Skip to main content
ARPIActions

Management Actions

Deterministic review prompts backed by the same governed data as the operating console.

Granite Chevrolet · 47 open review prompts

Granite Auto Group is fictional. Operating figures are synthetic.

Every warehouse record in this project is synthetic. Granite Auto Group and its three stores are fictional. No real dealership, customer, employee or lending data exists anywhere in the project.

Dataset v17 · developmentAs of 31 December 2025Real-engine validation pending

Real-engine validation pending. Neither accepted ADR-0008 path has recorded a result, so nothing on this page may be read as evidence that the Power BI semantic model has been validated. This console renders exported SQL figures; rendering a number in HTML proves nothing about a DAX measure.

The evidence behind both statements

How this is built, governed and validated

What an action is, and what it is not

Each prompt below exists because a condition written down in advance holds in this dataset version. It is a reason to look, not a finding, a recommendation of business action, or evidence of a real-world condition.

Nothing here claims a cause. A rule states that a condition holds and shows the exported values that made it hold; why it holds is what the drill-through is for.

The queue is stateless. It is rebuilt from the data every time the export is regenerated, so it holds no history and carries no acknowledgement, assignment, completion or due date. Reloading this page reconstructs the same queue; nothing remembers what was clicked.

No language model, learned model or scoring heuristic takes any part in producing it. Every word of every prompt comes from a rule template and every number from an exported column, so any action can be recomputed by hand from files in the repository.

12 of 30 permanent rules are enabled. The other 18 are retained and switched off, each with the audited reason the project cannot evaluate it honestly — the evidence is absent, it exists only at a different grain, or the condition is one an earlier data-quality gate already prevents.

Every threshold this register owns is a project default for a fictional dealer group. None is an industry benchmark, an OEM standard, a best practice or a compliance requirement.

The review queue

Ruleset 911c6bd4a226
1of 47 open review prompts shownClear all filters

Counts are of the whole queue, so selecting one facet does not renumber the others. Open means present in the data this console is serving, as of 31 December 2025: the queue is rebuilt whenever that data is and holds no workflow state, so nothing here is assigned, acknowledged, completed or overdue.

What the gross change decomposes into

Total gross change, December 2025

In December 2025, total gross increased by $1,389 against the month before. The bridge attributes +$16,463 to unit volume, -$14,068 to front PVR, -$1,006 to back PVR.

  • Volume effect+$16,463
  • Front PVR effect-$14,068
  • Back PVR effect-$1,006
  • Rounding of the exact components$0
  • Total change+$1,389
Read total gross change, december 2025 as a table
Total gross change, December 2025. In December 2025, total gross increased by $1,389 against the month before. The bridge attributes +$16,463 to unit volume, -$14,068 to front PVR, -$1,006 to back PVR.
ComponentRoleAmount
Volume effectAttributed movement+$16,463
Front PVR effectAttributed movement-$14,068
Back PVR effectAttributed movement-$1,006
Rounding of the exact componentsAttributed movement$0
Total changePeriod total+$1,389
How this decomposition works

The bridge is computed in SQL by reporting.vw_gross_change_bridge and carried through the export. It is a SEQUENTIAL decomposition: each effect is measured with the earlier ones already applied, so the order is part of the method and a different order would apportion the same change differently. The bridge attributes; it does not establish cause.

Effects smaller than $500 are grouped into a single remainder rather than listed — display materiality — project default. Grouped, never dropped: the listed effects and the remainder sum to the period change exactly.

What meets a review rule right now

Severity, then domain, then store, then rule

The order is a property of the data rather than of the reader: everyone looking at the same figures sees the same sequence.

  • HighAccountingGSA-001Review role: Controller

    General-ledger control variance against the stock schedule

    $1,250.00 exception amountHigh-severity variance threshold — project default 1000 USD

    Reconcile the general-ledger balance and the stock schedule behind this control account.

    Open the evidence behind this
    Evidence, threshold detail and the rule
    Exception amount
    $1,250.00
    Exception date
    2025-09-30 exception date
    Exception code
    ACC-GL-VARIANCE
    Exception detail
    Used Vehicle Inventory at Granite Chevrolet of Nashua on date key 20250930: GL balance 758234.03 against subledger balance 756984.03, signed variance 1250.00. BOTH SIDES ARE VALID DATA. Positive means the control account carries more than the stock schedule supports; negative means the schedule carries more than the account. This is a reconciliation finding to investigate, not a broken record.

    Both sides of the comparison are valid data. The sign states which side carries more, not which side is wrong, and a variance is a finding to investigate rather than an error.

    Rule ACT-ACC-001 · accounting exception ACC-GL-VARIANCE:gl_reconciliation:20250930-1-2 · exception date basis

How this queue is produced

Where the rules live, and what keeps them honest

Rules live in config/dashboard/action_rules.yaml as data, not code, and are evaluated once at export time against the datasets this console already publishes. The queue is then a governed artifact with its own hash, exactly like a dataset.

The rule file is an input to the published data. Changing a threshold changes the queue even though no business fact moved, so the export check re-derives the queue from the current rule file and fails if the committed one differs. Ruleset 911c6bd4a226 produced what is shown here.

Facet counts are counts of the queue on this page. They are presentation figures derived from the rows, not new measures, and the generator refuses a count that disagrees with the rows it counts.

Every drill-through is checked against the console’s own route registry before the queue is written, so a link here cannot point at a retired route or carry a parameter its destination does not read.