Can you please describe your situation?

• Kim Esbensen TOS: it is not possible to ascertain the status of an individual sample (representative sample or worthless specimen) by any feature - a rep. sample is the result of a representative sampling processes, the characteristics of which are that is is correct (all incorrect sampling errors have been eliminated and for which the correct sampling errors have been reduced to an a priori acceptable threshold level. It is ONLY the sampling process which can be validated as representative. This is of course the SIMPLEST and at the same time also the most general definition.

• Rene Sterk I've seen different authors of sampling papers have different definitions. Some state that a "sample" must have been taken accurately and with a precision that is fit for purpose. I have also seen people stating that accuracy should be fit for purpose too. I adopt the former approach but interested to see what others think.

• Mark Murphy I have three criteria for representativity. As Kim points out first you must have a sample not a specimen. The golden (Gy's) rule here is that every fragment in the lot has an equal chance of ending up in the sample. Next systematic bias causes must reduced to an acceptable minimum (say <1-2% relative) - so you must have an estimate of sample recovery (mass recovered v mass of lot and some understanding whether the part lost invokes a systematic bias. Certified reference materials may help in the finals stages here but a correct sampling method is the best assurance. Finally, the precision of replicates (random errors) must be measured and deemed acceptable for sample decision purpose. In mining this can be high as +/-30% (including the hopefully small systematic errors) for all stages of a protocol - and hope for less than this. Gold is problem here as Ian will no doubt comment - but the problem is also an issue for the different minerals considered in deposits such as mineral sands.

• Jan E. Nesset I think Mark pretty much has it covered. Every particle has an equal probability of being selected, the protocol selected minimizes error due to bias, and the number and size of sampling increments gives the required precision. Don't forget that liberated mineral particles will help minimize random error.

• Dennis Lahondès Reproducibility of assay results.

• Ian Robertson Kim, I am devastated as a boss you are questioning my extensional powers to deem a sample is representative or not by looking at it. You are saying it is a process. I know a stick of core with a blob of coarse Au, a “director’s special” sample, is a process to enhance one’s budget, the bigger and more spectacular the blob of Au the more successful it is. The extensional powers must remain in the domain of some CPs (or QPs) judging by some statements in ASX Table 1 Section1 expressing the quality of drilling, sample preparation and analyses as Rene reminds us.
  Where does the process start?

  I have some SOPs from a former employer, a major international company, must be world class, probably world’s best practice too (sounds nice), I have cut and pasted onto our template with logo to disguise them and called SWP – safe work practices to emphasize our safety prowess. Mark the SWP tick all your boxes. SWP just like a recipe to follow. I believe in KISS.
  Dennis, unfortunately, the duplicates do not repeat. Must be just coarse gold! Looks like a bad start!

  Sampling is performed for a purpose (fit for purpose) to fulfill an objective using data of a quality (QA) to minimise the perceived risks. Does a "one size fits all" approach satisfy all purposes? e.g. JORC2012 Table 1 section 1 reference to "industry standard" - submission of <3kg sample for one pass preparation in LM5 and FA30.

• Graham Eccleson Gold levels in metal concentrates such as copper, lead and zinc can be grid sampled into 3kg sacks and then sample prep takes over to attain a 'representative' sample of the whole consignment, whether it be a 40te truck/ 30te container/68,000te ship.........the latter would involve autos auto-sampling during concentrate discharge, regulated and recognized by an accredited, UKAS, ISO standards etc.

• Steve Crutcher Without a detailed description of the entire sampling process, the equipment used during the process and the QA,QC methods employed, a "representative sample" may be a more of a suggestion, potentially indicative of wishful thinking.
  Rene, are you saying that accuracy does not have to be fit for purpose ?
  Mark, you say that the golden (Gy's) rule here is that every fragment in the lot has an equal chance of ending up in the sample. Could it be that the variable content value for each fragment makes the equal chance idea, impractical in many cases ?

  Ian Robertson The QA must at least be fit for the purpose to achieve the objective. Eg processing plants sampling for process control needs to be reasonably precise but not accurate. Sampling for metallurgical accounting needs to be both accurate and precise. See Lachance et.al. Detecting sampling biases in metal accounting, Sampling 2014 p109-119 & Pitard Sampling systems for process control & metallurgical accounting. Sampling 2014 p25-31. When companies enter into toll processing arrangements, the plant sampling systems & protocols designed for process control on own ore are used as a basis of a commercial transaction rather than sampling according to Amira P754 metallurgical accounting code of practice guidelines. This is the cause of many disputes regarding a third party's gold produced in toll processing whether by batching small lots or commingling ores as previously indicated by Mark Murphy in a previous discussion on Toll Milling. It is to be clear regarding purpose, QA and objectives.

• Mark Murphy Steve - good question. The fragment grade variability needs to be understood before attempting to take any sample. Theory of sampling gives a way of estimating whether you can assess whether you are collecting a representative sample (having a certain repeat precision and accuracy) as long as every fragment in the lot has a chance of ending up in the sample. The problem here is many geologist find Gy's equations hard to implement (the 'I did geology so I wouldn't have to do math' effect) and in the end the end the results is a 'location and size of the green' answer not where the hole and flag might be.

• Ian Robertson Mark agree mathematics is beyond us lesser mortals but unfortunately, it is used as an excuse not to practice Correct Sampling to minimize sampling bias (accurate). The rules to achieve Correct Sampling are simple enough with many diagrams to illustrate & explain principles of Correct Sampling without understanding the mathematical complexity. Agree TOS is based on each particle independently having an equal chance of selection. However, in RC drilling to control dust regardless of misting or water injection, it introduces a stickiness to particles, adhering to equipment (contamination) & to other particles or forming aggregates (causing segregation) which violate the assumptions on which the TOS is modeled. I have not seen but stand to be corrected that sampling has been modeled considering particle stickiness. Best controlled by avoiding misting or water injection in the first place using a sampling system such as Progradex to collect the dust as part of the sample.

  https://www.linkedin.com/groups/3088030/3088030-6250587477338394627

The experiences I have had show that it is necessary to know a lot about the beneficiation process that you want to apply. It sounds like a contrarian, but it's an important part of the decision-making process. Sometimes we choose too much detail for extremely robust plants or simplifications on extremely sensitive plants.
Another very important factor is to know the stage of the project, since from the exploration until the implantation of a plant there are many variables to be evaluated. Often we think of processing a medium quality ore, but depending on the development of the design and the implantation it is very possible that we have to think about extracts / panels of ore types, ie the samples should contemplate these variations.
I seek to do a critical analysis of deployed projects, mine operational difficulties and plant operation and return to the drafting of the sampling plan. I start by looking for an open-loop scan, then see if there is regionalization of behavior. Only after that do I determine if we are sure about the challenges.
Of course any and every sampling plan is based on a very robust knowledge of the model and geological controls of mineralization, as well as the detailed characterization of ore types. Simplistic models may abbreviate work, but can lead to irreparable errors in the future