Skip to content
PROTOTYPE — MOCK DATAFigures are generated fixtures derived from the registry. Not ACF actuals.
Menu (13)
EVALUATION GALLERY · SECTION 5.3 OF THE SCORING SHEET

Every chart, and the question it answers

Each entry names the question it exists to answer. If you cannot see the question in the picture, the chart has failed and should be deleted — deleting one is a success, not a failure (C5). Charts marked PART-MODELLED contain shape that ACF has not published; each says exactly which part, on the metric page it belongs to.

01 · THRESHOLD BAND

Is the regulatory breach unmistakable? (C2)

02 · TREND VS TARGET

Is the line going where the plan said it would?

03 · VARIANCE BAR

Which months missed, and by how much?

04 · CALENDAR HEAT

Did the batch land, every day?

05 · FUNNEL

Where does the dialling stop converting? (C3)

PART-MODELLEDTS-03TS-05
06 · VINTAGE CURVES

Is this year's book worse than last year's? (C4)

PART-MODELLEDCR-01CR-02
07 · BUCKET FLOW

Which bucket is the book leaking from?

PART-MODELLEDCO-01CO-02
08 · SCATTER RISK

Which dealers trip a two-part rule, not one threshold?

PART-MODELLEDCR-07NS-10
09 · PARETO

How many levers actually move approval rate?

PART-MODELLEDAP-06AP-08
10 · DISTRIBUTION BOX

Does the average hide the cases that breach?

PART-MODELLEDAP-03AP-05
11 · GAUGE BULLET

Is the FY provision plan being spent, or deferred?

Built as a bullet, not a dial. A dial spends a large area encoding one number and cannot show actual, pro-rata target and cap on one axis — and it costs an extra ECharts module we would be shipping to branch sites over SD-WAN.

12 · PERSPECTIVE RADAR

Are we balanced, or blind on an axis? (BSC's own chart)

People & Learning has no scorable metric, so its axis is a gap in the web, not a zero. A zero would say the perspective is failing when the truth is that it is unmeasured.

13 · PBT BRIDGE

Where is the plan breaking — which line moved?

14 · COVERAGE BAR

How much of this scorecard can we actually trust?

15 · HEAT GRID

A 29% company average — but which regions?

PART-MODELLEDNS-01NS-07
16 · BUCKET COMPOSITION

Is the SHAPE of the book changing, not just the roll rate?

PART-MODELLEDCO-01CO-02CO-03

HOW MUCH OF THIS IS MEASURED · Book totals track the disbursement series; the bucket MIX is modelled around ACF's published B1 resolved rate of 54.97%. ACF publishes resolution rates, not a monthly composition — a real bucket ageing extract replaces this wholesale (A-29 applies).

17 · LOSS TRIANGLE

How far has each cohort run — and where is the data we do not have yet?

PART-MODELLEDCR-01CR-02

HOW MUCH OF THIS IS MEASURED · The same numbers the vintage curves plot, in the shape consumer finance reviews them in. Blank cells are NOT YET OBSERVED, never zero — a zero in a loss triangle reads as 'no losses', which is the opposite of the truth.

18 · REGIONAL MOVEMENT

Which regions moved toward the target, and which did not?

PART-MODELLEDNS-01NS-02

HOW MUCH OF THIS IS MEASURED · Replaces a choropleth deliberately. A map would have cost a province GeoJSON to answer this worse: on a map the movement is a colour difference the eye has to estimate, here it is a bar length and the rows sort by it.

19 · DECLARED INCOME SPREAD

Where are the values no underwriter should have accepted?

PART-MODELLEDAP-01AP-02

HOW MUCH OF THIS IS MEASURED · Bin counts are modelled. What is NOT modelled is the tail: ACF's own data-quality review records declared incomes running to 500M VND/month, which is what R-D-12 exists to catch. The plausible ceiling of 100M is ours and needs Credit Risk to set it.

20 · CONNECTOR SEQUENCING

Which feeds land inside the cutover freeze? (D-30)

PART-MODELLED

HOW MUCH OF THIS IS MEASURED · Dates are the 05/06 sprint sequence laid against the cutover window in 00 §10. Two feeds cross the freeze and are drawn as blocked: that overlap is the finding, and it is the reason D-30 was registered as a timeline rather than a table.

21 · DECLINE MIX OVER TIME

Is the reason mix moving, or only its volume?

PART-MODELLEDAP-06FR-01

HOW MUCH OF THIS IS MEASURED · Two series are evidenced (134 score-band rejections, 23% fraud-coded); the other three are a taxonomy we partly invented and are drawn faint and dashed (Q-11, A-28). The shape worth reading is the fraud share doubling — which the static Pareto cannot show at all.

22 · QUEUE AGEING

How long does a submission sit before a checker sees it?

HOW MUCH OF THIS IS MEASURED · Computed from the approval queue. A control nobody has time to perform becomes a rubber stamp, so the four-eyes rule is only as real as this chart's right-hand bars — and a mean would hide them.

Built, and previously only visible on their own pages

These eight were built across F0–F8 and never appeared in this gallery, which under-reported the inventory by a third. Three are charts of the model rather than of a number; four came with Layer 4 and the integration workspace; one is inside every tile on the site.

23 · SPARKLINE

Is this number moving, without leaving the tile?

Direct Disbursement Ratio (DDR)
REGCG-01
67.4%
+37.4pp vs target 30.0%
MANUALPayment GW / T24⏱ T+1

Head of Corporate Governance·IT System Planning

HOW MUCH OF THIS IS MEASURED · The most-rendered chart on the site by a wide margin: one inside every tile, on every pack. No axes, no legend — it answers 'moving up or down' and nothing else, which is all a tile has room to ask.

24 · STRATEGY MAP

What is the causal story, and is it holding? (D-38)

HOW MUCH OF THIS IS MEASURED · Every edge carries its hypothesis status. Seven of nine are UNTESTED and drawn dashed — the map is a set of falsifiable claims, not decoration, and saying which are untested is the point.

25 · CASCADE TREE

How does a department's score roll into a division's?

HOW MUCH OF THIS IS MEASURED · The weights are on the nodes. Without them this is an org chart; with them it is the arithmetic of the roll-up, which is the thing a GM needs to argue with.

26 · LINEAGE GRAPH

Where did this number come from? (D-13 / D-37)

HOW MUCH OF THIS IS MEASURED · The audit artefact: source field → landing → staging → mart → semantic → tile. Blocked nodes are outlined in red with the diagram id that blocks them.

27 · DELIVERY CALENDAR

Did the department send what they said they would?

HOW MUCH OF THIS IS MEASURED · The same ECharts registration as the batch calendar, a different question. Missed windows are grey, not red: red is something that arrived and was refused, grey is absence.

28 · RATING DISTRIBUTION VS QUOTA

Does this team comply with forced distribution, and where is it off?

HOW MUCH OF THIS IS MEASURED · The quota is a line on each bar, not a second series — what matters is only whether a bar crosses its own ceiling. Only an over-quota bar takes status colour, because a rating of 'B' is not a bad thing.

29 · COMMITMENT RELIABILITY

Did the team do what it said, by the date it said?

HOW MUCH OF THIS IS MEASURED · Deliberately not a volume chart. Ticket counts are the obvious thing to plot here and the first thing that gets gamed; padding your commitments makes this line worse, which is the property that makes it usable.

30 · CYCLE-TIME CONTROL CHART

Is the process stable?

HOW MUCH OF THIS IS MEASURED · TEAM level only, and that is a design decision rather than a scope cut: an individual control chart invites comparison between people whose work is not comparable, and there is no version of that chart which is fair.

CHART INVENTORY

30 chart types are built. Fifteen shipped across F0–F6 and are above; seven are new in F9; eight existed since F4–F8 and had never appeared in this gallery, which was under-reporting the inventory by a third. The count is not the point, and reaching it was not the goal. Every one of the seven new shapes answers a question a metric in the registry already asks, and four candidates were refused on the way — bar races, nightingale roses, 3D bars and polar plots would each have moved the number and none of them answers anything. §5.3 of the scoring sheet asks whether a chart answers its metric's question or is decoration, and C5 says deleting one is a success. That still stands at 30.