is wrong. It will not be the last. A practitioner who says "fire assay is the referee method" without the qualification will eventually say it to someone who knows.

Correction two: there is no single "XRF reads N microns" figure

Bruker's own definition of information depth is "the thickness of the top layer from which a significant proportion (often 90%, 95%, or 99%) of the detected fluorescence originates" — it varies with the element, the emission line, the matrix and the instrument. The definition itself carries a convention choice.

The concrete anchor worth having: in Bruker's M6 Jetstream application note, working on Au-Lα, "the gilding must be thinner than 9 µm, as we can detect characteristic line of Ca and Fe below the gold layer."

Teach the range, not a point value: roughly 1–10 µm for gold alloys on Au-L lines. Treat a spec sheet claiming 40 microns of penetration in gold with suspicion — that number is doing marketing work. The conclusion is unchanged and stronger for being honest: XRF cannot see through plating and cannot see a tungsten core, and now you can say why rather than repeating it.

Sampling error is the dominant error term

The single most under-taught fact in the trade. A perfect instrument reading an unrepresentative sample gives a precise wrong answer — and almost every dispute in the scrap trade is a sampling dispute wearing an instrument's clothes.

Precision is repeatability. Accuracy is closeness to truth. A miscalibrated instrument is precisely wrong every time, which is far more dangerous than being obviously erratic.

An item has one composition. A parcel has as many as it has pieces — which is why sorting into visually consistent lots before sampling is the standard mitigation, and why a supplier who limits destructive testing to 2% of a parcel is imposing a real constraint on the confidence you can reach.

The tungsten arithmetic

Run it properly, with the measurement error carried through.

The test has ruled out a great deal. It has not distinguished the one pair you were worried about, and no amount of care with the same instrument will.

The best thing in this module

Published reference values give gold's density as both 19.30 and 19.32 g/cm³, depending on the source and the temperature assumed.

Before you have touched a balance, before any measurement error at all, the disagreement between two reference books about the density of gold is forty per cent of the difference you are trying to detect.

That is not a reason to abandon the test. It is a reason to understand exactly what the test can and cannot conclude — which is the whole subject of this module.

And note that karat densities are ranges, not constants. 18k gold is 14.99 g/cm³ copper-alloyed and 15.96 silver-alloyed: a 6% spread at the same karat. Always state the assumed alloy when you use one.

Two cheap orthogonal tests beat one expensive test

Gold and tungsten are almost identical on density. They are not remotely similar on anything else.

Conductivity: MIT 6.777J Material Property Database; ITT Reference Data for Radio Engineers, 6th ed. Velocity: Dakota NDT and Selfridge, two independent sources agreeing within 0.3%.

Ultrasonic velocity is the strongest single discriminator in the toolkit. Tungsten is about sixty per cent faster than gold acoustically, against a quarter of a per cent on density. That ratio is this module in one line — and it is why ultrasonic testing catches tungsten cores that scales and XRF both miss.

The design principle follows: choose the next test because it closes the specific gap the last one left open, not to add general reassurance. A step whose result would not change what you do next is decoration.

What the documented cases actually show

New York, 25 September 2012. Ibrahim Fadl of Express Metal Refining on 47th Street found four 10-ounce PAMP Suisse .9999 gold bars — carrying legitimate serial numbers — "hollowed out and filled with tungsten," then "expertly refinished to hide the deception." Tungsten was roughly 75% of bar content. Investigated by the US Secret Service, New York field office. Reported by Coin World, 3 October 2012.

Also documented: a Metalor 1 kg bar in the UK in 2012, drilled and filled with tungsten rods; and a 500 g tungsten-filled bar identified by Heraeus in Germany around 2010.

In the New York case the serial numbers were legitimate. The packaging is now the attack surface — a convincing card around a fake bar is cheaper to produce than a convincing bar. If your protocol authenticates the documentation and infers the metal, you have tested the wrong object.

Designing a protocol you can defend

A defensible protocol has four properties, and they are what a professional is actually assessed on.

1. Method selection appropriate to the item, the budget and the time. Nothing included for show.
2. Sequencing that is argued, not listed. Cheap-and-broad before expensive-and-narrow, and each result changes what happens next.
3. Blind spots stated specifically. "XRF has limits" is worth nothing. "XRF reads roughly the top few microns on Au-L lines, so it cannot see a core under this plating" is worth full marks.
4. A residual risk statement. What remains open after every step you have specified, and at what cost that gap could be closed. This is the criterion that separates a pass from a distinction, and "some risk remains" scores nothing.

What this module cannot tell you

It cannot tell you that an item is genuine. No protocol can. What it can do is let you state, precisely and defensibly, what you have ruled out, what remains open, and what closing that gap would cost — which is the only honest form a professional opinion takes.

Kept in this browser only — no account, and nothing is sent anywhere.

Educational content only. Not financial, investment or tax advice. Aurix does not buy, sell, store or broker metal, and does not quote executable prices.