Our earlier analysis measured which deficiencies predict a rejected dossier. This one asks the sharper question: in the same section, one dossier cleared and another stalled — what was different? We put the 42 dossiers that passed without a single clock stop next to the ones that did not.
In *What EFSA really checks*, we measured which deficiencies predict a non-favourable opinion. That analysis had one blind spot by design: it only looked at dossiers where the deficiency was present. It could tell you that a batch-consistency problem co-occurs with rejection — but not what the dossiers that *didn't* get flagged did instead.
This piece closes that gap. The question here is contrastive: take a single assessment section, find the dossiers that cleared it as submitted and the ones that got clock-stopped over it, and ask what actually separated the two. Not "which gaps are dangerous," but "in this section, what made the difference between a clean pass and a stalled clock."
We started from 239 published EFSA opinions and split them into two poles.
The clean-pass pole is strict: a plain favourable opinion with zero clock stops. Not favourable-with-conditions, not favourable-after-a-data-round — cleared without the panel stopping the clock once. Only 42 of 239 dossiers meet that bar. The clock-stop pole is the 143 dossiers that took at least one formal pause for missing data.
That ratio is worth sitting with. Sailing through is the exception, not the norm. The largest single outcome in the whole dataset is favourable *with conditions* — authorisation, but only after the panel extracted more. A clean pass — the applicant got it right the first time, across every section — happened in fewer than one in five dossiers. Most authorised novel foods were authorised the hard way.
So when we contrast "accept" against "clock stop" below, "accept" means this small, telling group of 42: the dossiers that gave the panel no reason to stop.
For each section we counted how many clean-pass dossiers had it judged adequate as submitted, against how many clock-stopped dossiers carried a deficiency there — and, among those, the single most common category of deficiency. The pattern is consistent: it is almost always one category doing most of the separating.
Toxicological — 16 dossiers cleared it as submitted; 90 clock-stopped ones carried a deficiency here, 59 of them on the genotoxicity battery. This is the widest gap in the data. What separated a clean pass was not a heroic tox package — it was a battery that was complete and run on material that matched the product. Where it broke down, it broke down concretely. In an early GM-maize opinion (ON-50) the record shows the panel "was divided on the need for an additional 90-day rat study with the MON 863 x MON 810 hybrid in order to complete its safety assessment." A missing study the applicant judged unnecessary; the panel judged otherwise.
Compositional — 20 cleared it; 78 clock-stopped, 40 on batch consistency. The dividing line here is whether the panel believes your production is reproducible. One plant-extract opinion (ON-6196) records the panel reading submitted consistency as the opposite of reassuring: "the results of all tested parameters ... are identical for all five presented batches produced at different time points in 2019, which is considered unrealistic." The clean-pass dossiers showed genuine batch-to-batch variation within specification. The stalled ones showed numbers too clean to be independently produced — or too few batches to judge.
Specifications — 23 cleared it; 59 clock-stopped, 49 on specification justification. The near one-to-one overlap between the clock-stop count and this single category is the tightest in the data: when specifications stall a dossier, it is almost always because they are not anchored to the batches actually tested, or a parameter is missing. A clean pass is a specification a sceptical reader could tie back to real analytical data.
Proposed uses — 25 cleared it; 43 clock-stopped, 33 on exposure modelling. The science can be sound and the section still fail, because the proposed conditions of use push intake past a safe level. In one botanical opinion (ON-5002) the panel's verdict was terse: "The Panel considers these MoEs for fortified foods and food supplements as insufficient." What separated a clean pass was an exposure estimate that survived the uses being requested.
Production — 25 cleared it; 42 clock-stopped, 30 on process representativeness. The recurring failure is a process description that does not let the panel confirm the studied material is the commercial material — and, strikingly often, an applicant who simply stops engaging. In one root-vegetable opinion (ON-3954) the record is blunt: "EFSA asked the applicant to provide conditions of the two-step steam heating process. In addition, the applicant was asked to provide the HACCP certificate for the manufacturing process. The applicant did not respond to EFSA's questions." A clean pass and a dead dossier can diverge on nothing more than whether the applicant answered.
Allergenicity — 18 cleared it; 26 clock-stopped, 20 on the allergenicity assessment itself — most often for protein-rich foods from new sources, where the panel expects the question to be addressed even when the applicant considers it remote.
Read down that list and the same shape repeats: one category, per section, does most of the work of separating a clean pass from a clock stop. That is the practical value of the contrast — it tells you where, within each section, the line actually sits.
Four categories in our data have no clean-pass dossier at all — not one favourable-with-zero-clock-stops opinion to hold up as the positive example. They are the frontier of novel food, and they are worth naming precisely:
- Cannabidiol (CBD) — four opinions, none favourable. The worst-performing category in the dataset. - Cultured-cell products — one opinion; authorised, but only after clock stops. - Insect protein — ten opinions. Every one was authorised (with conditions) — mealworm, house cricket, migratory locust. But not a single one cleared without the clock stopping; each took three to seven clock stops to get there. - Precision fermentation — two opinions, neither a clean pass.
The insect story is the honest nuance. It would be wrong to say these dossiers "fail" — they are the proof that new categories can be authorised. But there is no example, yet, of one clearing on the first pass. For an applicant in an emerging category, the realistic planning assumption is not rejection; it is that the clock *will* stop, more than once, and the timeline should be budgeted for it.
We tested how stable this contrast is by splitting the opinions into two random halves and re-deriving the discriminator in each. The category of deficiency that separates accept from clock-stop in a section — genotox battery, batch consistency, and so on — reproduces in 8 of 10 sections. That is the signal we lead with above.
The finer sub-label beneath each category — the exact flavour of the deficiency — reproduces in only 2 of 10. That is noise, and we do not build claims on it. So we report the section and its dominant category with confidence, and we treat anything more granular as illustrative, not measured.
The verbatim quotes matter for the same reason. Of the panel statements behind this analysis, 70% were verified word-for-word against the source opinion; where a quote could not be matched exactly, it is a paraphrase from our extraction and ranked below the verified ones. Every quote in this piece is one of the verified set.
The contrast points to a narrower, more actionable conclusion than "submit good data." For each section, there is a specific question that most often decides whether the clock stops:
- Toxicological — is the battery complete, and run on material that matches the product going to market? - Compositional — does your batch data show real, in-specification variation across independently produced lots? - Specifications — can every specified parameter be tied back to a batch you actually tested? - Proposed uses — does your exposure estimate survive the conditions of use you are requesting, not a narrower set? - Production — can the panel confirm your studied material is your commercial material — and are you prepared to answer follow-up questions rather than let them lapse?
And if you are in an emerging category — CBD, cultured cells, insects, precision fermentation — plan for the clock to stop. No one in that space has yet made it through in a single pass.
The outcome of each opinion — favourable, favourable with conditions, inconclusive, or unfavourable — and each clock stop are EFSA's own published records. The split into a clean-pass pole (favourable, zero clock stops) and a clock-stop pole, the per-section counts, and the deficiency categories are Borgh's structured analysis of 239 opinions, not an official EFSA classification. Two caveats carry over from our earlier work: the dataset spans both the current and previous novel food regimes, and only *published* opinions are analysed, so it is weighted toward applications that survived to a conclusion. The direction of each contrast is robust to a split-half test; treat the counts as well-grounded, not exact.
- Own structured analysis of 239 published EFSA opinions (2003–2026). Clean-pass pole: 42 opinions (favourable, zero clock stops). Clock-stop pole: 143 opinions. Discriminator stability tested by seeded split-half (top deficiency category per section reproduced in 8 of 10 sections). - Verbatim panel statements quoted from EFSA opinions ON-50, ON-6196, ON-5002, and ON-3954, each verified word-for-word against the source opinion. - Companion analysis: *What EFSA really checks* — the implied bar measured from the same corpus.
Paste a section and Borgh will flag the gaps that matter — with references to the relevant EFSA guidance.
Try a section →