Yes, you might want to start the investigation by weighing the rejects in addition to the samples. The total (sample plus rejects) will give you some insight into the variability of the RC rig's output, and the ratio of the two will highlight the effectiveness of the sampling process.

For sure the first step should be to weigh all splits and get the total.

That weight range seems excessive, although difficult to comment without witnessing operations first hand. In my experience, feedback from the client in the form of data is invaluable in improving and maintaining sample quality. Particularly the weights.

You could also plot an RSD plot of variance between weights vs variance between grades to see the effect it has on grade. Another consideration is accuracy of sample collection. When one sample interval is overdrilled one by 100mm, and then underdrilled by 100mm a 20% variance is seen. Easy to do when pen rates are fast.

Variability of samples weights from a single sample point alone does not really tell much about the source of variability, and if that variability is acceptable. Best is to compare primary and duplicate samples taken off the splitter, or primary and reject weights (sample recovery) to aid in determining if the source of variability is at the splitter, from drilling practices/configuration, or due to ground conditions. But sample weight variability isn't the full story. As pointed out, grade variability is also important - at the end of the day it is critical that grade is representative, not necessarily the weight of samples taken from the rig.

It is common to blame the splitter for variability in sample weights, but for most modern splitters, provided that the splitter is level and has not been damaged or excessively worn, and that samples are dry and properly fed into the splitter, then you should expect low variability from the splitter. If these are all good and ground conditions are not a factor (as you state), then drilling practice or rig configuration (air, bit tolerance etc) are the likely main cause of variability.

In the absence of duplicate or reject sample weights you could look to see if there are systematic variations down-hole or either side of drill runs / rod changes to help determine if the variability is from drilling rather than the splitter. As an example, when air capacity starts to become a problem, within a run you might see poor initial recovery (or declining recovery) followed by a"burp" of high recovery, or excessive sample return when the driller blows out the hole at rod changes (from sample not lifted to the surface falling back down the inner tube when the air is shut off).

Slightly unrelated but Increased sample size will add costs down the line including transport costs and lab splitting costs. Labs generally/often place a top limit on sample weight and any heavier samples will be split down incurring further costs. Worth checking.

In operations I've been involved in smaller split samples either mean poor sample recovery related to things such as bad ground conditions, the start or end of a rod, water ingress, clay, slopes etc. or alternatively drilling related such as splitter blockages, depth of hole and sample travel time through the hoses etc. Hole and hammer size affect sample volume directly so you could work out expected volume and weights for a start then check the drilling operation as it happens and work out the reason for the large variability.

In the end if you need a minimum sample weight there may be a case for re-splitting the reject/bulk to increase assay sample size.

This paper may help you out - Obtaining a Representative RC Sample — The Cone Splitter Versus the Tiered Riffle Splitter by B Catto and P Church (5th International Mining Geology Conference - Nov 2003). If you are an AuSIMM member, you can download it for free at https://www.ausimm.com.au/publications/epublication.aspx?ID=1321

If you are increasing the sample size simply by changing the sample/reject ratio on the splitter, then I would imagine that the variability is absolute. If you are changing the sample size by increasing the size of the bit, then the variability will be RELATIVE (edit).

In my experience: Depth, bad ground, etc have much more of an effect on sample size variations than splitter performance (we used a riffle splitter that was leveled at every hole, and cleaned out between rod changes). In a good hole, RC sample size generally increases with depth (1.5-2x mass by 166m). "Bad ground", water, or anything that can cause circulation loss are other major factors.

I am not sure that the variation of sample masses will be a bad thing, as long as each and every sample is large enough to be a representative sample.
"The interesting thing here is the mean grade of the larger sample population is higher than the smaller sample population."

Could the following also be also a possibility?

• deeper samples have higher mass sample return (ours consistently increased with depth)
• your deposit is at depth

therefore your larger samples have a higher grade?

Could there be geological reasons why you are getting a larger sample along with more gold?; is it easier to drill and return sample where mineralization is favourable?

We run a very large RC program in SW Nevada at the Mineral Ridge Mine, exploring for a large but low-grade gold system with numerous spotty pods of mineralization scattered through a contorted section of Paleozoic metasediments. The scope of the program involves 3 rigs, and annual footage is nearly 100,000 feet (nearly 20,000 5-foot samples). We use large sample bags - about 14 x 17", and typically collect about 13 # on average.

Based on your query, I ran the numbers for the last full year of drilling (2014) and can report the following:

Total # of samples: 19269
Average Weight: 13.01 #
Median Weight: 13.18 #
Modal Weight 14.48 #
Std Dev: 3.58
1. SD covers a sample weight range from 5.86 # to 20.14 #, which in fact includes 18,295 of the 19,269 samples, or 94.94% of the total.

In our project area, the rock is layered, uniformly hard, and devoid of unusual situations (broken ground, clay zones, and so forth), such that drilling progress is pretty consistent, day in and day out, with daily footage per rig of about 600 feet, and average bit life of 700 feet. In short, we have an extremely stable and predictable drilling environment, and it has been that way for over 4 years. We rarely have to change the pie-plate configuration on the splitter, and our duplicate samples (also of the same size) consistently support the gold values recorded in our principal sample. Historical records show that approximately 5% of the total exploration sample footage will yield gold grades of 0.010 opt or better, with virtually no nugget effect.

The few times we have encountered substandard sample weights, it has been easily identified as the fault of a new, inexperienced sampler not following the driller's procedural instructions, and those situations were remedied as soon as the received lab reports indicated that the weights were shy of expectations.

I know of a large resource drill out where we systematically weighed, analysed, and graphed both total sample and split weights (eg - graph by rod so you can see variation from the first to the last metre and determine if there is a pattern over all the rods in the hole). In my experience, by far the greatest contribution to weight variability was in the drilling - whether technique, ground conditions, sample collection you all mentioned, it came back to the drill rig. We were using 4 different splitters and did not find any significant bias for any of these when the split weights were analysed.

I have been in the field and just re-read your post and other comments, and think that most of us who initially commented missed the point.

It is clear that there is a systemic sampling bias, as indicated by the consistently higher sample weights in the primary sample, combined with consistently higher Au grade in the heavier sample. This points to a gravitational or centrifugal bias occurring within the sampling system, somewhere between the cyclone and the sample entering the sample bag. This bias is most likely to occur from one of three sources:

• non-level splitter tilted towards the primary sample shoot (most likely);
• damaged or off-centre cone resulting in more sample reporting to one side of the splitter; or
• uneven sample feed from the drop box onto the cone (damaged/blocked feed shoot or gate or, heterogeneous or off-centre sample distribution in the drop box, due to preferential "streaming" of sample from the cyclone).

The preferential deportment of gold to the primary sample shoot relative to the rest of the sample (indicated by higher grade) is likely due to the much higher density of gold compared to the rest of the sample (particle size may also have an impact), and is the smoking gun for a gravitational/centrifugal bias within the sampling system.

Larger sample size: I don't think that the higher grade is because you have a larger sample, but is instead due to gravitational bias (as discussed above). If it was a sample size issue then the smaller (duplicate) samples should show greater variability (less precision) that the larger sample, but should return grades that vary evenly, both higher and lower than the larger sample. So I don't think a larger sample size will help you. Presumably you would collect a larger sample by widening the sample cutters. Depending on where the sample is preferentially falling within the splitter, then opening the cutters could improve things, but could equally exacerbate the problem. Your focus should be on identifying the source of the bias and eliminating it.

Is it acceptable? I would say not. Modern sampling protocols are specifically designed to remove this sort of bias. Assuming that you use the primary sample for your ore reserves, the results you discuss will systematically over-estimate grade. True grade is likely to be somewhere between the primary and duplicate, but probably not the average, and is unlikely to be reliably factored.

What would I do? Check the sampling practice, in particular focus on the levelling of the splitter and rest of the system. If that is okay, then work up from there, checking the condition of the rest of the system: the cone (good condition and centred under the feed chute); the feed chute and gate; and, the cyclone. If all of that is good and you still can't resolve it, request that the drillers try a newer splitter. If the problem still persists then it must be the cyclone and the way it feeds to the cyclone.

I agree with May, ploting an RSD plot of variance between weights vs variance between grades to see the effect it has on grade is the way to go.