| Vose Software

Industry: Mining and Natural Resources
Product: ModelRisk
Application: Mine-Plan Sequencing under Geological and Price Uncertainty


Phase the Pit or Plan the Life: Probabilistic Sequencing for a Cu-Au-Mo Deposit

The deterministic LOM model handed two answers across the metallurgist's desk. Plan A — a 20-year uniform extraction at 12 Mt/yr blended head grade — returned an NPV of about $390M. Plan B — a phased plan, 5 years of high-grade (top-quartile) ore at 6 Mt/yr, then expand to 12 Mt/yr for 18 more years — returned roughly $370M. The two strategies sat within deterministic noise of each other and the decision drifted for a year. ModelRisk reframed the same question: not which plan has the higher mean, but which plan owns the better left tail and the earlier cash-flow profile that the project finance covenant actually rewards.

A 200-million-tonne copper-gold-molybdenum deposit at 0.65% Cu was put through the same simulation under both extraction sequences. The two plans land almost on top of each other in mean NPV ($436M uniform, $431M phased), but the shape of the distribution — where the left tail sits — is where the decision actually lives:

NPV distribution: uniform 20-yr plan vs phased high-grade-first

Where the grade really lives

Both plans draw from the same orebody, but they sample it differently. The uniform plan averages everything; the phased plan front-loads the top-25% grade zone at roughly 1.4× the deposit mean (~0.91% Cu) for five years before reverting to the diluted remainder (~0.57% Cu).

  • Tonnage — LogNormal, mean 200 Mt, σ_log 0.15.
  • Cu head grade — LogNormal, mean 0.65%, σ_log 0.22. Phase 1 draws from a shifted distribution centred on 0.91%; Phase 2 from a depleted distribution centred on 0.57%.
  • Recovery — Beta, mean 85%, sd 5%, bounded on [0, 1] as physics requires.
  • Cu price — LogNormal, mean $8,500/t, σ_log 0.22, matching the LME range of the last decade.
  • OPEX per tonne ore — LogNormal, mean $32/t, σ_log 0.20.
  • Capex — LogNormal, mean $750M (uniform full plant) or 60%/55% phased split, σ_log 0.20.
  • Permit delay — Triangular 0/12/30 months (phased; small plant easier to permit) versus 12/24/42 months (uniform; full-scale).

Why Monte Carlo, not a point estimate?

The two plans land roughly on top of each other in mean NPV, but the distributions tell a different story. Phased shows a tighter, more concentrated distribution. The uniform plan has a longer left tail driven by the combination of full upfront capex, longer permitting, and an extra ~2 years before first cash. Both plans carry a substantial chance of a negative-NPV outcome at the wide price and grade uncertainty modelled — roughly 41% for uniform against 37% for phased — but the phased plan's shorter left tail (P10 of −$901M versus uniform's −$1,170M) is the difference that the project-finance covenant actually prices.

Phased dominates the bottom of the curve

The same picture as cumulative probability:

Cumulative NPV — strategy comparison

Below roughly $580M of NPV the phased CDF lies above the uniform CDF — phased is the downside-protected choice across the entire lower half of the outcome range. Above that crossover the two curves swap, because the bigger Plant 1 in the uniform plan can monetise an upside copper-price scenario more aggressively, giving uniform the heavier right tail (P90 of $2,246M versus phased's $1,898M). The phased plan is the downside-protected choice; the uniform plan is the upside-leveraged choice. The choice depends on where the project finance covenant lives, not where the mid-case lands.

What actually moves the answer

Tornado: drivers of phased-plan NPV

Copper price and ore grade dominate, as always — these are the two LME-and-drilling inputs the project effectively bets on. Recovery ranks third, just ahead of unit OPEX and capex. Permit delay appears sixth — the phased plan's smaller starter plant carries a much shorter delay distribution, which is one mechanism by which it earns the tighter NPV curve. Tonnage is the smallest driver here because LogNormal σ on tonnage is the smallest, and tonnage interacts only weakly with the early-life cash flows that discount most heavily.

Where the strategies cross over

Sweeping the assumed copper price from $5,500/t to $12,000/t under both plans:

Phased plan dominates at low Cu prices; uniform catches up at high prices

At Cu prices below roughly $8,500/t the phased plan wins on mean NPV — its early cash flow at top-quartile grade carries the project through the bear scenario before the long-tail risk of running Phase 2 at $6,500/t copper. Above that crossover the uniform plan pulls ahead on the mean, because the full plant monetises the upside more efficiently; at $12,000/t copper the uniform mean of $1,890M beats the phased $1,759M. The two strategies are close on a trial-by-trial basis — the probability that the phased plan beats the uniform plan drifts down only gently with copper price, from about 59% at $5,500/t to 48% at $12,000/t, crossing 50/50 near the $9,000–10,000/t band. The mean lines diverge more than the win-probability does, because phased trades a thinner left tail for a thinner right tail. The project economist now states the recommendation conditional on the assumed price deck, not as a single answer.

What changed

  • Phased plan adopted as the base case, on the strength of the P10 NPV gap (≈ $270M shallower left tail than uniform) and the earlier first-cash-flow date (≈ 2.5 years faster on the median permitting draw).
  • Project finance term sheet rewritten around the phased capex profile — debt covenants now reference the Phase 1 free cash flow rather than the consolidated LOM, lowering the interest-rate premium by roughly 80 basis points.
  • Phase 2 expansion decision deferred to a future board paper, framed as a real option: exercise only if the price deck two years before expansion shows median Cu > $8,000/t over the following decade.
  • Permitting strategy reprioritised — phased plan let the project file Phase 1 permits in parallel with feasibility, compressing the delay distribution by roughly 9 months on the median.

ModelRisk Functionality Used

  • Shared random-number streams so that the uniform and phased plans see the same draw of price, grade, recovery, and capex in any given trial — the only way to compare strategies fairly on the same simulated future.
  • LogNormal price and grade fits with σ_log calibrated to LME and assay records.
  • Beta recovery with method-of-moments parameterisation, bounded by construction.
  • Phased cash-flow engine with discount, delay, and expansion-capex timing controlled by the same draws — no double-counting or independent re-sampling.
  • Sensitivity ranking that put recovery ahead of capex as the focus for the next year of metallurgical research.
  • Price-sweep crossover analysis at fourteen Cu-price levels, producing the strategy-preference curve.

A mine plan is not "the LOM schedule" — it is the conditional NPV distribution given a sequence of decisions that the geology and the market together police. Monte Carlo simulation in ModelRisk is what makes the comparison between two plans honest: same future, same draws, two strategies, and a left tail you can actually defend in front of the people writing the cheques.