Turn budget, forecast, and prior-period differences into controlled driver bridges, evidence-backed commentary, owned actions, and a reviewable verdict.
A variance bridge can add to zero and still tell the wrong story. The source populations may use different cut-offs. A residual may have been renamed "mix." A stakeholder's explanation may sound plausible but have no transaction evidence behind it.
This skill controls the comparison before writing the commentary. It ties base and actual amounts to named sources, keeps currencies and versions separate, recomputes every variance, and labels each driver as supported, reported, inferred, proposed, or unexplained. It also keeps arithmetic reconciliation separate from causal proof.
Use it for budget-versus-actual, forecast, prior-period, revenue, cost, headcount, and operating-expense analysis. The package includes a decision rubric, a canonical JSON record, and a read-only checker for references, percentages, signed driver bridges, evidence states, actions, gates, and verdict consistency. It does not invent materiality, FX rates, mappings, forecasts, approvals, or accounting treatment.
Input
Analyze Q2 actual revenue against the approved budget by product and region. The budget is $12.4M, actual is $11.7M, and the sales export has controlled volume and realized-price fields. Finance's summary says the whole miss is price, but one region changed product mappings mid-quarter and the mix bridge does not reconcile. Use the supplied materiality policy. Do not change the forecast.
Output
A source and mapping assessment; signed price, volume, supported mix, and explicit unexplained residual drivers; arithmetic tied to the $700K total variance; causal commentary limited to supported evidence; the mapping conflict kept open; forecast impact marked undecided; owned evidence and review gates; and a blocked or ready-for-review verdict.
Provide the entity, period, line items, comparison basis, source reports and control totals, currency and unit, sign convention, approved mappings or rates, materiality and review policy, and known business explanations with their evidence. The skill returns a controlled comparison, signed driver bridge, commentary with evidence states, actions, forecast implications, open gates, and the narrowest supported verdict.
Budget, forecast, prior-period, and actual reports with stable IDs, versions, parameters, periods, currencies, units, populations, control totals, and locators for the amounts in scope.
The approved comparison basis, sign convention, materiality or investigation rules, mappings, allocation and FX policies, completion rule, preparer, reviewer, and decision authority. Missing rules remain open gates.
Available transaction detail, operational measures, contracts, headcount records, rate sources, stakeholder explanations, forecast context, and owners for unresolved evidence requests or actions.
Required only for the optional local JSON checker. It uses the standard library, reads one record, prints findings, and makes no network requests or file changes.
SKILL.md; comparison, decomposition, evidence, action, and closure rubric; canonical variance-analysis record; read-only JSON checker
No reviews yet.
Input
Compare June cloud and contractor spend with the May forecast. Actual is $286K above forecast. Procurement calls $140K one-time, Engineering says usage grew after launch, and the cloud export covers only 24 of 30 days. No investigation threshold was supplied. Build the review for the CFO without inventing the missing days or a materiality rule.
Output
The partial cloud population marked not controlled for a full-month causal claim; actual-minus-forecast arithmetic; supported contract-rate effects separated from reported one-time and inferred usage drivers; percentage and materiality limits; an evidence request for the missing six days; bounded commentary; forecast and action owners; and no claim that the overrun is explained or non-recurring.
Input
Return the canonical JSON record for payroll actual versus budget. Include approved hires, delayed starts, vacancies, salary-rate changes, a disputed allocation, and a zero-budget new team. Preserve employee-population controls, keep the zero-base percentage not meaningful, make all driver sums reconcile, and leave review open because the allocation owner has not supplied approval.
Output
One checker-ready JSON object with controlled source and comparison IDs; recomputed line variances; supported hiring, timing, vacancy, and rate drivers; an explicit inferred or unexplained allocation residual; null percentage for the zero base; qualified commentary; owned actions and gates; and a verdict whose blocker IDs match every open finding and gate.
Trebuh
Updated Jul 13, 2026
Automated scan passed. It found 1 finding; highest severity info.
Scanned Jul 15, 2026.
Automated scanning reduces risk but cannot guarantee a package is safe for every environment.
Creator
Oolvane21
