Arcvue
Reporting & compliance—triggered by the close, not by a person
The month closed at 03:12. By 03:14 six people had the version of it they needed.
Arcvue Reporting builds the month’s statements and packages from the close event itself, and gives each audience the version it needs. This page follows Arcvue reporting one close through nine steps, from 03:12 to six people who did not have to ask.
Specimen Systems LLC · an example 120-person contractor · all figures synthetic
June closes at twelve minutes past three, and at that instant nobody knows.
Nothing is inherited yet. The last entry posts, the period is marked closed, and what exists is a database fact and no more than that. Everything this page describes follows from that one event, and the whole argument is about what happens between it and a person reading a number.
In most firms what happens next is a person. Somebody pulls the numbers, builds the pack, checks it against last quarter, formats it, and distributes it. The people who need the information wait on the person who produces it—and when that person changes role or leaves, the process breaks in a way nobody documented, because it was never written down anywhere except in their habits.
What this page counts. Not speed. How many hands the information passes through before it reaches the person whose decision depends on it. On Arcvue that count is zero, and everything else on this page follows from that.
Synthetic. An invented firm. All six packages carry the same numbers, and nobody assembled any of them.
The trigger is the close itself, which is why nobody is waiting fifteen days after it.
The database fact from stop 01 is the trigger. Not a schedule—a schedule is a guess about when the close will finish, and it is wrong in both directions. Run it early and the pack reports a period that had not closed. Run it late and everyone waited for no reason.
The monthly performance report goes out because June closed—not fifteen days after the books close, which is what it costs when a person has to assemble it. Arcvue compares the close date against the one it last saw, and a close that lands at 03:12 is noticed on the next cycle rather than at a date somebody picked. The board, lender, and division views are already there when their readers log in, each scoped to the reader. A schedule only fires if somebody set it and nobody moved it. The close fires because the books closed.
And those fifteen days were never assembly time. They are coding time—a month of transactions arriving and being classified at the end of it. Arcvue pulls transactions in as they occur and the AI Bookkeeper codes them as they arrive, so the numbers are current between closes as well as at them. The close is what makes the period final. It is not what makes the information available.
The same close reaches six audiences in six shapes, and none of them filters a generic pack.
One event at stop 02 produces six documents. The common alternative is one report sent to everybody, where each reader ignores the four fifths of it that is not theirs.
| Audience | Built for the decision | Does not receive |
|---|---|---|
| Board | Revenue, EBITDA, cash, covenant status, pipeline coverage | Contract-level detail |
| CEO and CFO | Full P&L by division, balance sheet, thirteen-week cash | — |
| COO | Contract performance, forecast status by division, operational metrics | Covenant and capital structure |
| CGO | Pipeline by division, coverage ratios, win and loss history | Balance sheet |
| Division director | Their own contracts and margins | Other divisions. Anything corporate. |
| Investor or lender | A live view of the agreed metrics, updated nightly | Everything not agreed |
Screen Arcvue reporting configuration, six audience views. Read this one row by row. The six views are alternative renderings of one close, not parts of a whole, so there is nothing to total. Basis the layout of each view is configurable without code. What is NOT configurable is the arithmetic underneath it—all six read the same closed period. Synthetic sample.
The most valuable column in that table is the one listing what each audience does not get.
Withholding is the product, not the limitation.
Read Exhibit 3 again and the right-hand column stops looking like omissions. A division director receives their own contracts and margins—nothing from other divisions, nothing corporate. Withholding the rest is what makes it safe to give twelve people financial information at all.
The usual alternative is to send the full pack to a smaller group, because the pack cannot be safely narrowed. So the people closest to the contracts—the ones whose decisions the numbers should be informing—get nothing, and find out at the quarterly review. Role-scoped reporting is what lets you send more information to more people, not less.
And it is the same scoping the rest of Arcvue uses. A lender sees the agreed metrics because scope is a property of the reader, not a separate export somebody prepares. An export prepared by hand is the thing that leaks—not because anyone is careless, but because a person under time pressure copies a tab.
A model reads the whole close and writes what matters, because a table does not tell you what to look at.
Six correctly scoped views arrived at stop 04. A correctly scoped table is still a table: the reader has to find the thing that matters in it, and the reader is a chief executive with four minutes.
So a language model reads the close. Not a template with the numbers dropped into it—it works through the statements, the contract portfolio, the indirect rates and the variances, decides which of them are actually material for that audience, and writes that in plain English. The board summary flags covenant headroom and major revenue movement. The CGO summary calls out coverage gaps and recent bid outcomes. A division director's highlights the contracts running furthest from plan. The insight is surfaced rather than left in the tables for somebody to find, and the tables are still there underneath it for the reader who wants to check.
The summary describes; it does not compute. Every figure it mentions is calculated deterministically and can be traced—the synthesis chooses what is worth saying, it does not decide what is true. A system that let prose produce its numbers could not show you where any of them came from.
The exhibit an investment bank spends weeks producing comes out of the same event.
The packages at stop 05 are recurring. The contract waterfall is the one people pay a bank to build: revenue decomposed year by year into active base, recompetes, and new growth, out to the end of the forecast window.
It is expensive to produce by hand because it needs contract-level backlog, period-of-performance dates, recompete probability, and a forecast, reconciled to actuals. Every one of those is already in Arcvue the moment the month closes—which is why the same decomposition the forecast publishes as funded, recompete, and go-get is available on demand rather than commissioned.
Screen Arcvue contract waterfall, generated on close. Basis these are the same three segments the forecasting page publishes, on the same target, because both are readings of one closed period. Both are computed from the same closed month, so they cannot disagree. Synthetic sample.
The auditor's list is generated from live data, which is a different document from one assembled under deadline.
Stop 06 produced the exhibit for the people financing the business. This one is for the people checking it. The prepared-by-client list is the schedule of everything an auditor wants supporting the numbers, and assembling it by hand is where a clean audit turns into a three-week exercise in finding things.
Generated from live data, it is a different artifact: not a folder somebody filled, but a set of queries against the ledger that just closed. And the difference shows up in the one place it matters—when the auditor samples a transaction and asks for the trail behind it, the answer is the same one screen Arcvue has been keeping all along rather than a reconstruction.
Assisted sign-off on the close, not a gut check. Before the period is marked closed, Arcvue surfaces what a reviewer should look at—unusual movements, accounts that did not behave like their history, items still in an escalation queue. The reviewer still signs. What changes is that they are signing against a list somebody computed rather than against a feeling that it looked about right.
Between closes, the things that move find the person responsible for them.
A month is a long time. Stops 02 to 07 all hang off the close, and the events that matter do not wait for it: a covenant approaching its threshold, a contract running away from forecast, an indirect rate moving outside its normal range, a solicitation scoring high against your capabilities.
Arcvue raises those in the application, scoped to the role that owns them—the same scoping as stop 04. Alerting everyone about everything is the same as alerting nobody—people learn to ignore it within a couple of weeks, and then the one that mattered arrives in a folder nobody opens.
The real test is what happens when the person who used to do all this leaves.
And that is what stop 01 was really about.
Two minutes from close to delivery is a nice number and it is not the point. The point is the count this page has been keeping: zero hands between the event at stop 01 and the person reading it.
Which only becomes visible on the day the analyst who built the board pack every quarter moves on. In the ordinary arrangement that is a small crisis: the pack lived in their habits, their file, their memory of which tab feeds which. Here it is not an event at all, because nobody was in the path to remove. A process that depends on somebody remembering is a process with a single point of failure that never appears on any risk register.
And every number in every one of those six packages ties. Every figure on this page originates somewhere else in Arcvue—the forecast publishes the same three waterfall segments, the program report publishes the same contract margins—because all of them are readings of one closed month. A reporting layer whose figures disagreed with the systems that produced them would be demonstrating the exact failure it claims to remove.