29 Suicides. Eight Victims Misidentified as Aggressors. And a Question the ACT Has Never Asked.
The ACT’s own domestic and family violence death review reveals a system capable of getting the most fundamental decision wrong: who is the victim and who is the aggressor. What is worse is that the Territory still does not measure how often it happens.
Domestic and family violence policy is understandably dominated by homicide.
But the ACT Government’s own fatality data reveals something that should fundamentally change the conversation.
Between 2020 and 2024, the ACT recorded 49 deaths in a domestic and family violence context.
Eleven were homicides.
Thirty-eight were not.
Twenty-nine were suicides.
In other words, suicides outnumbered homicides by more than two and a half to one. Over the longer period from 2000 to 2024, the ACT register recorded 168 fatal incidents: just 31 homicides and 137 non-homicide deaths. A_System_That_Cannot_See_Itself_Technical_Volume_7x10_interior.pdf
Yet the ACT’s domestic and family violence system remains overwhelmingly constructed around homicide and protection from another person.
That would be concerning enough.
The next finding is worse.
Eight dead victims had previously been identified as aggressors
The ACT Domestic and Family Violence Review’s 2025 report examined 13 deceased people whom it assessed, after reconstructing their histories, as victims of domestic and family violence.
Eight of those 13 had previously been misidentified by police or another agency as the predominant aggressor.
Two had been misidentified within three months of their deaths.
The consequences documented by the review were not trivial bureaucratic errors. They included criminal charges, loss of employment, removal of children and incarceration. A_System_That_Cannot_See_Itself_Technical_Volume_7x10_interior.pdf
Four of the 13 deceased victims had themselves been respondents to family violence orders in circumstances where the later review found evidence that the applicant was actually the predominant aggressor. A_System_That_Cannot_See_Itself_Technical_Volume_7x10_interior.pdf
These were people whom the system was supposed to protect.
Instead, in those cases, the system at some point put them on the other side of the ledger.
And then they died.
We do not know whether the eight misidentified victims were the eight who died by suicide
This is where precision matters.
Of the 13 deceased victims, eight died by suicide.
Separately, eight of the 13 had previously been misidentified as aggressors.
It is tempting to join those numbers.
The death review does not report whether the eight people who died by suicide were the same eight people who had been misidentified. A_System_That_Cannot_See_Itself_Technical_Volume_7x10_interior.pdf
That missing cross-tabulation may be one of the most important omissions in the entire ACT domestic and family violence dataset.
Because the government apparently holds both pieces of information.
It simply has not published them together.
Then there are another 21 suicides
The review separately examined 25 deceased people whom it classified as having perpetrated domestic or family violence.
Twenty-four were men.
Twenty-one of the 25 died by suicide.
Twenty had been named as respondents to a family violence order at some point. Nine were respondents to a current order when they died. A_System_That_Cannot_See_Itself_Technical_Volume_7x10_interior.pdf
Those figures require care.
Calling somebody a “perpetrator” sounds like a finding of guilt. In this dataset, it isn’t.
The classification was made retrospectively by the death review from the records available to it. Twenty of the 25 had been respondents to civil family violence orders. Only five were recorded as having contravened an order, and the review does not report how many had ever been convicted of a domestic or family violence offence.
As the technical analysis concludes, the safe statement is stark enough:
Twenty-one people took their own lives after being accused of using family violence. For most of them, no court ever determined whether the accusation was true. A_System_That_Cannot_See_Itself_Technical_Volume_7x10_interior.pdf
Some undoubtedly used violence. The review itself records substantial evidence of violence within the cohort.
The point is not to retrospectively declare them innocent.
The point is that nobody checked whether the classification system itself was sometimes wrong.
That is particularly difficult to justify when the same government review had already demonstrated misidentification in the opposite direction.
The ACT looked for classification error in one direction only
This may be the most extraordinary finding.
The death review asked whether people it ultimately identified as victims had previously been misidentified as aggressors.
It found that this had happened in 8 of 13 cases.
But it did not publish the reciprocal analysis: whether anybody whom it ultimately classified as a perpetrator had previously been wrongly classified.
The technical review is careful not to turn that absence into a gender claim. It expressly states that the ACT has no data demonstrating that men are more frequently misidentified than women, or vice versa. The documented misidentification among the deceased victims was predominantly of women. A_System_That_Cannot_See_Itself_Technical_Volume_7x10_interior.pdf
The criticism is simpler:
The ACT discovered that its classification process can get the fundamental question wrong, but has never measured its error rate.
Nine people died while subject to current family violence orders
There is another unanswered question.
Nine of the 25 deceased people classified as perpetrators were respondents to a current family violence order when they died.
Six of the 25 had police involvement within the week before death. In this dataset, “police involvement” includes responding to a domestic violence incident, serving an interim order, arrest, a welfare check or responding to a mental-health crisis. A_System_That_Cannot_See_Itself_Technical_Volume_7x10_interior.pdf
This does not prove that family violence orders cause suicide.
Indeed, the obvious starting hypothesis is the opposite: orders are made in already-dangerous situations, so we should expect people involved in those situations to have elevated risk.
But good risk management does not stop at the obvious hypothesis.
It tests it.
The ACT has not.
The ACT Suicide Register does not even contain a field recording whether someone was subject to a family violence order when they died. The Coroners Court confirmed that no review, audit, thematic analysis, research project or annual summary had examined correlations between domestic violence proceedings and suicide outcomes. A_System_That_Cannot_See_Itself_Technical_Volume_7x10_interior.pdf
That is not evidence that orders cause suicide.
It is evidence that the Territory has never built the data architecture necessary to find out whether its interventions ever contribute to harm.
That distinction matters enormously.
The system counts one kind of mistake and not the other
A family violence system has to manage two risks simultaneously.
The first is obvious and devastating: fail to protect someone who is genuinely in danger.
The second is less visible: wrongly classify someone, remove them from their home, separate them from their children, suspend licences, damage employment, initiate criminal proceedings or otherwise impose serious consequences on someone who was not the danger.
The first failure is visible. It can produce a homicide, a coronial inquiry, a death review and intense public scrutiny.
The second largely disappears into administrative records.
As A System That Cannot See Itself puts it, the incentives are therefore structurally asymmetric: one error is visible and attributable; the other is effectively uncounted. A_System_That_Cannot_See_Itself_Technical_Volume_7x10_interior.pdf
That is not an accusation of bad faith against police, courts or domestic violence workers.
It is a system-design problem.
And system-design problems can be fixed.
Start by counting
The first reform is almost embarrassingly simple.
Add a field to the ACT Suicide Register recording whether a deceased person was subject to a family violence order.
Cross-tabulate the information the Territory already holds by cause of death, sex, order status and victim/perpetrator classification.
Measure misidentification.
Publish the results.
Then test whether different assessors presented with the same facts actually reach sufficiently consistent conclusions about who is at risk and from whom.
We cannot prevent every error in a system dealing with incomplete evidence, frightened people, conflicting accounts and urgent decisions.
But we can insist that the system measures its errors, learns from them and stops repeating the ones it already knows about.
At present, the ACT cannot demonstrate that it does.
Read the evidence yourself
The evidence, source documents, proposed reforms and rewritten risk framework are set out in A System That Cannot See Itself: The Technical Volume. The shorter public edition presents the same argument in about 15 minutes. The accompanying Risk Handbook for Domestic and Family Violence provides a practical risk-management approach.
All three can be downloaded free from dfvrisk.com.au.
The technical volume is deliberately open to checking and correction: its evidence cut-off is 2 September 2026, and it distinguishes what the evidence establishes from what remains “not proven”. A_System_That_Cannot_See_Itself_Technical_Volume_7x10_interior.pdf
The subject is too important for slogans from either side.
We need the data.
Comments