Methodology

Shark Incidents is built around one principle: fresh information and reliable information are not always the same thing.

A new shark-bite report can be useful within hours, but important details may still be uncertain. A historical record can be decades old and still require further research. The database therefore stores the circumstances of an incident separately from the confidence placed in the record.

1. What counts as an incident record

A public record requires enough credible evidence to establish that a shark-human interaction was reported and to associate it with a meaningful date or date range and location.

A keyword match alone is never enough.

The database can include:

• bites or injuries attributed to sharks;

• provoked interactions such as handling a captured shark;

• watercraft interactions where the vessel, rather than a person, was the direct target;

• air/sea-disaster cases when shark involvement is part of the historical record;

• questionable cases when shark involvement is plausible but not established.

A sighting without a human incident is not an incident record.

2. Discovery

Automated discovery systems may monitor public news sources for phrases such as “shark bite,” “shark attack” and related language.

Discovery creates a private candidate. It does not publish a page.

The system must reject obvious false positives such as sports teams named Sharks, film or book references, anniversary stories that do not describe a new event, ordinary sightings and articles that simply discuss shark safety.

3. Duplicate reports

One incident can generate dozens of articles. Those articles are sources for one record, not dozens of separate map pins.

Potential duplicates are compared using:

• date and time;

• location;

• named beach or offshore description;

• activity;

• source-canonical URL;

• entities mentioned in the report;

• semantic similarity.

When duplicate incident records are discovered later, they should be merged rather than kept for SEO.

4. Source hierarchy

Sources are evaluated by their relationship to the event.

Primary authorities

Examples include lifeguard agencies, coast guards, police, emergency services and relevant local/state authorities. A direct public statement from the responsible authority can strongly support the core event.

Scientific and specialist sources

Scientific institutions and specialist shark-incident organizations are important for definitions, investigated historical records and aggregate context. Their copyright and access terms still apply; a trusted source is not automatically a freely republishable database.

Reliable local and national reporting

Professional reporting can establish what an authority said, provide follow-up information and show how a developing account changed.

Several sites repeating the same wire story or press release are not automatically independent corroboration.

Aggregators and social media

These are useful for discovery but do not establish a record by themselves unless the original primary evidence can be identified.

5. Verification status

Verification status answers: How strong is the public evidence supporting this record?

Reported

At least one credible source supports a plausible shark-human incident, but the core event or important details still require confirmation.

Use this status for fast/developing reports and possible bites where shark involvement is not established.

Corroborated

Use when either:

1. two genuinely independent credible sources support the core event; or 2. a primary authority directly confirms the core event.

Do not count copied versions of one article as independent evidence.

Verified

Use only after strong evidence supports the key facts of the record—for example a well-documented primary investigation or a recognized specialist incident source reviewed in context.

“Verified” does not mean every field is known. Species can remain unknown and coordinates can remain approximate.

Disputed

Use when credible evidence conflicts materially with the record or shark involvement becomes doubtful.

Retracted

Use when later evidence establishes that the public record should no longer be represented as a shark incident. Keep an internal audit trail and an appropriate public correction/redirect strategy rather than silently rewriting history.

6. Incident classification

Classification answers a different question: What kind of interaction does the evidence describe?

Unprovoked

The available evidence does not show that the person first attempted to capture, handle, feed or deliberately interfere with the shark.

This is a human-interaction classification, not a claim about the shark’s motivation.

Provoked

The circumstances materially involve deliberate human interaction with the shark, such as catching, landing, handling, feeding or attempting physical contact.

Watercraft

The shark interacts with a boat, kayak or other vessel. A watercraft event is not automatically a human bite.

Air/sea disaster

The record is associated with a shipwreck, aircraft event or comparable disaster in which the exact sequence can be difficult to reconstruct.

Questionable

The available evidence does not establish that a shark caused the reported injury or event.

Other / Unknown

Used when no more precise classification is justified.

7. Outcome

Outcome should be normalized to:

• Non-fatal

• Fatal

• No injury

• Unknown

Do not expose raw database enums such as `non_fatal` to readers.

A fatal outcome should be attributed to shark-related injury only when the evidence supports that relationship. Scavenging or uncertain post-mortem involvement must not be presented as a fatal shark attack.

8. Location precision

The map must never suggest more precision than the sources provide.

Allowed precision levels:

• Exact/coordinate-supported

• Beach/site

• Locality

• Region

• Offshore approximate

• Country only

• Unknown

If a report says “about 30 miles offshore,” a marker may be generalized to an offshore area; it must not be placed on an invented exact coordinate.

9. Species identification

Store both the raw species wording and the normalized taxon when appropriate.

Public confidence levels:

• Confirmed

• Probable

• Reported

• Unknown

Species must never be inferred solely from geography, typical local fauna or the severity of an injury.

Accepted taxonomy should be normalized through an authoritative taxonomic source such as WoRMS.

10. Activity

Retain the original wording internally and normalize where defensible into stable categories such as swimming, surfing, diving, fishing, spearfishing, boating or shark handling.

Activity is not the same as classification. A fishing-related incident can be provoked or unprovoked depending on what occurred.

11. Historical records

Historical data need a stricter disclosure because documentation quality changes over time.

A legacy URL can tell us that an earlier version of Shark Incidents once had a page; it does not prove that every field encoded in that URL is correct.

An old incident page is restored as indexable only after at least one credible source supports the event and core year/location.

Historical URLs that encode obvious spelling mistakes can redirect to a corrected canonical record after the underlying event is matched.

12. Geographic taxonomies

Earlier data sometimes treated raw location phrases as states or countries. The rebuild normalizes geography into:

• country/territory;

• state/province/administrative region;

• island/locality;

• marine/geographic region;

• free-text incident location.

A phrase such as “between England and South Africa” belongs to the incident location, not a state taxonomy.

13. Personal information and privacy

Incident pages should publish only personal details necessary for a useful, source-transparent record.

Names, ages and medical descriptions should not be copied merely because a news article contains them. Public interest, source authority, sensitivity and the age of the person involved should be considered.

Graphic wound photographs are not used.

14. Corrections and versioning

A public record can be updated when stronger evidence appears.

The system should store:

• what field changed;

• previous value;

• new value;

• source supporting the change;

• reviewer;

• timestamp.

The visible “last reviewed” date should change only after a meaningful review or content change, not simply because a cron task ran.

15. What counts can and cannot tell you

Incident totals describe documented events in the current dataset.

They do not contain a reliable denominator for swimmer-hours, surfer-hours, beach visits or other exposure across every location and period. Therefore the site must not convert raw incident counts into a personal probability or “danger score.”

16. Dataset scope

Every aggregate page should state:

• data-through date;

• verification statuses included;

• classifications included;

• whether questionable records are included;

• whether the page covers the whole reviewed dataset or a subset.

That scope makes comparisons reproducible and prevents a headline total from hiding its definition.

---