FILE
GC–2026–08
ENGINE
sealed kernel v6.0.0
CLASS.
TECHNICAL · PUBLIC
ACTIONS
211 with DOI
SCOPE
EU 2025/179
211
operational actions with a verified DOI: 108 stated explicitly by the paper and 103 inferred from an evidenced mechanism
0,933
area under the curve in Listeria over 903 genomes; 0.970 for agreement with phenotypic resistance in Salmonella
1.337
genomes downloaded by accession and re-measured with the kernel's real detectors, with zero errors
< 200 ms
to place an isolate against thirty-three thousand epidemiological profiles
+
+
+
+
What makes it unique
Three things that are not found together anywhere else: risk is measured from the isolate's own markers instead of being inherited from its clonal complex; tracing crosses global genomic surveillance with RASFF alerts on the same food matrix; and the result comes out as an executable action plan, not a gene list.
Two hundred and eleven actions curated by exhaustive literature work with DOI verified one by one, triggered by the specific marker the strain carries: presence, absence, complete island, truncated gene or point mutation.
01
The state of the art
Where the others hand over a list, GIFCloud hands over the work order.
Every tool in the field solves its own stretch well, and GIFCloud builds on several of them. The difference is the next stretch: turning the genome into a plant decision with its dose, its frequency and its DOI.
The consequence has been measured: the 35.5% of E. coli from food carries extraintestinal infection markers that a STEC-focused programme ignores entirely, and 26% of Salmonella serovars were listed as low risk while having hospitalisation rates of 31 to 64%.
02
Intrinsic risk of the strain
Not what its lineage usually does. What this isolate can do.
Standard practice attributes risk by clonal complex membership: the strain inherits its clone's reputation. Here risk is built from the determinants the isolate's genome actually carries.
The ecological strategy, which only this system issues
Crossing virulence with persistence yields a reading no other tool gives: nosotrophic, dangerous to the patient but short-lived in the environment; saprotrophic, settled in the plant with reduced virulence; and amphitrophic, virulent and persistent at once. The amphitrophic strain is the one that survives cleaning, colonises the facility and keeps its capacity to cause disease intact: the maximum industrial risk, and the one that justifies a different response.
In the August 2026 revalidation over 436 clinical isolates, 136 come out amphitrophic. Reclassification affects 6 of 436 — 98.6% do not change strategy — and each of those six has a case-by-case explanation: three enter through real SSI-2 in CC121 and CC475, two drop out through chromosomal homologues emrC mistaken for qacH, and one through incomplete SSI-1.
Where the clone already says it all, and it is declared
Resolution above the clonal complex is not uniform and is not sold as such. In hypovirulent clones it is there: qacH and SSI-2 vary within the lineage — 22.9% and 22.6% — and the index adds what the clone does not. In the Maury-Intermediate subset, by contrast, being amphitrophic amounts to belonging to CC7, CC3 or CC5 rather than CC6: there the index adds no resolution, and the report says so.
Likewise, LIPI-3 is fixed at 0% or 100% in ten out of ten major clonal complexes. Every marker enters with its prevalence, its area under the curve and its confidence interval, not with a round claim.
03
Epidemiological tracing
A single layer with global surveillance and recall alerts.
A One Health approach: the strain is compared not only against your own collection, but against public surveillance databases and the European food alert register, in the same query.
The cross-reference that changes the conversation
A plant isolate stops being an isolated datum: it is placed against surveillance profiles, documented outbreaks and RASFF notifications for the same pathogen, the same product category and the same time window. If what lives on the line matches what is being recalled in the matrix you produce, you know before the letter arrives.
The limit of the cross-reference
RASFF publishes no sequences: it contributes an alert layer — hazard, category, notifying and origin country, product and date — not a genetic profile. The match is epidemiological and contextual, never a strain identity, and it is labelled as such in the report.
04
How it works
From genome to work order, step by step.
A continuous index from 0 to 100 with four action levels: critical from 86 to 100, high from 66 to 85, moderate from 36 to 65 and low below 36.
01
The sealed kernel genotypes the isolate
Pure detectors of virulence, persistence, context and resistance over the genome, with a versioned reference set and an integrity check. Sealed means two analyses run under different rules know they are not comparable.
02
The index is weighted by who is asking
An industrial profile prioritising environmental persistence; a clinical one prioritising virulence and clonality; One Health prioritising resistance and clonality; or whichever the user defines. The same genome changes index because the question changes, and the formula with its weights travels in the report.
03
Three interpretation layers in parallel
The first looks up curated operational interpretations for the genes detected. The second runs eight reasoning patterns over the knowledge graph, three of them plant-specific: zone colonisation, sanitisation override and hurdle opportunity. The third formulates hypotheses when the first two are not enough, and every finding is annotated with the layer it came from.
04
The genotype triggers the actions
Every action has its condition: gene present, gene absent — the CRISPR case — complete island or operon, gene truncated by a premature stop codon, or point mutation in the quinolone resistance-determining region. Only those the real genotype justifies fire, and the number of actions scales with what the strain actually carries.
05
It is placed on the epidemiological map
Comparison against the consolidated surveillance base and against RASFF notifications for its pathogen, matrix and window, with supplier focus and the origin engine when several isolates converge.
06
The reports are issued
Four deliverables per strain, each written for its reader, downloadable as PDF. Third-layer hypotheses do not go into the audit report: they go to the curation queue, where an expert approves or discards them.
05
The four reports
A report does not catalogue measurements. It interprets and instructs.
The plant-facing prose is curated, not generated on the fly: deterministic, reviewable and with no latency. And it is declarative — "this strain carries such-and-such marker" — never generic advice in the conditional.
What these figures do NOT license anyone to say
The index is validated in Listeria monocytogenes — area under the curve 0.933 over 903 genomes. In the other species it is the same architecture extended, and tier-by-tier validation is still pending: the reports declare this as provisional by design instead of hiding it.
The distinction is also kept between the reliability of the algorithm, validated at genus level, and that of the composite index in each species. And in the action catalogue, tier 2 — an action inferred from an evidenced mechanism — is never presented as tier 1, which is what the paper states explicitly.
06
The contract
What goes in, what comes out, what it chains to.
In
The genome of an isolate — assembly or reads — with the plant metadata: facility, zone, product and the sanitisation protocol in force. The weighting profile is chosen in that same step.
Out
Index and action level with its breakdown, ecological strategy, epidemiological context with alert cross-referencing, action plan by standard clause, and the four reports as PDFs, archivable with a timestamp.
Chains to
It receives assemblies from HoloGen. The hypotheses it generates go to the curation queue and, once approved by an expert, return as knowledge available to every other module.
07
Technical sheet
Pathogens
The five mandatory under Regulation (EU) 2025/179: Listeria monocytogenes, Salmonella enterica, Escherichia coli across its pathotypes, Campylobacter jejuni and Campylobacter coli. Adding a new one requires no software development, only defining its genomic profile.
Action base
Two hundred and eleven actions: 67 Listeria, 55 Salmonella, 38 E. coli and 51 Campylobacter, plus 49 candidate enrichment genes that would widen operational inference once their detection is wired in.
Intervention axes
Biocide and its rotation, biofilm, pretreatment, sampling plan, strain interaction, additional analytics, process control, clinical management and product risk.
Throughput
Five to ten minutes per genome and eighty to one hundred and twenty genomes per hour in batch, with tabular export of more than one hundred and forty columns for integration with the laboratory management system.
Deployment
A full web platform or a command-line tool to embed in your own pipelines. Both run on local infrastructure or in the cloud, with no dependence on external services for the analysis.
Reproducibility
Sealed kernel with a per-file checksum and a global integrity check. The action engine lives outside the kernel and operates on the already-persisted result: adding actions forces no sample to be reprocessed.
Limit · validation
The composite index is validated in Listeria. In Salmonella, E. coli and Campylobacter it is an extension of the same architecture with tier validation pending, and the reports declare this case by case.
Limit · coverage
The action engine fires on what the kernel already genotypes. Genes in the base that are not yet detected — and the enrichment candidates — do not fire until their detection is wired in with its own reference sequence.
Limit · alerts
RASFF provides no sequences: a match with a notification is epidemiological and contextual, not a strain identity. And submission to the authorities is not automatic: it remains an operator action.
Limit · evidence level
Close to half the catalogue is tier 2: an action inferred from a DOI-backed mechanism and pending final curation. It is flagged on every action and never presented as tier 1.
Action base curated by exhaustive literature work with DOIs verified one by one against Crossref and PubMed. The Stress Islands and Persistence schemes of pubmlst_listeria_seqdef, Institut Pasteur. Maury et al. (2016) Nature Genetics 48:308–313 · Cotter et al. (2008) PLoS Pathogens 4:e1000144 · Vázquez-Boland et al. (2001) Microbes and Infection 3:571–584. Regulation (EU) 2025/179 · Regulation (EC) 2073/2005 · USDA-FSIS 10010.2 · EFSA STEC risk groups · Magiorakos MDR and XDR criteria · IFS and BRC.
The number opens the conversation. The action plan closes it.
BIOTECNO.org · Vitoria-Gasteiz
EU + Codex