DFV: A System That Cannot See Itself

Something rather unusual has just been published about Australia’s domestic and family violence system.

Not another opinion piece. Not another collection of anecdotes. Not another argument that the system is either fundamentally good or fundamentally broken.

Instead, A System That Cannot See Itself: The Technical Volume attempts something much more ambitious: a 582-page technical examination of the ACT domestic and family violence system through the disciplines of risk management, evidence, measurement, governance and systems design.

The full technical volume is available free from DFVRisk.com.au, alongside a much shorter public edition. It examines the development of the ACT system between 2016 and 2026 and asks a deceptively simple question:

How does the system know whether it is working?

The answer emerging from the research is uncomfortable.

The ACT has spent a decade building legislation, services, specialist structures, coordination mechanisms, funding programs, risk frameworks and review processes. Yet the central finding of the research is that the measurement and feedback mechanisms needed to establish whether those interventions are actually reducing risk — or sometimes creating unintended harm — have not developed to the same degree.

That distinction matters enormously.

Activity is not the same thing as effectiveness. Money allocated is not necessarily money spent. Money spent is not evidence of an outcome. A risk framework existing is not evidence that it reliably assesses risk.

And a system intended to protect people should be able to detect when it gets things wrong.

Four findings that deserve attention

The public edition distills the problem into several striking findings.

First, the review reports that none of 17 assessed requirements of AS ISO 31000:2018 were fully met by the ACT’s July 2022 Family Violence Risk Assessment and Management Framework.

That does not mean that nothing useful exists within the framework. It means something more specific: when assessed as a risk-management framework against those criteria, the review found that it did not fully satisfy any of them.

Second, the ACT’s 2025 domestic and family violence death review examined 13 deceased victims and found that eight had previously been misidentified as the person using violence.

That is an extraordinary figure because identifying who is actually at risk, who is using coercion or violence, and what intervention is appropriate sits near the heart of the entire system.

Third, the review points to the ACT Auditor-General’s 2024 finding that there was no performance monitoring strategy or evaluation framework for the Safer Families Levy.

This is not the same as saying that the money was wasted, and the authors expressly do not make that claim. It raises a different question: how can government demonstrate what was achieved by an intervention if the mechanisms for measuring its performance were never established?

Fourth, the review identifies 2019–20 as the last year for which the ACT Magistrates Court published protection-order application statistics.

Each fact can be debated in terms of its implications. But the underlying documents and data can also be checked. That is one of the strengths of the project: rather than asking readers to accept a political position, it repeatedly invites them to examine the underlying evidence.

The technical volume goes much further

The 582-page technical volume is where the project becomes particularly interesting.

It does not simply catalogue perceived failures. It attempts to reconstruct the system.

Its chapters examine funding, deaths, service capacity, court demand, interim orders, procedural fairness, identification of the person using violence, missing data, accountability and the relationship between family violence orders and risks that the existing data may not capture.

It then moves from diagnosis to engineering.

The volume contains a proposed rewritten ACT Family Violence Risk Assessment and Management Framework, including sections dealing with governance, screening, assessment, classification, children, risk-management planning, reassessment, records, reporting, validation and correction.

It also contains an interim-order pre-listing risk assessment instrument, risk criteria, a source-reliability and claim-credibility register, control-effectiveness tests, an annual investment and results ledger, draft legislative amendments, an implementation tracker and an inter-rater reliability protocol for testing family violence risk instruments.

In other words, this is not simply a 582-page argument that something should be done.

A substantial part of those 582 pages is an attempt to show what “something” could actually look like.

Perhaps the most important issue: can we tell when we are wrong?

Domestic and family violence policy presents an exceptionally difficult risk problem.

Fail to intervene when someone is genuinely dangerous and the consequences can be catastrophic.

Intervene incorrectly, against the wrong person or on an inaccurate understanding of events, and the consequences can also be profound: separation from children, loss of housing, employment consequences, financial damage, reputational harm, psychological distress and years of litigation.

That is precisely why a sophisticated system needs feedback.

It needs to know how often its initial assessments were subsequently shown to be correct or incorrect. It needs consistent decision criteria. It needs mechanisms for reviewing decisions. It needs to measure outcomes rather than merely activity. It needs to identify unintended consequences.

Most importantly, it needs to learn from error.

The technical volume makes the systems argument that without sufficiently defined criteria, thresholds and decision rules, consistency becomes difficult to measure. Without measuring consistency, validation becomes difficult. Without validation, errors become difficult to identify. And without identifying errors, systematic correction becomes almost impossible.

That is the significance of the title: a system that cannot see itself cannot reliably learn from itself.

What the report does not claim is equally important

One of the more impressive aspects of the research is its restraint.

The project explicitly says that it does not conclude that domestic violence funding has been wasted; that the system is biased against men or women; that family violence orders cause deaths; that false allegations are common; or that officials have acted in bad faith.

Where the evidence cannot establish something, the technical volume repeatedly uses the finding “not proven” rather than turning an absence of evidence into evidence for a preferred conclusion.

That distinction is particularly important in family violence policy, where public debate can quickly divide into ideological camps.

It is entirely possible to believe that domestic and family violence is a serious problem requiring strong intervention while simultaneously demanding that those interventions be evidence-based, measurable, fair and capable of detecting their own mistakes.

Indeed, those positions should reinforce one another.

This should be read beyond the ACT

The immediate subject is the Australian Capital Territory, but the questions raised are national.

Every Australian jurisdiction operates some combination of police responses, protection orders, specialist courts or lists, risk-assessment frameworks, support services, information-sharing arrangements and government-funded interventions.

The obvious question is whether those systems can answer the same basic questions.

How accurately do we identify risk?

How consistently do different practitioners assess the same circumstances?

Have the instruments we rely upon actually been validated?

How frequently is the wrong person identified as the predominant aggressor?

What happens after an incorrect assessment?

Do we measure the unintended consequences of intervention as well as the intended benefits?

Can governments connect expenditure with measurable outcomes?

And when the system makes a mistake, does that information travel backwards through the system so that the next decision is better?

Those are not questions against domestic and family violence reform.

They are questions necessary for good domestic and family violence reform.

A substantial contribution to the debate

People will undoubtedly disagree with parts of A System That Cannot See Itself. A work this large, dealing with an issue this contentious, virtually guarantees disagreement.

But that may be precisely why the technical volume is useful.

It places hundreds of pages of evidence, analysis, proposed instruments and draft reforms into the public domain and, unusually, makes the underlying work freely available for scrutiny. The published methodology distinguishes evidence from interpretation, records corrections and identifies where conclusions cannot presently be supported.

That creates something considerably more useful than another round of competing assertions.

It creates propositions that can be tested.

If a finding is wrong, it can be challenged with better evidence. If a proposed risk instrument is inadequate, it can be tested and improved. If government possesses data that answers questions the report says remain unanswered, that data can be published.

And if the central diagnosis is substantially correct, then the implications extend well beyond one territory.

Australia has invested enormous effort, money and institutional capacity in trying to reduce domestic and family violence.

The next stage of reform may need to ask a harder question than simply “What more should we do?”

It may need to ask:

“How do we know whether what we are already doing actually works?”

The public edition and the complete 582-page technical volume of A System That Cannot See Itself can be accessed free through DFVRisk.com.au.