A calculation is admissible proof, and the law says on what condition
Predictive microbiology has been a working discipline for forty years. What has been missing is the step that makes a model trained in a laboratory flask apply to a raw-milk cheese, and the document that carries the answer to an inspector
Four ways to prove a shelf life, and what each of them costs
A manufacturer may prove compliance by studying the characteristics of the product, by storing real batches for the whole shelf life, by deliberately inoculating the product, or by calculation
All four are admissible. Three of them are slow or expensive, and the fourth carries a condition that public calculators do not meet in most categories
Study of characteristics
A demonstration that the product cannot in principle let the pathogen multiply
- Cost per item
- low
- Time
- days
- Needs a microbiologist
- yes
- Covers
- part of a range
Durability study
Real batches are stored for the full shelf life and tested at intervals
- Cost per item
- €1,500–4,000
- Time
- 8–16 weeks
- Needs a microbiologist
- a laboratory
- Covers
- one recipe version
Challenge test
The product is deliberately inoculated with a controlled dose of the organism
- Cost per item
- €1,500–4,000
- Time
- 8–16 weeks
- Needs a microbiologist
- a laboratory
- Covers
- one recipe version
Predictive modelling
The behaviour of the organism is computed from the properties of the product
- Cost per item
- €250–900
- Time
- 3–7 days
- Needs a microbiologist
- ours, not yours
- Covers
- every version
What the matrix correction actually does to the numbers
A model is judged on two figures. The bias factor says which way it is systematically wrong, and the accuracy factor says by how much it is wrong on average
Switch the correction on and off below, and watch what happens to both
Validation sample
Every point is one product tested in a laboratory and also computed by the model
A bias factor below one would mean the model systematically underestimates how fast the pathogen grows. In a safety task that is inadmissible whatever the accuracy, so a category that has not passed the threshold keeps a widened interval and a larger margin until enough data accumulates
| Indicator | Target | Without correction | What it means |
|---|---|---|---|
| Bias factor | 1.00–1.10 | unconstrained | Deviation is permitted in one direction only, towards overstating growth |
| Accuracy factor | 1.15 or better | 1.35–1.50 | The average size of the divergence between prediction and observation |
| Safety margin | 10–15 % | 30–40 % | The margin is proportional to the width of the confidence interval |
| Days recovered | 3–6 | none | On a typical chilled product declared at 21 days |
Nobody stores food at a constant four degrees
The classical calculation assumes one temperature for the whole shelf life. The real chain is cooling after heat treatment, a warehouse, a lorry, a goods-in bay, a display cabinet, a journey home and a domestic refrigerator
The core integrates growth segment by segment and carries the state of the lag phase across the transitions, because an organism that started growing during a deviation does not begin waiting again when the temperature comes back. Below, one chilled ready meal at pH 6.0 and water activity 0.962 is put through three levels of knowledge about its own chain
The horizontal axis runs from production on the left to the domestic refrigerator on the right, and the width of each stage is its share of the total time
Moving from the first level to the third removes the margin step by step. The more precisely the chain is known, the less has to be allowed for the unknown, and the more days come back to the manufacturer
The engine answers what you are obliged to do, not what is interesting
Everything up to this point is science. This step is law: the product is assigned to a category under the criteria of the regulation, and the category decides the entire regime of proof
- The justification rests on the characteristics themselves, supported by the literature
- No laboratory test is prescribed at all
- The dossier is issued in 3 to 7 days
- Cost per item falls to €250–600, six to seven times below current practice
- The engine finds the highest-risk item of the category and tests that one
- Instead of five tests across a range, one confirmatory test is carried out
- It chooses between a characteristics study, a durability study and a challenge test, and sets the batches and sampling points
- The sampling protocol is written so that a laboratory accepts it without rework
Both branches end in the same artefact: a dossier with the justified shelf life, the reasoning, the verification plan and the audit trail. What differs is whether a laboratory appears in it, and how many times
Four texts hold the product up
Demand here is not created by fashion or by a trend in packaging. It is created by four instruments, and two of them changed within a single year
Regulation on microbiological criteria
It obliges a manufacturer to prove in documented form that a ready-to-eat product stays safe throughout its stated shelf life, permits predictive modelling as a means of that proof provided the model is established for the food in question, and requires reasonably foreseeable conditions of distribution, storage and use to be taken into account
The amended criterion for Listeria
For products capable of supporting growth, absence in 25 grams now applies throughout the whole period the product is on the market, unless the manufacturer has demonstrated to the competent authority that the level will not exceed 100 colony-forming units per gram. Without that demonstration the batch falls under a criterion chilled goods cannot meet in practice
Commission and Reference Laboratory guidance
The guidance of the European Commission in its December 2025 version, the February 2026 amendment to the technical guidance of the European Union Reference Laboratory and standard ISO 20976-1 set the protocols within which the engine builds a verification plan, and confirm that compliance may be shown by challenge tests, by predictive models or by physico-chemical characteristics supported by the literature
The food loss prevention laws
The Spanish law and the stricter Catalan law establish obligations on the loss prevention plan, on the hierarchy of priorities for handling surplus and on annual reporting, with penalties reaching €500,000. The loss plan module produces exactly the reporting a company is obliged to keep, from the same figure the safety justification is built on
Put a question to the scientific team
If you work with a matrix we have not named, or you have a validation result that disagrees with our figures, we would rather hear it than not
Questions on models, calibration, validation protocols and the acceptance of a calculation by an inspection go straight to the scientific director