Backlog burndown open cases by risk tier · projection to week 40

Tier 1 · High Tier 2 · Medium Tier 3 · Low

Recoverable value waterfall A$M · solid = decisioned · hatched = modelled

A$38M FY26 spend drag — do nothing

Pipeline funnel per-stage conversion · median dwell

DECISION OUTCOMES Cleared · code lifted 58% Maintained 31% Escalated to ECDD 11%

This week's considerations

{{ n.text }} Drill to evidence →

Sign-off panel · week 10 gate

Saved views

Tier

Ranked by risk-adjusted priority — statutory clocks always float to top · why this order?

CaseStop-code reasonPropRisk tier · top reason codeCode ageAssembly readinessValue bandState / SLA

{{ r.id }} {{ r.cat }} {{ r.prop }} {{ r.tier }} {{ r.code }} {{ r.age }} {{ r.readyPct }} {{ r.value }} {{ r.state }} ›

Self-exclusion & Crown Play Safe codes are out of re-vetting scope — they appear in the Register Explorer and route to the CPS process, never into this queue. No bulk actions by design — every decision is per case.

✓ {{ decidedText }} · routed to QC sampling · event written to append-only audit trail with model, prompt and checklist versionsView in Audit & Evidence →

← Queue {{ ws.id }} {{ ws.cat }} {{ ws.tier }} {{ ws.age }} · {{ ws.propFull }} · expected value {{ ws.value }} ASSEMBLY READINESS {{ ws.readyPct }}

Assembled case file

Play cadence · pre / post code

2019▼ stop code · Mar 20212026

Avg A$4.2k/mo rated play stopped at code — the revenue consequence and the behaviour baseline for the return decision.

Draft narrative LLM-drafted · analyst-owned Click any sentence to open its cited source · uncited sentences render amber and cannot be committed

{{ n.text }} ^{{{ n.cite }}}

CITED SOURCE {{ citeCard.ref }} {{ citeCard.sys }} ✕ Close

{{ citeCard.body }} retrieved {{ citeCard.ts }} · retrieval hash {{ citeCard.hash }} · model v2.3.1 · prompt v14

1 paragraph edited by S. Chen · 20 Jul 2026 09:02 · tracked in audit trail

Typology checklist SoW/SoF re-verification · guide v3.2 · pinned to case {{ checkCount }}

Waive requires a reason code:

Context timeline sessions · transactions · alerts · contacts · stop-code events — behaviour reads in sequence, not as isolated alerts

SESSIONS CAGE / TXN ALERTS CODE / CONTACT

Session / txn Alert Stop-code event Customer contact The 2021 "chip share" alert cluster resolves as one pattern on one strip — not three separate analyst-hours.

Risk summary — why ranked here

{{ ws.tier }} classifier v2.3 · pinned · calibration →

← lowers risk · raises risk → · plain-language reason codes, never raw model terms

Relationship context

Spouse — patron CR-2021-4418 Declared Mar 2021 case notes. Different surname & address; explains chip sharing flagged as transfers between strangers.

Host-declared relationship feed — not yet available (gap)

Outstanding information

SoW documents (bank stmts, tax return)✓ Received 02 Jul

Occupation verification letterDay 21 · nudge sent

Requested 29 Jun · email + registered mail · reminder automation on · code lifts through this workflow when satisfied

Decision

🔒 Locked — satisfy checklist first

Unlocked — analyst decides

Mandatory reason code:

Nothing auto-clears. The analyst decides; the system assembles, drafts and ranks. Routes to QC per sampling policy.

Pen-state handoff If only outstanding information blocks clearance, emit a handoff artifact to the host workflow — contact permission state, information required, approved framing. Nothing is contacted from this application.

Tier distributionwk-over-wk drift +0.4pt

Model version timeline

v2.3 · 28 Jun 2026 · recalibrated on decisioned trial cases · approved CRO · active — pinned v2.2 · 09 May 2026 · typology feature added (checklist tuning) v2.1 · 21 Mar 2026 · initial trial classifier

Retraining is a visible, logged event — never silent. Tiers pin to the prior version until approval.

Calibrationpredicted vs realized escalation Points on the diagonal = an honest classifier. Mirrors the XGBoost auto-triage pattern already trusted in transaction monitoring.

Aging heatmaptier × age bucket · open cases

Tier 3 dwell falling from the historical 60+ days toward target. Zero Tier 1 cases beyond 60 days — the evidence that faster does not mean riskier.

Consistency panels Team-lead view only · coaching material — no public league table

Analyst {{ a.id }} {{ a.tp }} /day

Overs — unnecessary work on low risk Unders — second risk missed, send-back

Week 10 gate — projected in the decision session Pipeline vs senior analysts · 164 historical cases · selection logic pre-approved by engagement risk partner 91.5% agreement · gate ≥90%

Agreement matrixpipeline recommendation × senior decision · click a disagreement to open the narrative diff

ClearMaintainEscalate

{{ m.n }}{{ m.pct }}

NARRATIVE DIFF — largest disagreement cell Pipeline under-weighted a screening signal in the foreign-PEP-adjacent typology (4 cases). Senior escalated; pipeline recommended maintain. → became checklist amendment CL-3.2-Δ7.

Time per casesenior manual vs assisted · one pair per case · spread shown, not just the mean

10× threshold

0 min150300450 min

Senior manual · median 262 min Assisted · median 24 min 10.9× · gate ≥10×

Documentation qualityrubric radar · blind-scored by compliance · gate: equal or better

Assisted Senior manualCitation integrity 5.0/5

Gate progress — four for four

✓

Coded population — 7,812 patronsfirst cut: withdrawal of licence vs stop code · click a block to open the cohort

Code age distribution2021–22 over-conservative cohort highlighted

Spend-before-code overlay: recoverable value concentrates in the gold cohort — lifetime outcomes where 90-day reviews would have done.

Pen-state cohort412 candidates

Stop-coded solely for outstanding information — requested, chased, never confirmed high risk. Feeds the host-workflow handoff; nothing is contacted from this screen.

Outcome flow — where the population goes as re-vetting proceeds1,264 decisioned to date

Back-book policy mapper NLP over historical free-text exclusion notes, mapped to the current time-bound policy. This application surfaces and evidences; security & surveillance owners decide.

Historical: {{ b.was }} → Current policy: {{ b.now }} Re-paper candidate

Review queue: 318 candidates whose historical outcome exceeds current policy · time-bounding began only in the last couple of years — the back book is large and entirely free text.

Remediation consistency: pipeline configuration checked against open AUSTRAC remediation findings before any case enters · last review 30 Jun 2026 · owner: AML Compliance Officer

Append-only event log every view, retrieval, claim, edit, checklist event, decision and refusal · live

AUSTRAC-review export Full evidence bundle per case or period: narrative with citations, checklist state, decision provenance, model/prompt/rule versions. The only export in the application — and it is logged.

No data-team extract required. The CRO's regulator preparation runs entirely from this screen.

Refusal logattempted actions outside policy — refused on screen, logged

✕ {{ r.what }} {{ r.meta }}

Purpose-limitation attestation

✓ Zero exports outside this console ✓ Zero fin-crime data reads from any marketing surface — confirmed architecturally ✓ Every SoW/SoF view carries a role, a reason and a watermark ✓ Government identifiers masked by default; 100% of unmasks logged with reason

Recovery curvecumulative expected vs realized return · A$M · realized populates in the pilot and is labelled as such

Cost to reviewanalyst-minutes per decisioned case · weekly

manual baseline · 262 min

Priced into the A$1.5–3M review-cost savings claim · assembly effort displaced toward judgment.

Recoverable value ledger one row per cleared patron · de-identified · Finance co-signs every claim

PatronHistorical spendReturn prob.Expected valueStatusFinance co-sign