Grade Control and Reconciliation Software Problems: The 7 That Cost Real Money, and How to Avoid Them

Why does the build start before the definitions are agreed?

The biggest scope failure in reconciliation is starting development while the meaning of a tonne is still contested. It happens because the disagreement is invisible until you try to automate it. Everyone uses the word tonnes, everyone assumes their own basis, and the differences only surface when a system forces two numbers into the same column.

The decisions that have to be made first are not technical. Wet against dry. In situ against broken. Which density applies where, and whether the survey conversion uses the model's density or the plant's assumption. Where and how moisture is measured. What happens to a stockpile that spans a period boundary. Those are technical services and metallurgy decisions, and a development team cannot make them, but a development team will absolutely encode whichever answer it hears first.

The consequence of skipping this is a build that argues with the plant in a new way rather than resolving the argument. The consequence of doing it is a shorter project, because two workshops before kickoff save more time than any technology choice.

Then make the definitions impossible to lose. Every quantity in the system carries its basis, and every conversion is explicit and logged with its source. If a survey volume becomes tonnes at a density of 2.68, that number sits on the record with where it came from. It sounds pedantic until the first time a factor moves because somebody changed a density assumption in a spreadsheet cell and nobody could see it.

What goes wrong when block models and historical factors are migrated?

Reconciliation compares against a model that is itself a moving object, and that is the migration trap specific to mining.

Models get re-estimated, reblocked, re-domained and revised as new drilling arrives. Depletion happens on a survey schedule that does not match the model update schedule. If the comparison silently uses the current model rather than the model that was current when the ore was mined, your history rewrites itself every time the resource geologist publishes, and last year's reported factors can no longer be reproduced.

Historical factors bring their own problem. Four years of spreadsheet tabs contain numbers whose inputs are gone: the model version has been superseded, the density assumption was edited in place, and the stockpile balance at the period boundary was a judgement somebody made on a Friday. Loading those figures as history gives you a trend line built on five different methodologies, and any conclusion drawn from it is unsafe.

The fixes are versioning and honesty. Treat models as first class objects with effective dates, and pin every reconciliation record to the model version it was computed against, so reruns are reproducible and you can also deliberately rerun a past period against the current model to isolate what model change alone did to the number. Load historical factors only where the inputs survive, mark the rest as legacy figures rather than comparable data, and start the trustworthy series at the point the definitions were agreed.

Why do the historian, laboratory and fleet integrations break after launch?

The plant, the laboratory and the mine were instrumented for their own purposes, none of which was reconciliation, and each drifts differently.

The plant historian holds tag values rather than business events, so a shift total is something you compute rather than read. Tags get renamed during a control system upgrade, a weightometer is replaced and reports through a new tag while the old one keeps returning a stale value, and a totaliser resets on a schedule that suits the operator rather than your reconciliation period. None of that errors.

Laboratory data drifts when methods change, when an instrument is replaced and reports to a different precision, or when results arrive against a sample identifier that no longer matches how ore control is keying its samples. Fleet systems drift when a truck is reassigned, when onboard scales are recalibrated on some units and not others, and when a dispatcher changes a destination code.

The controls that hold this together are provenance and reconciliation rather than better connectors. Every value carries its source tag, its method and its timestamp, so an instrument change is visible instead of absorbed. Draw and production quantities reconcile daily and a mismatch raises a queue rather than being smoothed at month end. Alert on staleness, since a tag returning the same number for nine days is the classic failure and it looks perfectly healthy in a chart. And ask any prospective developer to name the specific historian, laboratory system and planning package they have integrated, and what broke.

What happens when ore control at the digger and stockpiles are not covered?

Two gaps recur, and between them they usually explain most of an unattributed factor.

The digger is the first. A dig line is marked on a plan, translated to flagging or an in cab screen, and then reality intervenes. Blast movement shifts the ore boundary by several metres while the markup was based on pre blast positions. The operator on night shift cannot see the flagging. A truck tips to the wrong stockpile because the dispatcher was busy. Each of those is ore loss or dilution and none of them is recorded anywhere reconciliation can see. A per load record carrying the source polygon, the ore control classification, the instructed destination and the destination actually tipped turns mis tips from an anecdote into a count per shift, and counts get managed. Where blast movement monitoring is in use, the moved dig lines have to be the ones both the digger and the reconciliation use, otherwise you are comparing a plan nobody executed against an outcome nobody predicted.

Stockpiles are the second. Grade through a rehandled stockpile with partial reclaim is genuinely difficult, every site does it differently, and it is the most common thing left out of a first release. Leaving it out is defensible. Leaving it out silently is not, because the balance movement then lands in the unexplained residual and makes every other attribution look worse than it is.

Should you build custom or configure what you already own?

If you are a single site, single commodity operation already standardised on Datamine with disciplined data and a stable factor framework, do not build. Reconcilor is purpose built for this and will get you there faster and cheaper than a discovery phase. The same logic applies if you are committed to Micromine or Hexagon MinePlan and your reconciliation needs are modest next to your planning needs.

Before assuming custom, find out what you already hold and do not use. Historians commonly retain years of tag history nobody has queried. Fleet management systems often record instructed against actual destination already, unreported. Laboratory systems usually export quality control results that are being reviewed monthly by hand. A surprising share of the reconciliation gap is data that exists and is not joined, and joining it is cheaper than collecting it again.

The build case appears when two or more of these are true. Factors are computed in a spreadsheet only one person can run, and that person is not junior. Inputs live across systems your modelling suite cannot read. The factor has been off for more than two quarters with no attributed cause. You operate several sites whose factor definitions differ enough that group comparison is meaningless. Or you want daily reconciliation rather than monthly, which is where the operational value sits.

How do hidden costs get into the quote?

Reconciliation quotes go wrong in five places.

• Source system openness. A plant historian, a laboratory system and a mine planning package are three separate integration problems, and an older installation may need extraction engineered rather than configured.
• Stockpile modelling. Tracking grade through rehandled stockpiles with partial reclaim is difficult, site specific, and frequently assumed to be included.
• Multi site or multi commodity scope. Factor definitions must be normalised before they can be compared, which is a technical services workshop rather than a coding task.
• Underground scope. Development and stope reconciliation carry their own logic that does not transfer from open pit, and adding it later is close to a second project.
• Undecided definitions. The single biggest driver we see, because every unresolved question about basis becomes rework once someone in the plant disagrees with a number.

Digital Heroes delivery experience puts a reconciliation engine covering ingestion from the resource and grade control models, survey, fleet and plant historian, explicit basis handling, versioned model pinning and monthly plus daily factors with attribution at $70,000 to $150,000 over 10 to 16 weeks. A full platform adding ore control markup and digger destination capture, laboratory ingestion with automated quality control evaluation, stockpile balances by material type and multi site rollups runs $180,000 to $420,000 across 6 to 12 months.

What separates a build that works from one that fails here?

Working builds produce an attribution rather than a number. Of the gap, this much is mis tipped loads, this much is timing between survey and model depletion, this much is stockpile balance movement, this much is evidenced scale drift, and this much remains unexplained. The residual is the honest headline and it should shrink as instrumentation improves. A single factor tells you there is a problem. An attribution tells you whether to spend money on grade control drilling, dig line marking, ore loss controls or scale calibration.

They run daily, not only monthly. A monthly factor is a post mortem. A daily factor with attribution behaves like a control system, catching a scale drifting on three trucks within a week rather than at quarter end.

They use machine learning for one narrow, testable job: anomaly detection over payload and assay distributions, which surfaces instrument drift and sampling bias earlier than a scheduled calibration will. Grade prediction sold on top of a data set this messy is not worth funding until the data set stops being messy.

And they settle ownership in writing before kickoff, covering the repository, the cloud accounts and the right to hire another firm. Reconciliation history is evidence supporting public reporting under codes such as JORC or NI 43-101, and it should never sit in a system you cannot get it out of.