One number, four altitudes.

How four documents drift apart

They stop reconciling, and nobody notices until two of them are in the same room.

It happens through ordinary steps. The client report is prepared first and includes an item that was still being confirmed. The internal report is produced two days later without it. The dashboard was refreshed on a third date. Each document is defensible and the three totals differ by a little.

The cost of that is not the discrepancy. It is that the first person to spot it now has a reason to check everything else, and the credibility of all four documents drops together.

Aggregation, not authorship

The principle that prevents it is that the four documents are not written separately. They are aggregations of the same underlying figures at different levels, and the only thing that changes between them is how much is rolled up.

Which means the detail has to sum to the summary exactly, and any exception — something excluded, something provisional — is applied at the source rather than in one document. If an item is provisional, it is provisional in all four.

The practical version of this is a single cut, at a single moment, from which all four are produced. The reports differ in level. They never differ in content.

What each altitude actually needs

The mistake in the other direction is to produce the same document four times with different amounts of detail deleted.

The foreman's version needs what to do, not how the project is doing. The construction manager's needs comparison between areas, which is a different cut rather than a shorter one. The director's needs trend and exception: what changed and what is at risk. The client's needs position against obligations, which is a contractual frame the other three don't use.

So four documents, four shapes, one set of numbers. The shape follows the decision each reader has to take

, and only the numbers are shared.

The fifth document

There is nearly always one more than the four, and it is the one that causes the discrepancy.

Somebody keeps their own tracker. A discipline lead maintains a spreadsheet because the official report doesn't show what they need. A client representative builds their own summary from the detail. None of these is unreasonable — each exists because the official set didn't answer somebody's question.

They are worth finding rather than banning. A private tracker is a specification: it tells you exactly which view is missing from the four, written by the person who needed it. Adding that view is usually easy, and it removes the source of the next unexplained difference.

Improving the sheet, or working round it

If you are building it, work from one cut-off and one data set. Write down which report is produced from which and at what level, so that nobody adds a fifth from a different source.

Inherited: the four almost certainly come from different places. Rather than rebuilding, check whether they agree for one month. If they do, you have less of a problem than you feared. If they don't, the size of the difference tells you whether this is a timing artefact or two separate systems that have drifted.

System design

One cut, four views. What matters is not the format of each document but that all four are produced from the same extraction at the same moment.

Record	Produced by	Required quality	Verified against	Feeds
The extraction	Project controls	One cut, one moment, one data set	The cut-off	All four reports
Look-ahead view	Construction	Activity and area level	The extraction	What to start
Internal report	Project controls	Discipline level, same numbers	The extraction	Where to move resources
Executive summary	Project controls	Trend and exception, same numbers	The extraction	Whether to intervene
Client report	Project controls	Contract position, same numbers	The extraction	Entitlement, payment

The third column is the same in every row on purpose. If any report is produced from a different extraction or a different date, the four stop reconciling and nobody finds out until two of them meet.

Practical insight

Take last month's client report, internal report and dashboard, and find the same headline figure in all three.

If they match, note the date each was produced from — you may be lucky rather than systematic. If they don't, work out whether the difference is timing or content, because the fix for each is entirely different and only one of them is quick.

Key takeaways

• Four readers need the same figure at four levels: activity, discipline, trend, and contractual position.
• Produced separately, the four drift apart through ordinary timing differences.
• The damage is not the discrepancy. It is that all four lose credibility at once.
• Reports are aggregations of one data set, not four separate documents.
• Exceptions are applied at the source, so a provisional item is provisional everywhere.
• Four shapes, one set of numbers. The shape follows the decision the reader has to take.
• On an inherited setup, test whether they agree before rebuilding anything.

Records born here. The extraction itself, dated · the four views · the exception list applied at source.

What is coming next

Whatever shape those four documents take, the same handful of figures ends up on the front of each. Every one of them describes a period that has already finished.

Next: the indicators themselves — the ones that report the past, the ones that predict the next month, and the one that always lands just under target.