Two credible casualty totals can differ without either organization having counted carelessly. One may record identified civilian deaths in a particular area; another may estimate all direct deaths across a conflict. Before asking which number is larger, ask what each number measures.
The most useful casualty figure has a clear population, place, period, category and method. Its accompanying explanation should also say whether it is a verified minimum, a reported total or an estimate. Without those boundaries, precision can be misleading: a number printed to the last person may still describe only the cases accessible to its producer.
Start with the people and categories
Killed, injured and missing are separate statuses. Casualties sometimes combines deaths and injuries, while a particular military or statistical system may use a different definition. A headline that replaces casualties with deaths can therefore change the claim substantially.
Missing means a person's whereabouts or fate is unresolved. It should not silently become a death count. On October 15, the Romanian Naval Authority update reported by AGERPRES listed five people rescued and seven missing after the Wine Trader fire, replacing an earlier four-safe, eight-missing account. One person's status changed; the total aboard remained 12. The rescue chronology illustrates why a changing tally is not automatically an inconsistency.
Civilian and combatant categories need their own rules. A recorder may have enough evidence that someone died but insufficient information to establish their status. Retaining an unknown category is more accurate than forcing every case into a politically convenient classification. Later evidence can change that classification without changing the total number of deaths.
A verified minimum is a bounded claim
The OHCHR guidance on casualty recording emphasizes source assessment, corroboration, consistent categories and quality control. Its recommended UN standard asks whether there are reasonable grounds to believe a recorded event occurred as described. That is a stated evidentiary threshold, not a promise of complete access to every death.
Recorders may build individual records, incident records or both. Names, dates, locations and circumstances help connect accounts, but they also create duplication risks. Two hospitals may describe the same person after a transfer. Several outlets may reproduce one authority's list. A name may appear with different spellings. None of those situations supplies an additional victim merely because there is another entry.
Every Casualty Counts' recording standards call for source retention, duplicate controls, corroboration and correction procedures. A transparent method lets readers see what qualifies for inclusion and how disagreements are handled.
Even careful recording can miss inaccessible places, unidentified people or incidents whose evidence has not yet arrived. A verified minimum means at least that many qualifying cases have met the chosen threshold. It is not a claim that no one else died. Conversely, a larger reported total does not become verified simply because the smaller count is incomplete.
Estimates answer a different question
The Uppsala Conflict Data Program's battle-deaths codebook, version 25.1, defines its scope around direct deaths in combat-related events involving warring parties. This can include military personnel and civilians killed in those events. It does not equal every death caused by a war's wider effects.
UCDP publishes low, best and high estimates to represent uncertainty in the available reporting. Those figures are products of its coding method; readers should not assume that they are statistical confidence intervals or that the high estimate is an absolute ceiling on reality. New information can change the series.
A different method may estimate excess mortality by comparing observed deaths with an expected baseline. Such studies can capture consequences that incident reporting misses, including deaths associated with disrupted healthcare, disease or hunger. Their results depend on the population, baseline, survey or registration data and analytical assumptions.
OHCHR's technical guidance on conflict-related deaths distinguishes documented direct deaths, estimated undocumented direct deaths and broader excess-mortality measures. Adding figures from those categories without checking overlap can count the same deaths twice. Comparing them without aligning the period and population can create an apparent contradiction where the studies asked different questions.
Read revisions before judging the trend
A revised total can include deaths that occurred weeks earlier but were verified only now. An increase between two publications is therefore not necessarily the number killed between those publication dates. Event date, reporting date and verification date serve different purposes.
Corrections can also reduce a total: duplicates may be removed, a person reported dead may be found alive, or a case may fall outside the dataset's rules. A revision history is useful evidence of how a system works. Its existence alone proves neither manipulation nor accuracy.
Before comparing two numbers, align six things: population, geography, time period, type of harm, verification threshold and whether the figure is a minimum or estimate. Then examine source independence, access limits and correction practices. As with other battlefield claims, the question is not simply who published the figure. It is what evidence supports the particular conclusion being drawn from it.