Capstone: The Close-to-Board Simulation

What is easy to underrate is that each component can be correct on its own and the package can still fail, because the board does not read the pieces in isolation; it reads them together. If the summary says revenue rose on one story and the variance section shows a different one, or if a figure reads $410,000 in one place and $450,000 in another, the reader stops trusting the whole package, even though most of it is right. Getting each piece right is necessary. Getting the pieces to agree with one another, and getting the summary to tie back to them, is a separate discipline layered on top.

The package is a reconciliation problem rather than four writing problems. The one-page summary is a claim about the components underneath it, so every headline figure in the summary should trace down to the component it came from, and any figure that appears in two components should read the same in both. That through-line, from the summary down to each supporting piece and back, is what makes a package defensible in front of a board and an audit committee.

Each component has a purpose and a standard it is held to, and the capstone carries those standards forward unchanged.

• Close and flux (Module 2). Close-ready commentary that ties every figure to the trial balance, screens by materiality as a first pass, and names only drivers the data supports. The material movements this month were a price-led revenue gain, a freight-rate cost spike, an engineering run-rate step, and a one-time legal item.
• Variance and driver tree (Module 3). The revenue beat decomposed into price and volume by line, with at least one alternative hypothesis so the story is not a single confident guess.
• Exception review (Module 7). A small set of rules run against the general-ledger extract, surfacing the entries that warrant a look, each with the rule it tripped and a proportionate disposition, on a reperformable audit trail.
• Technical memo (Module 8). One issue worked to a supported position, with the alternative treatment addressed and ruled out, and the open point flagged rather than hidden.

Each of those modules left you a worked example of its finished output. Those four examples are your destinations again here; the capstone asks you to reach all four in one month and then bind them together.

This integration belongs in an AI course, rather than sitting as four separate exercises, because the method you have practiced generalizes cleanly to a chain: the deterministic math still lives in each component's template or query, the folder is still scoped to what the task needs, the human checkpoint still sits before the deliverable, and the whole thing is still a reusable pattern rather than a one-off. A model is genuinely useful at running four practiced workflows over one dataset and drafting a first-pass summary that pulls their headlines together, which is the slow, assembly-heavy part of a board package.

What the chain does not hand to the model is the judgment that the pieces agree. Reconciling the components to each other, and confirming the summary ties to them, is the cross-check that stays with you. As in every module, the model is a fast preparer here, not an approver. The rest of this module builds the workflow around that split: let the model draft and synthesize, and keep the reconciliation and the sign-off human.

The most useful thing you can hand a model for each component is an example of that component's destination, and you already have all four. Module 2 left you a finished flux commentary, Module 3 a finished variance explanation, Module 7 a finished exception note, and Module 8 a supported memo. Point the model at the right example for each piece and you are asking it to produce more of a known shape rather than to guess the house style four separate times. The path runs from one month of Meridian data to one reconciled board package, with the four prior examples anchoring the shape of each part.

The deterministic split you learned once now applies four times over. The variance bridge holds the price-and-volume decomposition, the flux template holds each account's dollar and percent change, the exception rules run mechanically against the ledger, and the recognition measure on the technical issue is a defined calculation. Each of those is deterministic work with one correct answer, so each lives in a template or query the model narrates on top of, rather than something the model recomputes in prose.

This matters more in a chain than in a single module, because now there are four sets of numbers that all feed one summary. If the model recomputes any of them in prose, it can drift from the template, and a drifted figure does not just weaken one component, it breaks the reconciliation between that component and the summary. Keeping the math in the templates keeps one source of truth per piece, which is what makes the cross-component tie-out possible at all.

One figure carried across two components makes the integration concrete. The Industrial Fittings revenue gain of $410,000 shows up in both the flux commentary and the variance explanation, so it is exactly where consistency is won or lost. The chain carries it in four steps, each with a distinct owner.

• Variance bridge, deterministic. The bridge decomposes the Industrial Fittings movement and shows the full $410,000 is price, with volume on the line flat. There is one correct split, so the bridge holds it and the model does not recompute it.
• Flux template, deterministic. The same line's dollar and percent change come from the flux template, and the $410,000 there is the identical figure, sourced from the same trial balance. Two components, one number.
• Narrate, non-deterministic. The model describes the gain the same way in both places: a price-led increase with flat volume. The words can differ in phrasing, but the driver and the figure do not.
• Reconcile, judgment. You confirm the summary states the $410,000 once, that both the flux and the variance component tie to it, and that neither describes it as a volume or demand gain. This is the cross-check the model does not own.

The capstone folder is deliberately one period: the current and prior trial balances for the flux, the price-volume bridge for the variance, the GL extract for the exception review, and the technical-accounting issue for the memo. Nothing from other months, nothing unrelated. Scoping to a single month is what keeps the four components consistent, because they are all describing the same close; pull in a second month and the pieces start answering different questions and the summary has nothing clean to reconcile to.

Minimum necessary context does double duty here. It keeps each component focused, as it did in every module, and it keeps the package reconcilable, because a summary can only tie to components that all sit on the same period's data. The folder is small on purpose, and the reason is integration as much as quality.

Because a board package runs every period, the prompt sequence that chains the four workflows is worth saving as a reusable skill: the four destination examples, the deterministic-math rule per component, the one-month folder list, and the instruction to draft a summary that ties to the pieces. Saved that way, next month's package starts from a known shape rather than a blank page, and the skill standardizes the assembly the same way the individual module skills standardized each piece.

Pair it with a composite checklist that adds the cross-component checks to the per-component ones you already run: does every summary figure trace to a component, do any two components disagree on the same figure, does an unconfirmed item in the memo carry its provisional flag through to the revenue lines. The per-module skills make each draft repeatable; the composite checklist makes the reconciliation repeatable, which is the part the chain adds.

Download the folder and run the whole chain in whatever AI you use. The folder holds one month of Meridian: the October and September trial balances for the flux, the price-volume bridge for the variance, the GL extract for the exception review, and the technical-accounting issue for the memo. Run each workflow you have practiced, let each component's template hold its math, and then write the one-page board summary that ties the four together. Come back for the cross-component validation before you would call any of it board-ready. The company and every number are fictional.

The heart of capstone validation is reconciliation across the pieces. Take each figure in the one-page summary and trace it down to the component it came from, then check that any figure appearing in two components reads the same in both. A single untied summary figure, or a single number that disagrees across two pieces, is enough to send the whole package back, the same way one untied figure sent a single flux commentary back in Module 2.

Confirm, too, that uncertainty propagates honestly. If the technical memo's position is still waiting on legal, the revenue figures downstream of it carry that provisional flag, and the summary should say so rather than presenting a settled number. Then run each component's own checks underneath: the flux ties to the trial balance, the variance split ties and carries an alternative, the exceptions are real with no clean entry flagged, and the memo cites correctly and rules out its alternative. Work the checklist below against your package before you would ever call it board-ready.

The summary below is what the CFO carries into the room. It leads with the material story, states each shared figure once, and ties every headline down to the component it came from, so a board member who reaches for a component finds the same number.

Revenue. The month's revenue gain is price-led, not demand-led. Industrial Fittings rose $410,000 on the October list price increase with flat volume, and that figure is the same in the flux commentary and the variance bridge. Precision Components added $800,000, but that piece is volume from one large order, and its recurrence is not yet confirmed, so the summary flags it rather than banking it into the run-rate.

Cost and margin. Inbound freight rose $150,000 on higher carrier rates. Because the revenue gain was price with flat volume, that freight step is a margin headwind rather than a cost that scaled with sales, the same reading the flux commentary carries.

Operating expenses. Engineering payroll rose $87,000 on two hires, a run-rate step that will persist. Legal and professional expense rose $150,000 on a one-time settlement, which the summary labels one-time so the board does not extrapolate it. Both figures match the flux component.

Controls. The exception review flagged four ledger items for follow-up: a possible duplicate payment of $18,450.50, a $27,310.22 weekend freight posting, a $50,000.00 round-dollar consulting entry, and a $96,000.00 revenue debit to investigate. None is a confirmed error; each has an owner and a next step, and the count in the summary matches the review's audit trail.

Open technical item. Revenue on the custom build is presented as over-time recognition, which depends on an enforceable right to payment that legal is confirming from the cancellation clause. The related revenue is therefore provisional, and the summary says so, so the board sees the one number in the package that is not yet settled.

The lab is clean by design: one tidy month, seeded anomalies, and four components that were built to fit together. Real packages are messier, and integration strains in three specific places worth naming.

• The late change that breaks reconciliation. A component finalized after the summary was drafted, say a freight accrual revised up from $150,000 after a late invoice, leaves the summary quietly out of tie. The summary is a claim about the components, so any component that moves after it is written has to push a change back up, or the package ships internally inconsistent.
• The two pieces that tell different stories. The variance section can call the revenue gain price-led while a hurried summary calls it demand-led, and both can read fluently on their own. On the same $410,000 line, that is a contradiction a board will surface, and it is invisible unless someone reads the two components against each other rather than each alone.
• The technical position that moves the numbers underneath it. If the custom-build memo flips from over-time to point-in-time recognition after legal reads the cancellation clause, revenue on that contract shifts out of the month, which changes the trial balance the flux ties to and the base the variance is measured against. A single accounting judgment upstream can silently break the reconciled figures three components downstream, which is why the provisional flag has to travel with those numbers.

On top of these, any one component can still produce output that is fluent, specific, and wrong, which is why the per-component checks and the cross-component reconciliation are both in the workflow. Integration does not remove the controller's judgment; it asks that judgment to hold across four pieces at once instead of one.

Rate your own package against the composite rubric below. It spans all eight course learning objectives at once, since the capstone is where they come together: approach, governance, workflow design, evidence, validation, decision quality, framing, and improving the result. Score it against the weakest link in the chain rather than an average, because a board package is only as defensible as its least reconciled piece. Your scores roll up to the workflow maturity dashboard on the course hub.