Guinea: a 3,000 t/d gold operation held at around 93% recovery

This is the case where the integrated model is easiest to see. Xinhai carried the engineering, the equipment supply, the construction and then the operation, providing management and key technical personnel on site.

At 1.2 g/t, the margin between a plant that works and one that does not is thin, and it is decided in the leaching and adsorption circuit. The reported result is an overall recovery held stable at around 93%, which is a function of grind size control, leach residence time and carbon management rather than of any single piece of equipment.

Cross-process coordination is what the operation contract actually buys: mining supplies the plant with feed of a known character, and the plant is tuned for it, instead of each side optimising separately.

Here's the awkward bit of any enquiry. You want a capital number and a schedule. What decides both is a body of test data that may not exist yet. So the honest first answer to “what would a plant like this cost” is a list of what we need from you.

Why gold punishes a single composite

Gold is harder on sampling than most ores, for two reasons that are specific to it. Coarse gold turns the sub-sampling and splitting step into a source of error in its own right — JORC's sampling criteria single it out as a case with inherent sampling problems — and oxide, transition and fresh material from the same deposit can behave differently in cyanide, in reagent demand and in grind requirement. Send separate charges for each weathering horizon and each ore domain rather than a blend that averages them away; what a representative programme needs in general is set out under test work and research.

The assay and test suite

• Head assay by fire assay, plus silver, copper, arsenic, sulphide sulphur and organic carbon. Those last four decide whether cyanidation behaves at all.
• Gravity-recoverable gold. Coarse gold changes the flowsheet and it also changes how your samples have to be split.
• Bottle-roll or diagnostic leach at more than one grind size — with reagent consumption logged, not just recovery. Consumption is what shows up in your operating cost.
• Oxygen demand and lime demand measured through the leach, not assumed from it. Both size the aeration and reagent plant, and both are cheap to record alongside a bottle roll.
• Carbon behaviour on your own leach solution: adsorption isotherm and loading kinetics. Carbon inventory, elution frequency and regeneration duty come off those curves, and they set the size of the adsorption train and the acid wash and kiln that sit behind it.
• Comminution indices measured on the same material, so the mill and the leach aren't sized off two different orebodies.
• Settling and rheology. Thickener diameter, filtration and tailings deposition all come out of this and nowhere else.

The site constraint that gets left out

The heaviest single lift your access route allows, the port handling limit, and the road weight limits between port and plant. That set decides how much of the plant is welded in a workshop and how much is welded in your yard. Leach tanks to 20 m diameter and mills to 7 m diameter sit inside Xinhai's own fabrication range, described under equipment manufacturing, but whether a given vessel travels whole or in plate is settled by your road, not by the workshop. Decide the fabrication split while the layout is still on paper: it moves site welding hours, crane and rigging requirements and the construction sequence, and none of the three is cheap to change once steel is ordered.

Xinhai reports roughly 200 test programmes a year across more than 70 ore types, run in a CNAS-accredited laboratory, and the same house feeds the design. The discipline coverage that handover reaches into sits on full lifecycle mine design.

Four errors are worth naming. None of them are exotic, and every one of them is decided long before the first foundation is poured.

Watching the mean instead of the spread

A thin margin at this grade is real. What breaks plants at 1.2 g/t, though, is not the mean — it's the spread. A deposit averaging that figure with tight metallurgical behaviour is an easier plant to run than a higher-grade one whose recovery swings by domain, because the second one forces you to design for a case you cannot predict. Ask what the variability looks like, domain by domain, not what the average is. The reporting codes already put that question in writing, which means the material often exists before anyone thinks to ask a plant designer for it. In the JORC Code, Table 1 Section 1 “Sampling Techniques and Data” covers sample preparation, sub-sampling quality control and the quality of assay data, and its criteria apply to every later section; Section 4, on Ore Reserves, lists among its criteria the nature, amount and representativeness of the metallurgical test work, the metallurgical domaining applied and the recovery factors that go with it — to be addressed on an “if not, why not” basis. Canada's Form 43-101F1 asks for the same ground at Item 13: where metallurgical testing has been carried out, a technical report discusses the nature and extent of it, the basis for any assumptions or predictions regarding recovery estimates, and the degree to which the test samples are representative of the deposit as a whole. Ask for whatever a project already holds under those headings before you ask anyone for a capital number.

Treating nameplate as production

A 3,000 t/d design figure is a design figure. Annual tonnes are that number multiplied by availability and by utilisation, and the gap between the two lives in crusher downtime, wet-season feed handling, power interruption and spares lead time. Size the surge capacity and the spares holding against that gap, not against a perfect year.

Buying the flowsheet and forgetting the reagent chain

Cyanide is not a line item. It's a supply chain with its own storage, handling, training, emergency response and decommissioning obligations, all of which an international code already spells out, and on a remote site with a wet season it's also a stockholding decision. Lime, grinding media and fuel behave the same way. On a site like that, a washed-out road can cost more production days than any single equipment failure, and the fix is inventory policy rather than engineering.

Assuming an operating team appears

Recruitment and training run longer than the construction tail, and they are the part of the programme that rarely appears on the bar chart. Decide the staffing model before you fix a commissioning date, because that decision sets three things you cannot retrofit late. The training plan, and how far ahead of first feed it has to start if operators are to learn the circuit on the circuit rather than on the ore. The recruitment runway for whichever roles are scarce near site — control-room operators, leach and reagent supervisors, maintenance planners — each with its own visa, medical and induction lead time. And the spares holding, which a crew still learning the plant consumes faster than a settled one, so a first-year list built off steady-state consumption will be short in exactly the months you can least afford it. The cooperation models available for the staffing side are on contract mining and operation.