Methodology
How this tool reads RASFF data, what it calculates, and where its limits are. Source: European Commission RASFF public data.
The Rapid Alert System for Food and Feed (RASFF) is the European Union's network for notifying risks to human health found in food, feed and food contact materials. National authorities in participating countries raise notifications when an official control, a company's own check, a consumer complaint or a border inspection identifies a problem. The European Commission publishes a subset of these notifications.
The public dataset is a register of reported enforcement events. Each record describes one notification: a product, one or more hazards, a declared country of origin, the notifying country, a classification (alert, border rejection, information, news) and, where published, an official risk decision.
It does not describe how much of a product was traded, how many consignments were tested, or how many units were affected. Records can also be updated after publication, and a single notification may involve more than one hazard.
A notification count is a surveillance signal. It rises when a product is traded in high volume, when controls are targeted at a specific hazard–origin pair, or when a member state tests more intensively. It falls where controls are lighter — not necessarily where the underlying risk is lower.
For that reason, this tool calls the frequency component the Notification Frequency Index (observed signal frequency) and the combined output the Risk Signal Score. Neither is a probability of occurrence or an incidence rate.
A. Notification Frequency Index (0–100). Within the currently filtered peer set, each group's notification count is transformed with log1p(count) and divided by the same transform of the largest count in that set, then scaled to 100. The log transform stops one very large category from flattening everything else. Because it is relative, the index changes whenever filters change.
B. Severity Index (0–100). Calculated only from records whose official risk decision is one of the two verified, explicitly configured labels (serious = 100, not serious = 25), as the mean of those weights. Every other value — including potential risk, potentially serious, no risk, undecided, missing values and unrecognised text — is unconfigured: it is excluded from the Severity Index and counted against configured-score coverage. No weight is inferred or fabricated.
C. Risk Signal Score (0–100) = Frequency Index × Severity Index ÷ 100. Both components are always shown next to the score so a high rank can be traced to frequency, severity, or both.
D. Confidence. A separate indicator combining sample size (full weight at 30 notifications, 60% of the indicator) and configured-score coverage (40%). Confidence is displayed alongside the score and is never multiplied into it, so a small sample never silently deflates a score. Groups below 5 notifications are flagged as small samples.
Coverage and denominators. Configured-score coverage is the number of notifications with a configured weight divided by all notifications in the selection. Overview and emerging serious share divide serious notifications by all published decisions; score-table serious share divides serious notifications by configured decisions only. Published but unconfigured decisions and missing decisions both reduce score coverage, but are distinct source states. A group with no configured decisions has no Severity Index or Risk Signal Score (—), never zero risk.
Emerging signals compare the last 90 days with the preceding 90 days, requiring at least 5 recent notifications and at least 40% growth before a combination is surfaced.
Multiple hazards and origins. A single notification can carry several official hazards and several countries of origin. All of them are kept and used: in hazard, origin and combination views a notification is counted once per distinct hazard and once per distinct country, never more than once for the same value. Totals, serious share and severity always remain per notification, so expanding a record into several associations never inflates them. In the Hazard × Product × Origin view a combination is a notification-level association only: where a notification lists more than one hazard or more than one origin, the source does not state which hazard belongs to which country of origin, and no such link should be inferred from these figures.
| Risk-decision value | Weight | Note |
|---|---|---|
| serious | 100 | Verified official 'serious' risk decision. Highest configured weight. |
| not serious | 25 | Verified official 'not serious' risk decision. Low, non-zero weight. |
| potential risk | unconfigured (unknown) | Not yet configured: excluded from the Severity Index and counted as unknown in known-risk coverage. No weight is inferred. |
| potentially serious | unconfigured (unknown) | Not yet configured: excluded from the Severity Index and counted as unknown in known-risk coverage. No weight is inferred. |
| no risk | unconfigured (unknown) | Not yet configured: excluded from the Severity Index and counted as unknown in known-risk coverage. No weight is inferred. |
| undecided | unconfigured (unknown) | Not yet configured: excluded from the Severity Index and counted as unknown in known-risk coverage. No weight is inferred. |
| missing / not published / any other value | unconfigured (unknown) | Not yet configured: excluded from the Severity Index and counted as unknown in known-risk coverage. No weight is inferred. |
Only serious and not serious are weighted today, because they are the only labels we have verified against the official RASFF dataset. Additional labels will be weighted only after their meaning is confirmed; until then their notifications reduce configured-score coverage instead of receiving an invented weight.
- Trade volume. High-volume origins generate more notifications at the same underlying defect rate.
- Targeted controls. Reinforced checks on specific hazard–origin pairs mechanically raise counts for those pairs.
- Enforcement intensity. Member states differ in sampling plans, laboratory capacity and reporting practice.
- Missing and unconfigured risk decisions. A published decision can still be unconfigured. Both unconfigured published decisions and missing/nonpublished decisions reduce configured-score coverage and confidence; they are not assigned invented weights.
- Multiple hazards per notification. A record can involve several hazards; normalisation choices affect hazard-level counts.
- Country labels. Country of origin is as declared and can be re-attributed after investigation.
This tool supports prioritization and trend exploration; it is not a substitute for a company-specific risk assessment.
Source: European Commission RASFF public data.
This independent Hub reads its own stored official RASFF snapshot. Record dates, ingestion time and source modification dates are distinct. A stored snapshot does not establish unattended refresh, current source contents or freshness parity with another application. Failed reads show an error, never substituted demo figures. Only a genuinely empty store may display explicitly labelled demonstration data.
Independent analytical tool using European Commission RASFF public data. Not affiliated with or endorsed by the European Commission.