"Besides, even in the absence of that eagerness and want of thought, (which we have mentioned,) it is the peculiar and perpetual error of the human understanding to be more moved and excited by affirmatives than by negatives, whereas it ought duly and regularly to be impartial; nay, in establishing any true axiom, the negative instance is the most powerful." *

Predictive modeling in science and in other disciplines normally involves some specification of experimental control and application of common statistical methods. A population is specified prior to any experiment to establish such control; to minimize bias, subjective influences; and to allow for unbiased measures of test significance / design error. The population fixes parameters for (controlled) inference; the base for samples and related data. Control is established with the totality of a population, with exhaustive sampling and testing - not with selective analysis, fitting and presentation of narrowly-based data to support a preferred outcome.

Quantitative models, hypotheses etc. are usually statistically tested within limits of error and with some measure of dispersion. Uncertainty is a given. Any "point" estimate (related statistic, observation, opinion, guess, extrapolation, interpolation) might be presented as a substitute for a true value or a population parameter. Facile or speculative models of this kind can be made more complex, specialized and numerous, of course. On the other hand, measures of uncertainty or bias may not be presented or may not be known. Without definitive tests of location and dispersion, such conceptual models can only be presented as (statistically) untested or raw hypotheses. In the limit, only unbiased estimators will have the greatest probability of constraining the sought parameter; and otherwise will reliably report uncertainty, variation and dispersion.

In principle, a rigorous predictive design for mineral deposits will be supported by a pre- specified, and carefully characterized population of deposits - and some model of the behaviour of the (greater) mass of the silicate systems that host gold deposits. Only these foundations would allow for controlled evaluation of a single sample, a sampled width, a drill-intersection, a single zone, a single deposit, a single camp or a deposit class. The reverse approach might invite damaging bias; costly and repetitive errors - or repeated and unproductive modeling efforts!

Speculative models and controlled inference should therefore be distinguishable by constraint of uncertainty and bias. Of course, speculation may develop toward controlled (frequentist or Bayesian-based) inference where rejection of hypotheses (in the Baconian, inductive sense) will be within (quantitative) limits of test significance. On the other hand, untested opinion, hypotheses or models are more likely to remain open- ended, to pose further questions and perhaps to present intractable or conflicting generalizations with each new trial.

When uncertainty is unspecified or when potential biases are not constrained, the approach could be flawed or misleading. Selective testing, sampling, or characterization; selective elicitation of data or of opinions; selective presentation and disclosure will (inevitably) produce interpretations or results. On the other hand, these may only have a small probability of representing a true value or of providing a helpful predictive tool (figuratively, watching one's feet to see where the road goes).

Additional speculative deductions (each with some smaller probability), may increase uncertainty rather than confidence in the state-of-the-art or in the existence of predictable natural processes. Controlled inference should be recognizable by measures of uncertainty - not by degrees of confidence, certainty, knowledgeability, volume, nor exclusivity of opinion. Sampling and testing can never completely characterize a "universe" of possibilities, but comprehensive use of available (often inexpensive) data serves to distinguish controlled inference from merely speculative and possibly biased evaluation methods.

Whether the activity is pure research, applied research, technical development, continual quality improvement, benchmarking or business process improvement - ever growing information will demand increasingly effective management of choices and biases; risks and rewards. Visibility of activities to identify and to constrain uncertainty - not to avoid it - may be indicative of robust project control and organization. Tests of uncertainty are an important distinction between progressive science and speculation - as Sir Francis Bacon so clearly pointed out in 1620.

(See K.H.Poulsen, F. Robert and Dube', Geological Classification of Canadian Gold Deposits, Bulletin 540, Geological Survey of Canada, Ottawa, 2000 for a fuller discussion of these general approaches and issues);

geographic/ spatial analyses/ visualizations (GIS methods, 3D visualization technology) of selected regional- to deposit- scale characteristics (e.g. lithological, alteration, structural, known gold distributions) with imputed spatial proximities (characteristic analysis X-x,y,z);

thermobarometric (P-T-X) characterizations of multicomponent, multiphase (mineral-fluid) equilibria (P-T curves, tie lines, intersections, equilibrium tangent planes, phase diagrams, pseudosections) given (observed, experimentally derived, critically assessed, "internally consistent") thermodynamic data on specific (C+1) mineral end-members and their (singular) solid solution, given bulk compositions; related optimizations (multiple solid solutions, end-members; experimentally-derived, activity, free energy constraints; statistical least squares fitting and error measures); most directly for magmatic or metamorphic petrology but with constraints (P-T) for regional gold distributions

mass flux analyses, geochemical differentiation / fractionation/ partitioning simulations (X-oxide, X-trace element, X-rare earth element/ REE, X-isotope, X-gold); species X ratio distribution analyses and diagrams (e.g. AFM diagrams); some with, some without thermobarometric (P-T) inputs;

elicited (prior) methods (subjective probability, expert, neural networks, fuzzy logic) as they correlate expert knowledge of gold distributions variously against expert knowledge of mineralogical, geochemical, alteration, lithological, structural or geophysical indicators (characteristic analysis, bias analysis X-x,y,z); ;

Intensive sampling, data collection and broader acquisition practices can also be considered as an alternative, although costly state-of-art. Mitigation of cost and risk by acquisition of discovered and developed, rather than "grassroots" prospects may be an alternative for companies with cash flow from mining (smelting and refining) operations (major, integrated miners) and with access to large or derived pools of risk capital.

genetic classifications are limited in their ability to characterize physical conditions at the time of gold deposition with confidence; to clearly distinguish exclusive classes; and to maintain consistent terminology;

host rock classifications do not account for mineralization processes that are vein-related and not specific to lithology, nor do they account for lithological diversity; volcanic, sedimentary, plutonic or metamorphic processes which also play some part in the mineralization process; industry nomenclature may vary

geochemical/mineralogical classifications, where hydrothermal and wall-rock alteration processes are proposed as keys, may constrain the lithological diversity issue of host rock classifications. Any hydrothermal and wall rock alteration classification again constrains interpretation of broader volcanic, sedimentary/placer, plutonic and metamorphic processes in mineralized areas (or unmineralized areas); industry classification systems may vary with respect to analysis of mineralogical, geochemical data

economic prototyping of large, well-known ("world-class") deposits and the form of gold distribution, may not typify an intended class of deposits and may not allow classification of other extraordinary deposits; unmineralized areas are not characterized;

tectonic classifications of deposits are difficult to apply in a practical way and suffer overlapping classification problems (whether pre-, syn-, post-orogenic, or more generally metamorphogenic); scales of property assessment and regional tectonic characteristics are different and may limit evaluation of tectonic parameters with respect to property characteristics;

geophysical (indirect) methods require geological and spatial model inputs (with their own limitations) to define attenuation characteristics, to distinguish "noise" and to obtain a unique solution on conductor (etc.) configurations; extrapolations, interpolations or correlations may be less reliable with certain discontinuities or in some orientations; regional-scale geophysics offers some (experimental) control, but is costly and generally subsidized by government to support property-scale exploration activity;

structural analyses of deformation regimes are focused on mechanical properties of lithologies and changes from regional, system and system segment perspectives (controlled, in this respect), but may not be sufficiently predictive without other geological inputs. Identified structures may variable with respect to mineralization potential;

geographic/ spatial analyses/ visualizations (GIS methods) may impute spatial proximities where discontinuities are a specific problem for lode gold distributions. Generally, functional continuity is a precondition for mathematical and statistical models; for extrapolations or interpolations. Spurious correlations may be a problem with attributes related to trace, fractional or partial segments of the silicate system mass;

geostatistical and geomathematical treatments may impute spatial distributions of gold at zone-scale without regard for discontinuities; spatial models may be uncontrolled from experimental, regional, or deposit population perspectives; characterizations of mineral resources or reserves may not be explicit with respect to statistical significance or biases

thermobarometric characterizations of multicomponent, multiphase equilibria provide some constraints on thermodynamic estimations and some implications for predictive models of gold distributions but suffer from errors of "closure", non-singular results and problems of retrograde reactions; P-T-X simulations may be incomplete with respect to other system processes which are also physicochemical (e.g. system deformation/ mechanical work; alteration/ system segmentation);

mass flux, geochemical differentiation / fractionation/ partitioning simulations may characterize mass transfers in systems or segments of systems; may impute system mass differentiation with trace components (e.g. isotopes, REE /rare earth elements, enriched/depleted element levels in mineral), but other system changes (physicochemical energy changes) may not be characterized; innumerable indicators could be specified, but reliability, statistical and predictive significance of any particular indicator for gold distributions may be issues. Used as a predictive indicator, gold (present in parts per billion or parts per million of the silicate mass) is notoriously misleading, circular and subject to biases. In practical terms these are issues which force intensive and costly sampling, and detailed analysis (in commercial practice). Other geological inputs (with unresolved issues described above) are required;

elicited (prior) methods (subjective probability, expert, neural networks) may utilize weightings, inputs; incomplete or non-discrete characteristics (e.g. fuzzy logic) which may be judgmental or subjective. Subjective influences are normally constrained in statistical inference. In principle, minimum-bias estimators (and discrete characteristics) will have the greatest probability of constraining the sought parameter.

intensive sampling, data collection or broader acquisition practices are dependent on availability of risk capital; on (cyclical, "grassroots" exploration) activity of risk-capital-dependent, junior mining companies, on relative cost with respect to in-house, grassroots exploration; on investor interest in gold exploration (vs. other commodities, high technology etc.); on tax relief and fundamental geoscience research assistance; on a specialized (and aging) workforce for a cyclical, commodity-based and traditional sector. It is indicative that government flow-through tax subsidies for mineral exploration and public funding of geosciences research provide extraordinary assistance to mitigate the situation - whatever the state-of-the-art. Not without its own risks, the sector has faced legal and governance issues at times of heightened investor and regulatory sensitivity. Deficiencies in the state-of-the-art, in historical and recent practices have been costly, time-consuming or damaging to industry, to investors, and to the profession: