Run It Twice

Stand at the criterion line with a reading in your hand. It says 2,850 mg/kg and the criterion is 3,000. Do you pass that soil?

The honest answer, with any conventional field tool, is that you do not know — and worse, you have no way to find out until the laboratory report arrives, by which time the crew has gone home. The reading is a single number. It carries no information about how much confidence it deserves.

What closes that gap is not a better sensor. It is a duplicate.

The soil moves under the number

Soil is not homogeneous, and the scale at which it varies is finer than the scale at which we sample it. Split a single bag of coarse sand in two and analyse both halves and the results routinely differ by 10 to 30%. That is not instrument error. Both halves were measured correctly. They were simply not the same material.

On a recent chloride remediation programme, every duplicate pair was recorded and plotted. The spreads ran from ±5% to ±20% across the run — and one pair, FS-02, did something that ought to be required reading for anyone who has ever argued with a field result.

One split of FS-02 sits above the criterion. The other split of the same soil sits below it. Same bag, same operator, same instrument, same minute. Near the line, that is not a sensor miss — it is the soil itself, and no amount of analytical precision will make it go away.

Every field programme in the industry has been making pass-or-flag calls on soil that behaves like this. The difference is whether the programme measured it or assumed it away.

Conventional tools hide it. Measuring it is the breakthrough.

A colour-change strip or a conductivity pen gives you one number per sample and no mechanism for asking how repeatable that number is. There is no duplicate protocol, no calibration record, no way to distinguish a reading that is solidly clean from a reading that happens to have landed on the clean side of a coin flip. The variability is still there — it is simply invisible, and it surfaces later as a failed confirmation nobody can explain.

The alternative is not to eliminate the variability, which is impossible, but to measure it. That is what running duplicates does. Each pair tells the system how variable this site's soil actually is, right now, in this matrix — not a textbook assumption about sand, but this sand, today.

This only works if the process itself is not adding noise. Every sample goes into a single-use extraction tube with a measured conditioning solution and a temperature-controlled extraction, so every sample on every site sees identical conditions. Remove the process as a variable and what remains in the spread between two splits is the soil. That is the measurement you actually want.

From spread to probability

Measured heterogeneity is what powers the number that changes how a field lead works: a probability of exceedance attached to every reading. Instead of a bare value that turns out right or wrong weeks later, each sample arrives with its own odds.

Look at how cleanly the two populations separate. Everything the instrument passed carried a probability of exceedance of 29% or less, and nearly all of those sat at 0%. Everything it flagged carried 53% or more. There is a gap in the middle, and samples that land in it are visible as borderline at the moment they are read — so they can be re-tested, duplicated or stepped out on the spot rather than discovered later.

What this does to the error profile

The effect on the programme's classification record is the part that matters commercially. Across 28 laboratory-confirmed samples, 27 calls matched the laboratory — 96.4% decision accuracy — and there were zero false negatives. The instrument never passed soil the laboratory would have failed. The single disagreement was a false positive: a conservative over-call on soil the laboratory returned just below the criterion.

That is exactly the error you want a screen to make. Against the USEPA field screening limits of no more than 5% false negatives and no more than 20% false positives, the programme sat comfortably inside both. But the more interesting result is the one that does not appear in the summary table: not a single disagreement was a surprise. Each was visible as borderline in the field, at the moment of reading.

The stress test

A fair challenge is whether this holds at a threshold where heterogeneity bites harder. The programme's laboratory-paired results were re-assessed against a 100 mg/kg criterion — thirty times stricter than the programme's own, and close to the method's 50 mg/kg reporting floor.

The two false negatives at the stricter threshold deserve the detail, because they make the case rather than undermining it. Both are splits of a single sample that the instrument read at 68 to 76 mg/kg where the laboratory read 103 — three per cent over the line, inside the soil's own variability band. In live deployment the criterion is set in the application and the probability of exceedance is recalculated against it, so a 68 to 76 mg/kg reading against a 100 mg/kg line would carry a meaningful non-zero probability. It would be flagged for a duplicate or laboratory confirmation, not passed with confidence.

That is the whole argument in one sample. The instrument's job near the line is not to be right every time — no method measuring heterogeneous soil can be. Its job is to know when it might be wrong, and to say so while someone can still act on it.

The One Thing to Remember

The breakthrough is not a more sensitive sensor. It is running duplicates under a standardized process, so the site's own variability becomes a measured quantity — and every reading arrives with the confidence it deserves attached to it. A tool that cannot do this is not making fewer errors than one that can. It is simply unable to see them.