Two curves govern your first season. Utilisation is how many days you run. The yield curve is what comes out while you run them — and almost nobody models the second one.
The ramp the published models assume, and the one you will get
A year-two utilisation above roughly 50% is not credible for a first-time single-origin processor without a committed RCN buying facility. The standard model project reports assume considerably more, and a business plan built on their curve will report a shortfall to its lender in year two for reasons that were predictable at financial close.
The yield curve is the one nobody writes down. Your first weeks will give kernel outturn below your steady-state figure, a whole-cutting rate below the machine’s specification, a peeling first-pass rate near the 70% factories report rather than the 80–95% suppliers claim, and a manual finishing load at the top of the 20–40% band.
None of that is a fault. It is the tuning trial continuing under production conditions.
First-season targets that are achievable
No published source gives a first-season yield curve, defect trajectory or availability figure for a cashew plant. The table below is engineering judgement — steady-state figures discounted by the observed gap between vendor claims and reported factory outcomes. It is offered because planning a first season against steady-state numbers is worse than planning against a stated estimate.
| Metric | Steady state | Planning figure for season one |
|---|---|---|
| Production KOR (kg saleable kernel / 100 kg RCN) | 22–24% | 1–2 points below your own tested figure |
| Whole cutting rate | 85–90% | Bottom of the band, or below, for the first weeks |
| Kernel breakage at shelling | Under 3% | Above it while blade settings are found |
| Peeling, first pass | 80–90% | Plan at 70–80% |
| Manual finishing after peeling | 20–40% | Top of the band |
| Capacity utilisation, year two | — | At or under 50% without a committed buying facility |
| Grade mix | Predominantly wholes | Expect more splits and more scorched than your model assumed |
Set targets against these and beat them. Set them against the supplier’s specification and you will spend the season explaining a shortfall.
Calibrate to your buyer’s specification, not the supplier’s demonstration
The supplier demonstrated the machine on nuts it selected, to a standard nobody wrote down. Your buyer has written one down, and it is the only standard that matters.
Get it in writing before commissioning ends, and set the line to it:
- Grade definitions and the defect tolerances against them
- Moisture at packing — typically 3–5%
- Microbiological limits — under AFI specifications, Salmonella negative in 375 g, Listeria monocytogenes in 125 g, E. coli non-detectable
- Aflatoxin built to the ready-to-eat tier rather than the higher “to be sorted before sale” figures much guidance quotes
- Packing atmosphere — commonly a minimum 60% CO₂ with the balance nitrogen, which the trade calls nitrogen flushing even though the specification does not
One process point falls out of the moisture specification and is missed regularly. The dryer takes kernels to 3–4%, humidification puts one to two points back so they peel without shattering, and packed kernel must be 3–5%. Moisture therefore has to come down again between peeling and packing. If your commissioning runs do not include that final conditioning step, your first container arrives out of specification for a reason nobody at the plant can explain.
The first shipment
The sequence is: buyer’s written specification → your production sample → buyer’s sample approval → the production lot → pre-shipment sample drawn from that lot → inspection → documentation → stuffing → sailing.
What usually goes wrong, in the order it happens:
The approved sample is not the shipment. A sample hand-picked from your best hour gets approved and the bulk cannot match it. Draw the approval sample from a real production lot on a documented sampling rule, and retain a sealed counter-sample of every lot you ship.
Documentation lags production. The document set for a kernel shipment is not assembled in a day, and destination registrations run on foreign timetables — US facility registration, and China’s GACC registration, where nuts and seeds sit in a catalogue you cannot self-register into. These belong on the project critical path, not on the shipping clerk’s desk.
Records that were never kept. Metal detection results by lot, moisture at packing, the cutting test on the RCN that made it. A buyer’s first claim is usually met with a request for records, and “we did not log it” reads as an admission.
Moisture and transit. Kernels packed at the wrong moisture, in a container crossing climate zones, arrive as a claim. So does a container stuffed with a broken door seal.
One full container load is the practical minimum, at 14–15 tonnes of kernels per 20-foot container. If your first-season output cannot fill one to a single specification, plan a consolidation with another processor rather than shipping a part load of mixed grades.
The post-commissioning review
Thirty to sixty days after the first shipment, and in writing. Compare four things against what you approved at financial close:
- Actual capital cost against budget, line by line, with the overrun causes named
- Actual yields against the model — production KOR, whole rate, breakage, peeling, grade mix
- Actual headcount and labour cost per tonne against plan, including absence rate
- The schedule — what slipped, why, and what the slip cost
Then take three decisions with it. What changes in next season’s buying specification, now you know what your line does to which calibres. What changes in recruitment timing, from your own learning curve. What goes into the preventive maintenance schedule, from what actually broke.
Circulate it to your lender before the lender asks. A processor who arrives with a candid variance analysis in month four is a processor who gets a second facility.