The Child Protection Test · Discussion 09
What would a system capable of proving its decisions look like?
The final test is not whether the system has policies. It is whether a high-impact decision can later be reconstructed: what was known, where it came from, what contradicted it, what support was offered, what alternatives were tested, who challenged the reasoning, who approved it and what happened when a control failed.
We already require more recording than many people realise
Current Oranga Tamariki policy says case records must be full, accurate, timely and capable of showing social-work analysis, decision-making and the rationale behind decisions.
Its recording guidance says significant decisions should be evidenced and transparent, including how information was gathered and interpreted. Current assessment policy requires reasons for intake decisions to be recorded and approved by a supervisor.
Professional supervision policy also requires casework decisions and rationales to be recorded.
So the proposal in this discussion is not “start keeping records”.
The proposal is to make the critical safeguards structured enough that an auditor, court, whānau member, supervisor or independent reviewer can tell whether they actually occurred.
1. Evidence provenance
Every material proposition should retain its source and status.
Source: who supplied it and when?
Type: direct observation, allegation, professional opinion, document, historical note or inference?
Verification: what was done to confirm it?
Currency: does it still describe the present situation?
Dispute status: has the family or another professional challenged it?
Correction history: has it been amended, qualified or accompanied by a statement of correction?
Without provenance, repetition can make weak information look strong.
2. Disconfirming evidence must have a home
A system designed only to collect evidence supporting the current concern will naturally reinforce its first hypothesis.
The file should therefore contain a dedicated field for material that cuts against the proposed conclusion.
That could include successful engagement, contrary professional evidence, current clean testing, stable housing, an available whānau caregiver, a child’s different account, evidence that an allegation was not substantiated, or proof that a required service was never available.
Discussion 05 showed that OT already requires favourable information to be included in without-notice section 78 affidavits.
The same intellectual discipline should be visible earlier in assessment.
3. Support-before-separation ledger
Where the statutory and practice framework expects support, alternatives or whānau resources to be considered, the record should capture more than “referred to service”.
What specific issue needed to change to reduce risk?
What service, payment, housing intervention, treatment, respite, transport, disability support or whānau option was identified?
Was it actually available in the required location and timeframe?
When and how was it offered to the family?
Did the family refuse, accept, partially engage, or face a practical barrier?
Did the service start? If not, why not?
What changed after the support was delivered?
That prevents “support offered” from becoming a box that conceals a six-month waitlist.
4. Resource failure must be coded separately from family refusal
This should be a non-negotiable data distinction.
A parent declining available treatment is different from a parent being unable to access treatment.
A family failing to attend transport-supported contact is different from transport not being arranged.
A whānau placement failing assessment is different from the assessment not being completed in time.
If those states share the same operational outcome code, the system will produce misleading evidence about family behaviour and service effectiveness.
5. Decision checkpoints for coercive escalation
High-impact steps should trigger mandatory decision checkpoints.
For example:
before a without-notice application;
before rejecting a viable whānau placement;
before changing a plan from reunification toward permanence elsewhere;
before relying on a disputed historical allegation as a current risk factor;
before concluding that a parent “failed to engage” where a service-access barrier is recorded.
The checkpoint should require a reviewer to answer defined questions rather than simply add an electronic approval.
6. Record the challenge, not only the approval
Supervision is valuable when it tests reasoning.
An auditable system should therefore record what alternative hypothesis or contrary evidence was considered, not merely that a supervisor approved the decision.
If the practitioner and supervisor disagreed, that disagreement should not disappear.
It may ultimately be resolved in favour of the more interventionist option. But preserving the challenge improves accountability and future review.
7. Give whānau a structured right of response before the record hardens
Where safety permits, families should be shown the material findings that will drive a high-impact decision and given a defined opportunity to identify inaccuracies, missing context and alternatives.
The system should then record:
what the family disputed;
what evidence they supplied;
what was corrected;
what remained disputed;
and why the decision-maker accepted one account over another.
This does not give a parent veto power over a safety decision.
It gives the system a better evidential record.
8. Make the FGC plan executable
Discussion 06 showed why FGC participation and resources matter.
Every action in a plan should have an owner, deadline, dependency and escalation rule.
If a provider is responsible for treatment, the provider commitment should be as visible as the parent’s obligation to attend.
If the service is unavailable, the plan should automatically return for reassessment rather than silently continuing until the family is labelled non-compliant.
9. Separate the case reviewer from the decision chain
Not every case needs an external investigator.
But defined high-impact or disputed decisions should be capable of independent internal review by someone who was not part of the original chain of approval.
Triggers could include:
repeat without-notice applications;
a serious complaint alleging misleading evidence;
material correction of information after a court filing;
an Ombudsman finding;
a substantiated professional complaint;
or a pattern of promised supports not being delivered.
10. Findings must propagate back into the system
If a complaint, court process or oversight investigation establishes that material information was wrong, the correction should not live only in the complaint file.
The operational case record should identify the finding.
Relevant downstream documents should be reviewed.
People to whom the information was disclosed should be notified where the law requires or permits.
And if the control failure may be systemic, quality assurance should test comparable cases.
11. Build consequence into control failure
A policy that can be breached repeatedly without escalation is guidance, not an effective control.
Consequences should be proportionate and evidence-based.
They might include mandatory retraining, increased supervision, correction of a record, case re-review, management action, commissioning remediation, contract consequences, professional referral or systemic review.
The purpose is not punitive theatre.
It is to make the control real.
A minimum auditable decision record
What statutory or policy threshold was being applied?
What material facts supported it and where did each fact originate?
What evidence cut against the proposed conclusion?
What did the child and family say, dispute or propose?
What practical assistance was required, offered, available and delivered?
What less intrusive options were tested and why were they accepted or rejected?
Who independently tested the reasoning and what issues did they raise?
Who held delegated authority for the final decision?
Which parts of the plan depended on other agencies, providers, funding or capacity?
When did the decision need to be reconsidered and what events triggered early review?
What happened when material information later changed?
What occurred if a mandatory control was breached?
This is the point of the entire series
The Child Protection Test has not argued that children should never be removed.
Some children face serious and immediate danger. The State needs lawful powers capable of acting quickly.
Nor has the series argued that every social worker, iwi provider, court or government agency fails.
The argument is narrower and more demanding.
When the State intervenes in a family at the highest level, it should be able to prove the integrity of the pathway that produced that intervention.
Not just the final allegation.
The evidence.
The contrary evidence.
The support.
The alternatives.
The participation.
The supervision.
The approval.
The review.
And the remedy when one of those safeguards fails.
A child-protection system should not ask families, courts or the public to trust that the safeguards happened. It should be designed to show us.
Primary and official sources
Oranga Tamariki Practice Centre — case recording policy ↗
Oranga Tamariki Practice Centre — recording casework and evidencing decisions ↗
Oranga Tamariki Practice Centre — assessment and recorded decision rationale ↗
Oranga Tamariki Practice Centre — professional supervision and decision records ↗
Oranga Tamariki policy — without-notice applications and favourable information ↗
Office of the Privacy Commissioner — correction of personal information ↗
Current case-recording, assessment and supervision material checked 1 September 2026. The controls described as existing requirements are sourced; the proposed structured audit architecture is a governance proposal developed in this series, not a claim that current law already mandates every field or checkpoint.