Five layers
The bottom two layers are installed once per enterprise; the top three are configured per industry. Select a layer to see its composition.
Digital twin of the source
The asset is decomposed into elementary accounting units: leaching cell and well for in-situ leaching, block and face in ore mining, compartment and cutting area in forestry. Each unit carries an approved reserve, recovered fact, remainder and regime.
Click a cell to inspect its state. Values are illustrative.
What the model stores and calculates
- Reserve
- Approved reserves and resources per accounting unit, with version and approval date
- Fact
- Actually recovered volume and grade, collected from instruments and assay reports
- Remainder
- Reserve minus fact at any date; the full change history is retained
- Regime
- Flow, pressure, reagent concentration, L:S ratio, solution composition, downtime
- Hydrogeology
- Filtration properties, well pattern, movement of pregnant and barren solutions
- Provenance
- Every figure references the instrument, document or test report it came from
The domain model differs by industry; the calculation core is shared. Moving into a new industry changes the description of the accounting unit and the set of sources, not the architecture.
Predictive analytics
Automation shows what happened. Machine learning shows what will happen. The model is trained on the historical data of the specific asset and returns a confidence band rather than a single point — otherwise a forecast cannot be used in production planning.
Illustrative model. On site the curve is built on the actual regime history, site hydrogeology and assay data.
What the ML layer forecasts
- Recovery
- Recovery and yield profile over the planning horizon under current and alternative regimes
- Excursions
- Probability of technological solutions leaving the design contour, from observation well data
- Anomalies
- Behavioural deviations in shipment, intake and short weight — by behaviour, not a fixed threshold
- Failures
- Failure forecast for pumping, conveying and processing equipment from operating hours and regime
- Quality
- Expected grade and settlement weight of a lot before the test report arrives
- Validation
- Back-testing on closed periods before the model enters the production contour
- Research partner
- The calculation architecture and the forecasting-model approach are developed jointly with Henan Polytechnic University — China's first mining university
Volume and weight balance
The platform reconciles three figures: extracted at the source, recorded in transit and received at intake. The discrepancy is localised down to shift, section, machine and operator.
Calculated from your inputs. The platform does not remove the discrepancy — it shows at which interface and on which shift it arises.
How a discrepancy is found
- Metering points
- Every handover between owners and sites is described as a metering point with an instrument and an owner
- Tolerance
- Each point carries an allowed error: scale class, moisture, natural loss, measurement method
- Reconciliation
- Figures are matched by lot and by period; anything beyond tolerance is recorded, not averaged away
- Analysis
- The discrepancy is broken down by shift, section, machine and operator, and becomes a task
- History
- Every reconciliation and explanation is stored alongside the calculation
Language interface
An engineer asks a question in plain text. The platform looks for the answer in production data and enterprise regulations and returns it with a link to the source and the query parameters. Models are open-source and run on the client's own hardware.
Illustrative interface. The data in these examples is synthetic — on site, answers are built from your telemetry, test reports and regulations.
Confirmation loop
The platform performs no irreversible action on its own. Every control decision passes three steps, each written to the log.
- DraftCALCULATION AND IMPACT
- ConfirmationOPERATOR · ROLE AND RIGHT
- ExecutionCOMMAND TO THE SYSTEM
Modules
Thirteen types on one data contour. The first five form the working minimum, including the forecasting module; the rest are added as needed. Expand a row to see input, output and integrations.
Data sources and integrations
The list of interfaces for each asset is fixed during the survey. Below are the types of sources the platform works with.
Instruments and equipment
- Flow meters, pressure and level sensors at process nodes
- Vehicle, rail and belt scales
- On-board machine systems: harvesters, excavators, loaders
- Fleet telematics and trip records
- Sampling and laboratory testing: grade, moisture, ash content
- Satellite and UAV survey of boundaries and stockpiles
Enterprise systems
- Existing automation level and the process data historian
- Accounting contour: receipts, movements, shipment, cost
- Laboratory information system and quality logs
- Regulatory base: procedures, designs, instructions, orders
- Primary documents for lots and contracts
- Manual entry where no instrument exists: forms per metering point
Statutory systems
- National timber accounting and transaction system
- National grain lot traceability system
- Precious metals and stones traceability system
- Veterinary certification system
- Sectoral fisheries monitoring system
- National environmental information system
Deployment and security
The product is designed for enterprises that restrict data leaving the perimeter. No external cloud services are used inside the contour.
Where it runs
- Full deployment on the client's own hardware
- Open-source language models hosted locally
- Operation without a permanent external network channel
- Updates delivered as a package, with no remote access into the contour
Access and data
- Role model: access by asset, section and function
- Encryption of storage and transfer inside the contour
- Immutable log: queries, answers, confirmations, submissions
- Corporate data never leaves the enterprise perimeter
Industry requirements
- Compatibility with internal information security procedures
- Independent verification of calculations from the log
- Environment separation: test, pre-production, production
- Handover of source data to the client in open formats
How deployment runs
Work starts with a survey, not an installation: until metering points and tolerances are described, any balance figure is arguable. Stage durations depend on the number of assets and the state of the instrument park and are fixed in the pilot programme.
Survey
We describe the material movement chain, metering points, instruments, data formats, procedures and tolerances.
Connection
We collect data from instruments and enterprise systems, populate the register of objects and load historical periods.
Calibration
We tune the source model and tolerances, train and validate forecasting models, reconcile against a closed period.
Operation
We move to regular balance, forecasting and reporting, enable the language interface and train the staff.
We will show the platform on your data
A demo takes about an hour: we walk through your accounting chain and show the balance calculation, the forecasting layer and the language interface on a close industry example.