ChainVision: From Pit to Port

Probabilistic
planning for
mining value chains.

ChainVision is an integrated framework for hierarchical probabilistic optimisation of mining and mineral processing operations. From iron ore to lithium. Calibrated to operator data.

P50 P10 P90 MINE Site A P50 / P90 / P10 MINE Site B P50 / P90 / P10 STOCKPILE Blend COMMINUTION Crush & grind PROCESSING Separation P50 / P90 / P10 REFINERY Primary P50 / P90 / P10 LOGISTICS Port P50 / P90 / P10 SALE SALE ILLUSTRATIVE · VALUE CHAIN AS A GRAPH OF STOCHASTIC NODES Each node carries its own probability distribution. Uncertainty propagates through the chain to outputs that matter: production, grade, recovery, throughput, cost, revenue.
Calibration accuracyPilbara Pilgangoora ±1%
Calibration accuracyLynas Mt Weld ±3%
Demo library14 scenarios across commodities
CoverageAll commodities, all chain stages
OutputsProbability distributions, not point estimates
OriginAustralian deep-tech
Calibration accuracyPilbara Pilgangoora ±1%
Calibration accuracyLynas Mt Weld ±3%
Demo library14 scenarios across commodities
CoverageAll commodities, all chain stages
OutputsProbability distributions, not point estimates
OriginAustralian deep-tech
The Problem

Your chain is
connected.
Your tools
are not.

Upstream decisions propagate downstream in ways that aren't modelled until they become operational problems. A mine sequence change affects ore grade. Grade affects recovery. Recovery affects concentrate quality. Concentrate quality affects refinery throughput.

Planning teams manage this complexity with disconnected tools, manual spreadsheets, and point-estimate scenarios. The result: capital decisions made without understanding the probability distribution of their downstream consequences.

P50
Most plans report a single number for production, recovery, throughput, or revenue. The P10 and P90 exist; they just aren't calculated.
Site, corporate, and executive views drift apart when each layer recalculates from its own assumptions. Probabilistic coherence across levels is hard.
1%
A single percentage point of recovery at a major operation is worth a great deal annually. Knowing where it lives requires modelling the whole chain, not one node at a time.
ChainVision

The probabilistic view your value chain is missing

Configure the chain once. Run thousands of simulated futures. Understand the full probability distribution of outcomes across every node, every product stream, every planning period.

ChainVision · illustrative view
Primary product · P50
distribution
P10 / P50 / P90 across horizon
Probability of meeting target
distribution
under shared stochastic drivers
Revenue · P50 / month
distribution
multi-product, integrated
Confidence-weighted constraint
Across simulated futures
which nodes constrain most often
Primary product · P10 / P50 / P90 over the planning horizon
P10 P50 P90 Budget
Mine
Site A
P50 / P90 / P10
throughput distribution
Comminution
Crush & grind
P50 / P90 / P10
throughput distribution
Processing
Separation
P50 / P90 / P10
recovery distribution
Logistics
Port
P50 / P90 / P10
throughput distribution
Capabilities

Built for the full complexity of real operations

01

Stochastic value chain

Operations are modelled as a graph of stochastic nodes (mine, processing, refinery, logistics) with full uncertainty propagation through the chain. Outputs are probability distributions over the metrics that matter (production, grade, recovery, cost, throughput), not point estimates.

02

Multi-product streams

Most mining operations produce more than one product. ChainVision treats every product stream as part of the integrated value chain, with revenue distributions reflecting genuine operational uncertainty rather than deterministic averages.

03

Confidence-weighted constraints

Across thousands of simulated futures, ChainVision identifies which nodes constrain the chain most often. Confidence-weighted constraint analysis surfaces capital priorities that are invisible to deterministic planning.

04

Calibration to operator data

ChainVision is designed to be calibrated against operator actuals, not run on textbook defaults. Calibration accuracy has been demonstrated against publicly available production data: Pilbara Minerals Pilgangoora ±1%, Lynas Mt Weld ±3%.

05

Portfolio-level value chains

Site-level uncertainty aggregates into corporate-level views under shared stochastic drivers like commodity prices and FX. A publish-subscribe layer keeps planning views consistent across the organisation.

06

Outputs built for planning conversations

Each simulation produces interpretable outputs designed for planning conversations, not statistical artefacts that require translation. Probability distributions, confidence-weighted constraints, and scenario comparisons are presented in the language planners and executives already use.

How it works

From configuration to insight in minutes

01

Configure the chain

Build the operation's node topology (mine, processing, refinery, logistics) and connect the streams. Set parameters from operating data or testwork, with uncertainty distributions defined per parameter rather than reduced to a single value.

02

Calibrate and simulate

ChainVision is calibrated against operator actuals where available, and runs across thousands of simulated futures. Uncertainty propagates through every node so that downstream metrics inherit the full distribution of upstream variability.

03

Plan with the distribution

Outputs are probability distributions over production, recovery, throughput, cost, and revenue. Confidence-weighted constraints and scenario comparisons surface the decisions worth having a conversation about, at site, portfolio, and executive level.

Validation

Calibrated against publicly available operator data

Two named Australian mining operations, both validated against publicly available production data.

±1%
Pilbara Minerals · Pilgangoora
Calibration accuracy
±3%
Lynas · Mt Weld
Calibration accuracy
Planning horizons

Two horizons.
One framework.

Budget & capital planning

Model the probability distribution of outcomes over the full planning horizon. Understand the achievability of an annual budget before committing to it. Identify which nodes drive the most variance in production and where capital investment has the highest expected return.

  • Probability distributions per metric, per period
  • Probability of meeting target, not a single P50
  • Confidence-weighted constraints across the chain
  • Scenario comparison at site and portfolio level
  • Outputs framed for planning conversations
PRIMARY PRODUCTION · ILLUSTRATIVE Budget P50 Jul Sep Nov Jan Mar May Jul TARGET PROB. P(meet) distribution-based

In-period tracking

Compare in-period actuals against the simulated forecast distribution as the period unfolds. Variance is interpreted in the context of the original uncertainty band, not as a deterministic miss against a single number, and the chain-wide implications of trends propagate through the same model used to set the plan.

  • Actuals interpreted against the original distribution
  • Forward projection from current actuals
  • Node-level deviation in chain context
  • Re-forecast as new actuals arrive
WEEKLY ACTUALS vs PLAN · ILLUSTRATIVE Plan W1 W2 W3 W4 W5 W6 W7 W8 W9 W10 ← Forecast → On/ahead Warning Behind
About

Copula Labs.

Copula Labs is an Australian deep-technology company applying advanced stochastic methods to industrial planning problems. ChainVision is our first product.

Founded by Yazan Arouri, PhD in optimisation, with 8+ years across stochastic modelling, mathematical optimisation, and machine learning in industrial planning. Fulbright Scholar at the University of Texas at Austin. Based in Melbourne, Australia.

Plan with the
distribution.

If you operate in Australian mining, or advise, fund, or partner with companies that do, we'd be glad to talk.

yazan.arouri@copulalabs.com