Engineering & permitting

SpecificationWidely cited industry acceptance criteria (Read & Stacey guidelines, as reproduced in Gaida et al., the Bingham Canyon Mine paper cited below) are guideline values, not a binding rule: bench scale minimum static factor of safety 1.10 with a maximum probability of failure P[FoS<=1] of 25-50%; inter-ramp scale 1.15-1.2 at 25% PoF (low consequence), 1.20 at 20% (moderate), 1.2-1.3 at 10% (high); overall slope 1.2-1.3 at 15-20% PoF (low), 1.30 at 10% (moderate), 1.3-1.5 at 5% (high). A second published set, from Pothitos and Li, 'Slope Design Criteria for Large Open Pits - Case Study' (https://papers.acg.uwa.edu.au/d/708_21_Pothitos/21_Pothitos.pdf), not the paper cited below, expresses the same idea purely in probability terms: bench 10-50% (10% preferred), overall or inter-ramp 1-3%, and below 1% (values of 0.5% and 0.3% are cited) where permanent infrastructure sits within the failure runout.

Where it falls downThe criteria assume the strength inputs are real. Data needed: oriented core or televiewer structural data, rock mass classification logging, laboratory UCS and triaxial strengths, discontinuity shear strength (direct shear on natural surfaces), groundwater pressures from vibrating-wire piezometers, and a structural model that resolves major faults. Done on too little data, the strength distributions are guesses, the probability of failure number is decorative, and the usual failure mode is a wall that performs to plan for several benches and then unravels where an unmapped structure daylights. A factor of safety computed on assumed strengths carries no more confidence than the assumption.

SpecificationNIOSH roof bolt design (coal): suggested maximum intersection span IsG = 20 + 0.26(CMRR) feet; suggested bolt length LB = (Is/13)(log10 H)((100-CMRR)/100)^1.5; bolt intensity ARBS = (Lb x Nb x C)/(Sb x We) with suggested ARBSG = (5.7 log10 H) - 0.35 CMRR + 6.5, and a practical minimum ARBS near 3.0. Where CMRR is 50 or greater, successes outnumber failures roughly 6 to 1; CMRR below 40 at shallow cover and below 45-50 at deeper cover show persistently high failure rates. The document cited below is a coal roof-bolting study and carries no hardrock design values; hardrock rock-mass classification, span and stability-graph criteria must be taken from their own sources and are not reproduced here.

Where it falls downThese are empirical relations calibrated on case databases, not physics. They need honest geotechnical logging (CMRR or RMR/Q measured, not estimated from memory), depth of cover, and entry/stope dimensions. Applied outside the calibration range - very weak rock, high horizontal stress, seismically active ground, or excavations far larger than the database - they under-call support. Bolt bond strength also varies sharply with rock quality, so anchorage must be pull-tested on site rather than assumed from a catalogue value.

SpecificationAACE International Recommended Practice 47R-11, the mining and mineral processing cost estimate classification, is the reference framework (a recommended practice, not a regulation). Class 5: 0-2% project definition, conceptual planning, capacity-factored or parametric method, accuracy low -20% to -50%, high +30% to +100%. Class 4: 1-15% definition, screening of options, equipment-factored or parametric, -15% to -30% / +20% to +50%. Class 3: 10-40% definition, funding authorisation, semi-detailed unit costs, -10% to -20% / +10% to +30%. Class 2: 30-75% definition, project control baseline, detailed unit cost with forced detailed take-off, -5% to -15% / +5% to +20%. Class 1: 65-100% definition, bid/check estimate, detailed unit cost with detailed take-off, -3% to -10% / +3% to +15%.

Where it falls downThe SEC definitions in 17 CFR 229.1300 describe what each study must demonstrate but state no numeric accuracy or contingency percentages - those come from the AACE practice and from company or lender convention, so do not quote a percentage as an SEC requirement. The accuracy band is a function of how much engineering has actually been done, not of what the cover page calls the study: a document labelled 'feasibility study' built on Class 4 deliverables carries Class 4 error. Mining projects also carry geological uncertainty that the AACE bands do not cover at all.

SpecificationPractical kriging plan parameters (published geostatistics teaching notes): the search radius should be set at the variogram range or larger, sized so the required number of data are found; for final estimates about 40 search data in 3D, or 24 in 2D, is a reasonable maximum; the number of composites taken from any one drillhole should be restricted so a single hole cannot control a block. Kriging variance is a robust measure of data spacing; kriging efficiency is low in sparsely sampled ground and high where drilling is dense.

Where it falls downKriging needs enough composites to fit a variogram - typically several hundred well-distributed samples per estimation domain - and it needs domains that are geologically real, because a variogram fitted across two mixed populations is meaningless. On too little data the variogram cannot be fitted, the practitioner falls back to inverse distance or a borrowed variogram, and the result is a smooth model that understates both high and low grades (conditional bias): the mine then mines tonnes at a grade lower than the model promised. Indicator kriging needs enough data at each threshold, so the upper thresholds fail first. Nearest neighbour should never be used for reporting - only for validating that the estimated model reproduces the declustered mean.

SpecificationA published worked comparison at 5 m bench height tested SMU sizes of 10x10, 15x15, 20x20, 25x25, 30x30, 40x40 and 50x50 m (all 5 m thick) against a 1 x 1 x 5 m reference model. The 20 x 20 m block matched ore and waste tonnage (grade differed by 2%: 7.22% estimated against 7.35% actual); a 6 x 6 m block - read off the plotted relationship rather than simulated - matched ore grade (tonnage differed by 4%: 2.37 Mt against 2.42 Mt). As SMU size increases, reported ore tonnes fall and reported ore grade rises - the grade-tonnage curve rotates.

Where it falls downBlock size cannot be chosen freely: too small relative to drill spacing and the estimate is a smoothed interpolation dressed up as selectivity; too large and genuine selectivity is thrown away. The classic rule is that block dimensions should be a meaningful fraction of drill spacing, not a fraction of the ore geometry, and that selectivity claimed in the model must be achievable with the grade control data density and the equipment actually on site. Reporting a resource on an SMU smaller than the mine can dig overstates recoverable tonnes and grade.

SpecificationThe mapping is exact: 'There is a directed arc from node i to node j if block i cannot be extracted before block j, which is in a layer right above block i', and by Picard's construction the source side of the minimum s-t cut is the maximum closure. The original Lerchs-Grossmann algorithm is pseudopolynomial, bounded by O(|V| x sum of positive block values) iterations at O(m) per iteration; strongly polynomial variants run in O(mn log^2 ||b||) (scaled) or O(mn log n) (lowest-label with dynamic trees). Pseudoflow is the min-cut/max-flow family that replaced it in practice; on mining data push-relabel was already found empirically to be 'substantially more efficient than the (original) LG algorithm'. Nested pit shells are generated by re-running the optimiser across a set of revenue factors, and the differences between shells become pushbacks.

Where it falls downThe optimiser only maximises undiscounted value. It has no concept of time, so the true shell that maximises NPV is usually smaller than the shell that maximises value; that is why nested shells and pushback selection exist. Inputs are the whole story: a shell is only as good as the block values, which are only as good as the grade model, the metal price assumption, the recovery curve and the slope angles fed in. Slope angles are usually simplified to a small set of zones, so the shell must be re-cut into a practical design with ramps before it means anything.

SpecificationLong-term scheduling is formulated as a mixed-integer program: precedence constraints enforcing extraction from top to bottom, mining capacity constraints, processing capacity constraints, and volume-flow constraints that limit period-to-period variation. A published benchmark ran six instances of 53,271 to 4,000,000 blocks, with annual discount rates of 5%, 10% and 15% depending on the case; instances were allowed 72 hours each; the smaller instances solved to optimality within hours, and good solutions within 3% of optimality were obtained for most instances in under ten minutes. Where a conventional heuristic pushback scheduler left optimality gaps of roughly 15% on two of the instances, the direct MIP stayed within 3%.

Where it falls downA million-block MIP is not solved exactly; it is solved to a gap, and the gap must be reported. Schedules built on a single deterministic grade model inherit that model's error, so a schedule can look robust and still miss mill feed grade in the early years, which is where NPV lives. Practical schedules also need minimum mining width, ramp access, equipment fleet continuity and stockpile rehandle rules that a pure block-precedence model does not contain; a schedule that ignores them is not mineable.

SpecificationUpstream: cheapest, but the embankment is founded on loose, saturated tailings. EPA's technical report on tailings dams states that raise rates 'less than 15-30 ft/yr' are most desirable and 'over 50 ft/yr can be hazardous', and that the discharge 'should contain no less than 40 to 60 percent sand' to build a competent beach. Upstream construction gives poor seismic resistance, limited control of the phreatic surface and is 'not appropriate in areas with a potential for high seismic activity'; the failure mechanism is stated plainly as 'low relative density with a high water saturation. can result in liquefaction of the tailings embankment'. Downstream: good seismic resistance, drainage can be built in, suitable for water storage, but needs a large volume of fill and covers a much larger footprint. Centreline: intermediate - resistant to seismic loading, drains can be incorporated, raises can be added faster than upstream or downstream, but it is 'not amenable to permanent storage of large volumes of water'. In all cases the design intent is that the 'phreatic surface should not emerge from the embankment and should be as low as possible near the embankment face', achieved with chimney and blanket drains two or more orders of magnitude more permeable than the surrounding fill.

Where it falls downRaise rate, beach width, sand fraction and phreatic control are the whole safety case for an upstream dam, and all four depend on operating discipline over decades rather than on a drawing. Needed data: cyclic and monotonic undrained strength of the deposited tailings (in-situ CPT with pore pressure, not just index tests), piezometric records, site seismic hazard, and deposition records that actually match what was planned. Where strength is assumed rather than measured, and where the tailings are contractive, the design can satisfy a static factor of safety and still fail by static or seismic liquefaction with no warning.

SpecificationFiltration to a typical moisture content of less than 20%, achieved with belt, drum, horizontal and vertical stacked pressure plate, and vacuum filtration systems. The source cited below is an independently maintained industry website, not a government or standards-body publication, and the throughput ceiling it gives is no longer current, so no tonnage limit is quoted here. It notes the method allows higher rates of rise than conventional impoundment, suits seismic areas because no water-retaining embankment is built, conserves water in arid climates and improves recovery of dissolved metals and process chemicals.

Where it falls downHigh capital and operating cost for the filtration plant; no operational water buffer, so the site water balance must be solved elsewhere; dust generation in arid climates; upstream diversion works still required for storm flows; and some sulphide tailings oxidise faster when unsaturated. Filter performance is sensitive to grind size and clay content, so a bench and pilot filtration testwork programme on the actual tailings is the minimum data requirement - throughput assumed from a similar ore is how filtered projects get undersized. Filter plant capacity has risen substantially since the cited page was written, so any published upper limit on throughput should be checked against current operating practice before it is relied on.

SpecificationBuilt as a stochastic model: uncertain variables are entered as probability distributions and daily precipitation and pan evaporation are sampled from distributions derived from historic climate records. A Monte Carlo run 'will typically run up to 1000 realizations', and results are reported as percentiles - the median (50th), the 5th and 95th, or finer - rather than a single number. The model advances in time steps, allows operating rules and seasonal inputs to change at specified dates, and is commonly coupled to a contaminant mass balance so that load divided by volume yields a predicted concentration series.

Where it falls downOnly as good as the climate record behind it; short or non-representative station records understate extremes, and a model calibrated on ten years of data says nothing reliable about a 1-in-100 wet sequence. Seepage and runoff coefficients for waste rock and tailings facilities are frequently assumed rather than measured. Groundwater components must be calibrated: BLM's Nevada State Office guidance - state-office guidance, not national BLM policy and not a federal rule - asks that a groundwater model 'be calibrated through appropriate techniques as defined under ASTM protocols' and lists the applicable ASTM standards (D-5447, D-5490, D-5609, D-5610, D-5611, D-5618, D-5880, D-5981); that requirement is at https://www.blm.gov/sites/blm.gov/files/policies/Attachment%201-%20Nevada%20BLM's%20Water%20Resource%20Data%20and%20Analysis%20Guide%20for%20Mining%20Activities%20(16pp).pdf, not in the paper cited below. A water balance run on a single average year, with no percentile output, is not a design basis.

SpecificationAcid-base accounting compares acid potential, calculated from sulphide sulphur (1% pyritic sulphur requires 31.25 t CaCO3 per 1,000 t of rock), against measured neutralisation potential. BLM's Nevada State Office guidance - state-office guidance for BLM land in Nevada, not a national federal rule - states the kinetic-testing trigger explicitly: a kinetic test is required if 'the NNP does not exceed +20 and/or the NP value is not at least three times greater than the AP value'. The same guidance notes that Nevada state practice requires a meteoric water mobility procedure - an ASTM leach test - to identify constituents that will mobilise under rainfall. Samples 'should be collected that represent interburden, overburden, and ore zone', and where a pit lake will form, samples must 'represent the ultimate pit surface or pit wall'. Humidity cell testing follows ASTM D5744, on at least 1,000 g of material crushed to pass a 6.3 mm (0.25 in.) screen, cycled weekly with dry air, humid air and a weekly 500 or 1,000 mL leach, and analysed for pH, alkalinity/acidity, specific conductance, sulphate and selected metals.

Where it falls downNeither the sample count nor the test duration is fixed by federal rule; state guidance asks only that 'adequate characterization data. illustrate the full range in geochemical characteristics representative of each major lithologic, alteration, and mineralization unit', which puts the burden on the programme designer. ASTM D5744 itself says 'test duration is determined by the user's objectives of the test'. Characterisation done on too few samples typically misses the small-tonnage, high-sulphide, low-carbonate units that drive the eventual water quality problem, and kinetic tests stopped too early miss the lag time to acid onset entirely - laboratory and field release rates 'could differ by orders of magnitude', so scaling factors matter as much as the raw rates.

Specification43 CFR 3809.401 requires operator identification, maps of exploration, mining, processing, waste disposal and access, 'preliminary or conceptual designs, cross sections, and operating plans', water management plans, rock characterization and handling plans, quality assurance plans, spill contingency plans, 'a general schedule of operations from start through closure', and plans for roads, pipelines and utilities. The reclamation plan must address drill-hole plugging, regrading and reshaping, mine reclamation including a pit backfill feasibility analysis, riparian mitigation, wildlife habitat rehabilitation, topsoil handling, revegetation, 'isolation and control of acid-forming, toxic, or deleterious materials', building removal and post-closure management. A monitoring plan must specify 'type and location of monitoring devices, sampling parameters and frequency, analytical methods, reporting procedures'. An interim management plan is required for temporary closure.

Where it falls downThe regulation sets content, not numeric design criteria - no slope angles, no cover thicknesses, no factors of safety. Those come from state programmes, from the agency's reclamation cost estimate, and from the design standards the operator commits to. BLM's Nevada State Office 3809 Reclamation Bonding Guidelines - state-office guidance, not federal law, not part of 43 CFR 3809, and not binding of its own force in other states - set out the mark-ups used there: engineering, design and construction at 8% up to $1 million, 6% from $1 million to $25 million and 4% above $25 million; contingency at 10% up to $500,000, 8% to $5 million, 6% to $50 million and 4% above; insurance at 1.5% of total labour; performance and payment bond premiums at 3% of operation and maintenance cost; contractor profit at 10% of O&M; contract administration at 10%, 8% or 6% on the same tiers; and a 21% agency indirect rate on contract administration. Under the same Nevada guidance, cost estimates and financial guarantees are reviewed at notice extension (every two years) and at least every three years for plans of operations; 43 CFR 3809.552 itself says only that BLM will review them periodically. See https://www.blm.gov/sites/blm.gov/files/documents/files/NV_3809_Reclamation_Bonding_Guidlines.pdf.