How the Novatug Maritime Incident Dataset Is Built

Summary

The Novatug maritime incident dataset is a researched collection of publicly documented maritime casualties. It is designed to make source material easier to compare, with particular attention to tug incidents and the mechanism behind tug capsizes.

The dataset is not a complete global accident registry. Its totals describe the records collected and classified by Novatug at a stated point in time. They must not be presented as worldwide totals, occurrence rates or proof of product performance.

Current dated snapshot

The figures below come from the staging dashboard API accessed on 26 August 2026. The latest record update returned by the API was 24 August 2026. [S1]

The dashboard defines the 88-case umbrella as tug incidents whose type contains gird or capsize. It includes 48 Capsize girding records, 2 Capsize not girding, 23 Capsize unknown mechanism, 13 Near capsize and 2 Tow capsize records. The umbrella figure therefore must not be described as 88 confirmed girding incidents. [S1]

These are dataset counts. A later API response may change them as records are added, removed, corrected or reclassified. The access date and definition should always travel with the figures. [S1]

What the dataset is for

The dataset supports three related tasks:

1. Preserve a traceable route from a summary back to its public source.
2. Compare incident mechanisms and recurring operational conditions.
3. Separate confirmed girding cases from capsizes whose mechanism is different or unclear.

This last distinction matters. Treating every tug capsize as girding would exaggerate the pattern. Treating an unclear mechanism as non-girding would hide uncertainty. The classification therefore uses an explicit unknown category. [S2]

Scope and inclusion

A record represents a publicly documented maritime incident for which enough information is available to identify the event and retain at least one traceable source. The collection includes tug incidents and other casualties relevant to port control, vessel interaction and towage safety.

Selection is research-led rather than statistically complete. Records were assembled from available public material, historical research and targeted searches. Differences between countries, time periods, languages, reporting practices and online availability all affect what can be found.

The presence or absence of an incident in the dataset therefore cannot be used to calculate a global accident rate.

Source hierarchy

Where several sources are available, Novatug gives priority to evidence in this order:

1. Official accident investigations, government publications and regulations
2. Classification society rules, port guidance and standards
3. Peer-reviewed or institutional research
4. Clearly attributed Novatug datasets, simulations and technical material
5. Manufacturer documentation
6. Maritime trade publications
7. General news reports

Lower-ranked sources can help identify an event or fill a factual gap, but they do not override a primary investigation. Material product claims require separate support and are not proven by incident counts. [S3][S4][S5]

What a record can contain

Depending on the available source, a record can include:

• incident date and location
• vessel names and vessel types
• incident type and tug involvement
• a short factual summary
• fatalities, injuries or other reported outcomes
• the assigned mechanism classification
• source links and source notes

Not every source reports every field. A blank field means the information has not been established from the retained material, not that the value is zero.

How tug capsizes are classified

The classification follows a conservative decision tree. [S2]

Was a tug involved?
└─ Yes: did the tug capsize?
   ├─ Yes: were towline or pushing forces the stated mechanism?
   │  ├─ Explicitly yes: Capsize girding
   │  ├─ Explicitly no: Capsize not girding
   │  └─ Not established: Capsize unknown mechanism
   └─ No: classify the applicable incident type

Capsize girding

Used only where the source describes the towline or pushing operation loading the tug transversely and overcoming its ability to remain upright. Relevant evidence can include the towline lead, towing geometry, assisted vessel movement or an investigation’s explicit use of girding or girting. [S4][S5]

Capsize not girding

Used where the source establishes another mechanism, such as flooding, weather or a stability failure unrelated to towline force. It is not assigned merely because a source does not mention girding.

Capsize unknown mechanism

This is the default when the available evidence does not establish the capsize mechanism. Keeping uncertainty visible is preferable to a confident but unsupported classification.

Related but separate classifications

Hydrodynamic interaction is kept separate where suction or pressure effects from a nearby vessel initiate the loss of control. A tow-hook release failure is recorded as an equipment failure and can coexist with a capsize classification. It does not by itself establish the capsize mechanism. [S2]

Manual review and evidence boundaries

Automated tools can help extract dates, names and candidate classifications, but they cannot make a safety-critical classification final. Any machine-assisted classification must be checked against the cited source before it is treated as complete.

The evidence boundary also applies to Novatug. The dataset can show patterns within the collected records. It cannot prove that a particular tug design would have prevented each casualty. Design and performance claims require their own engineering evidence, test conditions and stated limitations.

Corrections and versioning

Published statistics should identify the export date. When a source correction or reclassification changes a figure, the new figure should be published as a new dated snapshot rather than silently replacing the context of the earlier number.

A correction should record:

• the affected incident
• the field or classification changed
• the source supporting the change
• the date of the change

The public page should also identify a correction contact before publication.

Limitations

• The dataset is not exhaustive and has no stable global denominator.
• Public reporting varies substantially by country and period.
• Older incidents often have fewer accessible primary sources.
• Fatality and injury fields reflect what retained sources report.
• Classification depends on the detail available in those sources.
• Dataset patterns show association within the collected records, not universal causation.

How to cite the snapshot

Use the dataset name, Novatug as publisher, the export date and the relevant scoped count. For example:

Novatug maritime incident dashboard, accessed 26 August 2026: 432 records in total, including 175 tug incidents. Of those tug incidents, 88 had a type containing girding or capsizing terminology.

Do not omit the access date or describe the 432 records as all maritime incidents worldwide.

Frequently asked questions

Does the dataset show that 88 tug incidents worldwide were caused by girding?

No. It shows that 88 of the 175 tug records returned on 26 August 2026 had an incident type containing gird or capsize. Only 48 of those were classified exactly as Capsize girding. [S1]

Why not classify every uncertain capsize?

Because an unsupported classification would distort the pattern. Capsize unknown mechanism preserves the boundary of the available evidence.

Are the 36 fatalities a global total?

No. They are the fatalities recorded within the 48 exact Capsize girding records in this dated dashboard snapshot. [S1]

Can the dataset prove that the Carrousel Rave Tug prevents casualties?

No. It can identify recurring mechanisms and inform engineering questions. Prevention claims require separate design, test and operational evidence.

Explore the evidence

• Browse the maritime incident archive
• Explore the maritime analytics dashboard
• Review the tug incident classification rules
• Read how tug girding can lead to capsizing

Sources

• [S1] Novatug, Maritime dashboard incidents API, accessed 26 August 2026; 432 records returned, with latest record update 24 August 2026.
• [S2] Novatug, Tug Incident Classification, classification decision tree and definitions.
• [S3] UK Department for Transport and Maritime and Coastguard Agency, A Guide to Good Practice on Port and Marine Facilities, updated 15 April 2025.
• [S4] UK Marine Accident Investigation Branch, Girting and capsize of tug Biter, 2024, updated 2026.
• [S5] Transportation Safety Board of Canada, Tug George H Ledcor, M18P0230, 2019.