Resource Model vs Mine Reconciliation: Why Your Model Doesn't Match Reality

I was called into a gold mine in Kalimantan two years ago that had been running for 18 months. The resource model said the first year should deliver 1.4 g/t Au at 2.1 million tonnes. The mill reported 1.05 g/t at 2.0 million tonnes. That’s a 25% grade shortfall — about $30M in revenue the model promised and the rock didn’t deliver.

The mine’s response was to blame the resource model. The resource geologist’s response was to blame the mining dilution. The metallurgist’s response was to blame the sampling at the mill. Everyone was partly right, which is the uncomfortable truth about reconciliation: the gap between model and reality is almost always multi-causal.

This post is about how to actually diagnose and close that gap. Not the theory — the field practice.

Why models don’t match mines

A resource model is an estimate. A mine is a measurement (noisy, but a measurement). The gap between them comes from five sources, and you have to identify which ones are active on your project before you can fix anything:

1. Smoothing and conditional bias

Kriging smooths. By design, it underestimates highs and overestimates lows. If your model has a 1.4 g/t mean but the deposit has a high nugget effect, the actual mining will recover the highs (which the model understated) and the lows will be left in the wall (which the model overstated). The net effect on the mill feed depends on the cutoff grade and the selectivity.

The fix: tighter search neighborhoods, smaller block sizes, and proper compositing that preserves the grade variance rather than averaging it away. The conditional bias can also be diagnosed in the block model validation — the slope of regression in cross-validation tells you if smoothing is a problem.

2. Block size vs SMU

The block size in your resource model is almost certainly larger than the Selective Mining Unit (SMU) — the smallest volume the mine can actually select and route. A 10m × 10m × 5m block model on a deposit mined with 5m benches and 2.5m flitches means the mine is making selectivity decisions at a finer scale than the model represents.

The grade-tonnage curve at the block size is more optimistic than the grade-tonnage curve at the SMU. The blocks average out internal high and low grades; the SMU doesn’t, so the SMU sees more extreme grades — more high-grade ore AND more low-grade waste. At any given cutoff, the SMU recovers less ore at higher grade than the block model predicts.

The fix: re-estimate (or post-process) the model at the SMU block size, or apply a correction factor derived from a conditionally-simulated grade-tonnage curve. Don’t pretend a 10m block is a 2.5m SMU.

3. Dilution

Mining dilution is material below cutoff grade that gets mixed into the ore stream during blasting, loading, and hauling. Two types:

• Internal dilution — low-grade material within the block that can’t be separated at the mining scale. This is a function of block size vs SMU (above).
• External dilution — waste from outside the block, picked up because the blast pattern, the loader operator, or the dig line isn’t precise.

External dilution on Indonesian gold mines typically runs 5–15%. On narrow-vein epithermal deposits in Sumatra, I’ve seen it hit 25–30% when the vein is <2m wide and the mining equipment is 3m+ buckets. The resource model assumes zero external dilution. The mine delivers the full dilution to the mill.

The fix: apply a dilution factor to the resource model when converting to reserves (JORC and KCMI both require this), and — operationally — tighten the grade control to reduce the dilution at source. Blast hole sampling, ore marking, and loader GPS all reduce external dilution.

4. Ore loss

The mirror image of dilution: ore that’s left in the pit or underground (in the wall, in the floor, mixed into the waste dump) and never delivered to the mill. Ore loss on typical Indonesian open-pit gold mines runs 3–8%.

Ore loss and dilution move together. Tight grade control reduces both. Loose grade control increases both. The reconciliation has to account for both — a model that “matches” because ore loss offsets dilution is not actually matching.

The fix: reconciliation that tracks both the delivered grade (mill) and the delivered tonnage vs the mined volume (survey), so ore loss and dilution are quantified separately, not lumped.

5. Grade control vs resource model

This is the one most resource geologists miss. The resource model is built at ~40m drill spacing. The grade control model is built at ~5m blast hole spacing. These are not the same model and they should not be expected to agree at the block level.

The grade control model is more accurate for the short-term mining decision. The resource model is more representative for the long-term resource statement. The reconciliation question isn’t “does the resource model match the mill?” — it’s “does the resource model match the grade control model, and does the grade control model match the mill?” Two reconciliations, not one.

Reconciliation factors — the numbers that matter

A proper reconciliation tracks three factors, typically on a monthly or quarterly basis:

• F1: Resource model vs grade control model. Measures the resource estimation accuracy.
• F2: Grade control model vs mill. Measures the mining and sampling accuracy.
• F3 = F1 × F2: Resource model vs mill. The overall reconciliation factor.

Each factor has a grade component and a tonnage component:

A well-run mine has F3 between 0.90 and 1.10 on both grade and tonnage. Outside that range, something is systematically wrong and the gap needs diagnosis, not acceptance.

The Kalimantan case — what actually happened

Back to the mine I mentioned. 25% grade shortfall over 18 months. We ran the reconciliation:

• F1 (resource vs grade control): 0.88 on grade. The grade control model was returning 12% lower grades than the resource model. The resource model had over-smoothed — a wide search, no top-cut, and 10m blocks on a nuggety vein. The high grades that the resource model smeared across blocks weren’t there at blast-hole density.
• F2 (grade control vs mill): 0.95 on grade. The mill was 5% below the grade control model — acceptable, mostly dilution from the loading unit.
• F3 (resource vs mill): 0.88 × 0.95 = 0.84. The mill delivered 84% of the modeled grade. Matches the 25% shortfall (1.4 × 0.84 = 1.18, vs 1.05 reported — close, the rest was rounding and moisture).

The diagnosis: 70% of the gap was the resource model (F1), 30% was mining dilution (F2). The fix was to rebuild the resource model with a tighter search, apply a top-cut, and re-estimate at the SMU block size. The mining team tightened the ore marking and the loader dig lines. Six months after the rebuild, F3 was 0.96.

This is one of the resource estimation mistakes that cost mining companies millions — building a model that’s smooth and defensible at audit but doesn’t represent the selectivity the mine actually achieves.

Closing the loop — the grade control feedback

The reconciliation is only useful if it feeds back into the resource model. The loop:

1. Mine the block.
2. Survey the mined volume (pit survey, typically monthly).
3. Sample the blast holes (grade control model).
4. Reconcile grade control model vs mill (F2).
5. Reconcile resource model vs grade control model (F1).
6. Update the resource model — where F1 is consistently off, the resource model parameters need revisiting (search, top-cut, variogram, domains).
7. Update the reserves with the reconciled dilution and ore loss factors.

Most mines I’ve reviewed in Indonesia do steps 1–4. Few do step 5. Almost none do steps 6–7 — the resource model gets built once at feasibility and never reconciled back. By year 3, the model and the reality have diverged so far that the model is useless for planning, and the mine runs on grade control alone until the next resource update.

The discipline is to run F1 reconciliation quarterly and feed it back into the resource model annually. This is also one of the block model validation steps — the validation isn’t done at sign-off, it’s done continuously through the mine life.

Practical reconciliation — what to track

For every reconciliation period (I recommend monthly), track:

• Mined tonnes (from pit survey — volume × density)
• Milled tonnes (from the mill weightometers — dry tonnes)
• Milled grade (from the mill head assay — typically a weighted average of shift composites)
• Contained metal (milled tonnes × milled grade — the number that matters for revenue)
• Grade control model tonnes and grade for the same period
• Resource model tonnes and grade for the same period

Calculate F1, F2, F3 for grade, tonnage, and metal. Plot them over time. A stable reconciliation is a flat line at ~1.0. A drifting reconciliation (F3 declining month over month) is a model that’s losing touch with the orebody — fix it before it costs you.

How Orebit GeoSuite helps

The GeoSuite Reconciliation module (in development, part of the Phase 03 Resource Estimation suite) supports:

• Period reconciliation — monthly and quarterly, with F1/F2/F3 on grade, tonnage, and metal
• Reconciliation dashboard — time series plots of the three factors, with control bands at 0.90 and 1.10
• Grade control model import — blast hole assays, surveyed mining polygons, mill feed data
• Resource model update workflow — flag the resource blocks where F1 is consistently off and queue them for re-estimation
• Dilution and ore loss tracking — separate quantification, not lumped
• Audit-ready reconciliation report — period summary with all factors, charts, and commentary

The existing Resource Estimation module handles the model-building side — variography, kriging, validation — and the reconciliation module closes the loop back from the mine. Together they cover the full estimate-to-mill lifecycle.

Try Core free → · Get the bundle — Rp 99K → (single modules Rp 49K · one-time · yours forever)

Bottom line

Your model doesn’t match the mill because models smooth, blocks are bigger than SMUs, mines dilute, ore is lost, and the resource model and the grade control model are different things. The fix is not to build a “better” resource model — it’s to measure the gap, decompose it into F1 (model accuracy) and F2 (mining accuracy), and feed both back.

Run the reconciliation monthly. Diagnose the dominant factor. Fix the model or fix the mining, depending on which is broken. A model that’s reconciled annually against the mill is alive. A model that’s never reconciled is a feasibility study fossil — and the mine will quietly stop trusting it.

Reconciliation gap you can’t diagnose? Email hello@orebit.id with the F1/F2/F3 numbers and the deposit type — I’ll tell you where I’d look first.