SITELIVE 
What shipped, and what it has produced
PLS commissioned the P680 crushing and ore sorting facility at its Pilgangoora spodumene operation in Western Australia during the September 2024 quarter, telling the ASX that all outstanding items were completed on schedule. TOMRA Mining, which supplied the sorters, describes the installation as the largest lithium ore sorting plant in the world: ten machines — four COM Tertiary XRT units for fines, three COM XRT 2.0 units for mid-size particles and three PRO Primary Color units for coarse material — with sorting capacity above 1,000 tonnes an hour. The operating record is now three financial years deep and disclosed in regulatory filings rather than brochures. In the September 2024 quarter PLS reported lithium processing recovery of 75.3 per cent against 72.2 per cent the quarter before, attributing the improvement to plant stability, continuous improvement work and the use of ore sorting technology. In the June 2026 quarter it reported recovery of 76.8 per cent and noted that improved ore sorter performance provided continued operational flexibility, closing an FY26 in which production reached 879.5 kt, 17 per cent above FY25.
The mechanism: atomic density, one particle at a time
The dominant sensor in hard-rock sorting is X-ray transmission. Two X-ray energy levels are fired through each particle on a conveyor and the differential attenuation is resolved into a measure of specific atomic density — the same physics as an airport luggage scanner, applied at hundreds of tonnes an hour. Because it reads through the particle rather than off its surface, XRT is largely indifferent to dust, moisture and dirt, which is what makes it viable on run-of-mine rock. Particles are stabilised on a belt, scanned, then rejected in flight by banks of compressed-air ejectors: TOMRA's COM XRT 2400 2.0 carries up to 384 nozzles at a 6.25 mm pitch and is rated to 500 t/h across a 10–125 mm size range, while the finer COM Tertiary XRT covers 4–60 mm at up to 150 t/h. Both require the feed to be screened into narrow fractions with a maximum-to-minimum size ratio of roughly three to one.
The change is to the mine plan, not the plant
The reason sorting matters is upstream of the sorter. Rejecting barren host rock before the mill means the mill sees higher and more consistent feed, which in turn means the pit can be mined less selectively. PLS put this in writing in its June 2025 quarterly: the ore sorting facility continues to contribute to a reduction in mining costs by allowing a higher proportion of contact ore — a blend of ore and host rock from the ore boundary — to be processed through the plant rather than stockpiled and reclaimed later. That is a dilution-tolerance decision, made in mine planning, enabled by a piece of process equipment. TOMRA and PLS both describe sorting at Pilgangoora as having moved from an optimisation layer bolted onto the flowsheet to a core process step, with PLS's head of operations crediting it with unlocking the full potential of the orebody at a one-million-tonne-per-annum production rate.
The same idea at bucket scale
Sorting does not have to happen in a plant. MineSense's ShovelSense mounts high-speed X-ray fluorescence sensors directly on cable shovels, hydraulic shovels and front-end loaders, grades the contents of each bucket and signals the fleet management system in real time to send the truck to the mill or the waste dump. At Copper Mountain in British Columbia, MineSense reports that in under six months of 2020, across a Hitachi EX5500 shovel and two Komatsu loaders, the system diverted 1,555 truck loads — about 342,000 tonnes — of economic copper ore out of the waste stream, worth C$3.6 million and a four per cent improvement in ore recovery, while rejecting 380 loads of non-economic rock, worth C$440,000 and a one percentage point reduction in dilution. Modelling single-bucket grades and positions also exposed high-grade stringer veins that the blast-hole grid had missed entirely. Freeport-McMoRan presented a comparable case at SME's 2024 MineXchange, reported by International Mining, describing more than ten per cent of trucks being re-routed at its Chino mine in New Mexico.
How solid is the evidence
The two evidence bases are not equivalent and should not be read as such. PLS's numbers are lodged with the ASX, repeated quarter after quarter, and sit inside audited annual accounts; they are the strongest public proof that large-scale sorting survives contact with a real operation. The MineSense figures are vendor-published, though with a named operator attribution from Copper Mountain's then chief operating officer, and the Freeport data came from an operator's own technical presentation rather than a filing. The honest summary is narrower than the marketing: sensor-based sorting works where the orebody presents genuine sensor contrast and where the operation's real problem is dilution or selectivity. It is not a general-purpose upgrade.
The honest limits and economics
Fines cannot be sorted. TOMRA's finest belt machine starts at 4 mm and the standard XRT units at 10 mm, so undersize bypasses the circuit and goes straight to the mill with whatever waste it carries. Feed must be screened to a roughly three-to-one size ratio, which means additional crushing, screening and conveying capacity, the space to put it and the power to run it; particle singulation on the belt sets a practical ceiling on throughput well below nameplate. Misplaced particles are real losses, because rejected rock leaves site with metal in it. And ramp-up is not free: PLS's December 2024 quarter saw production fall 14 per cent to 188.2 kt and recovery average 72.1 per cent, with the company citing the transition to its P850 operating model together with planned downtime for P680 crushing and sorting ramp-up, and noting that Pilgan plant recovery was impacted during the period by that ramp-up and optimisation. Two quarters of disruption is the price of the following two years.
What it means for operators
The sequence that works is test-work first, flowsheet second, purchase order third: bulk and particle amenability testing on representative material, including the size fractions you will actually feed, before any capital case is written. Sorting integrated into the flowsheet from the outset behaves differently from sorting retrofitted around an existing circuit, which is the clearest structural lesson from Pilgangoora. For quarries and mid-tier operations the same physics applies at smaller scale to contaminant removal and to salvaging marginal stockpiles. And whichever scale you work at, the value only shows up if reconciliation is honest — block model to dispatch to sorter to mill — because a sorter that nobody reconciles is a machine generating an unverifiable claim.