Anomaly-surfacing dashboard

A highlight is permission to spend executive time, not a verdict. When a series is statistically notable against its locked target, the cell carries four cites in one place: the actual, the target, the vintage, and the method. If any of those four is missing, the cell does not go in the packet. Color without a cite is not quality.

The actual is the observed value for the period under review, taken from the same extract the program already uses for official reporting. The target is the value the appropriation, statute, grant agreement, or published performance plan already set. The vintage is the as-of date on both sides, including whether the actual is preliminary, revised, or final, and whether later adjustments are still expected. The method is the comparison rule that decided the cell was notable: which test, which window, how missing periods are treated, and whether the rule looks at a single period or a run of periods.

Do not invent a significance percent to dress the cell. The named method and its documented parameters are the cite. If the rule cannot be named in the cell or in the footnote the cell points to, the highlight does not ship. Executives need to know which number was compared to which number, as of when, under which rule.

A program performance analyst still briefs. The cell ranks attention. It does not replace the sentence that explains whether the gap is operational, definitional, a lag in reporting, or a mismatch between the unit of the actual and the unit of the target.

Scoring against a moving numerator or an unofficial denominator produces noise that looks like signal. Lock the actuals extract and the target table before the comparison runs. Same period, same unit, same population. If the target is a rate and the actual arrives as a count, convert once, in a documented step, and freeze both sides. Re-running the score after someone refreshes a column is a new vintage. Treat it as one.

The target must already exist in the appropriation language, the performance plan, or the published agreement. Inventing a target the appropriation never set is the fastest way to manufacture a red cell. A stretch goal from a slide deck is not a target. A prior-year actual is not a target unless the statute or plan says it is. A peer median is not a target unless leadership adopted it as one. If no target exists, the series is not eligible for this dashboard. Route unexplained movement to a variance explainer that can describe change without implying a miss against a number nobody authorized.

Platforms already in government reporting stacks, including Microsoft, Palantir, OpenGov, and SAS, can host the extract, the score, and the packet. Treat them as the warehouse and the presentation layer. They are not the authority for what counts as a target. The lock lives in the performance file and the finance extract. The dashboard reads those locks. It does not invent them, and it does not silently substitute a forecast for a target because the target cell was blank.

A demand forecast can inform next year's plan. It is not this year's target unless leadership adopted it. Keep a service demand forecaster upstream of planning, not inside the score for the current cycle.

Consider a quarterly outlay series for a grant program, used here as a structure check, not as a scored result. The actual is the ledger close for that quarter from the official extract. The target is the quarterly share already stated in the award or the apportionment, not a round figure typed into the dashboard because the cell looked empty. Vintage is the close date plus whether later adjustments are still expected. Method is the documented comparison against that locked path, including the named rule for how many consecutive periods matter. There are no dollar figures and no percentages in this example on purpose. The work is the lock, the four cites, and the refusal to score a series that has no authorized target.

Empty is a status, not a defect. Most series in a government portfolio will be ordinary in any given cycle. Leave those cells blank. Filling them with green, with a hyphen, or with a rounded "within range" badge trains executives to scan color instead of cites. It also implies a finding where the method found nothing notable.

The product's job is to prioritize review time. A wall of highlighted cells means the gate failed. Either the method is too sensitive, the targets are poorly specified, vintages are mixed across rows, or someone painted ordinary series so the page would not look sparse. Fix the lock and the method. Do not lower the visual threshold so that more cells light up.

When a series is ordinary, the row can still appear in the packet as context. The highlight column stays empty. That emptiness is what makes the few cited cells readable. If a later revision changes the actual, that is a new vintage. Re-score. Do not edit last cycle's highlight in place.

Treating the highlight as the finding skips the job. A notable residual is a prompt. The finding is the operational story: delayed drawdowns, a definition change in the actual, a timing mismatch between obligation and outlay, a population that shifted after the target was set, or a reporting lag that will close next vintage. The analyst writes that sentence. The dashboard does not.

A red cell with no vintage is incomplete. Without vintage, the executive cannot tell whether they are looking at a preliminary close, a revised figure, or a stale extract sitting next to a current target. Do not ship that cell. Recite the vintage or drop the highlight. A cell that cites actual and target but not method is the same class of defect: the reader cannot reproduce why the row was singled out.

The briefing names the four cites, then the candidate causes, then what would change the cell next cycle. It does not ask the executive to trust the color. If the packet will be public, strip internal scoring language and use a public-facing data summarizer so a highlight rule does not leak as a public verdict. Statutory packets that must assemble narrative and tables together belong in a legislative compliance report assembler, which should consume the same locked actuals and targets rather than a second extract that will drift.

Quality on this page is completeness of the cite, not intensity of the color. A highlighted cell without actual, target, vintage, and method is incomplete. An invented target is a false miss. A highlight treated as a finding is a skipped briefing. Ordinary series painted anyway are noise that burns the next review cycle.

Keep the comparison narrow. Locked actual versus locked target. Method named. Vintage stamped. Empty when ordinary. Analyst on the brief. That is the working product a program performance analyst can defend in a hearing, a budget huddle, or a weekly ops review.