The evidence of nearly every project failure exists in writing, in real time, before the failure. The industry files it unread. This is the discipline the other seven run on.
Illustrative engagement telemetry — live engagements display from the actual record.
Research by FMI and Autodesk has estimated that roughly 95% of the data captured on engineering and construction projects goes unused, and put the cost of "bad data" — inaccurate, incomplete, inaccessible, or simply never read — at some $1.85 trillion to the global industry in 2020, with 14% of all construction rework attributable to it. That is the quiet scandal of the sector. The warning almost always exists, in writing, in time. It is filed.
The reason is arithmetic, not negligence. A large project generates tens of thousands of documents — submittals, shop drawings, RFIs, invoices, variations, notices, minutes, letters. A conventional oversight team of six, ten, or fifteen people cannot read them. So the industry sampled: spot-check the invoices, review the critical submittals, summarise the month. In 1995 that was a reasonable response to a human limit. Maintaining it now, when machines can read every document on arrival, is a choice — and the owner pays for it.
What sampling costs is specific. A letter that was actually an instruction, filed as correspondence. A notice period that ran while nobody was counting. A submittal cleared against a superseded revision. None of these are dramatic on the day. They are the compound interest on unread documents, and they are collected years later, with interest, in a claim.
Every document is read against what governs it. Not merely stored and versioned — checked. A submittal against its specification section and the current approved drawing. An invoice against contract terms, order, and verified progress. A letter against the clauses it engages and the clocks it starts. The machine layer does that on arrival; people receive the exceptions with the references already attached.
Everything is connected. A variation is linked to the RFI that preceded it, the drawing revision that caused it, the notice that governs it, and the invoice that will eventually carry it. That web is what allows a question asked two years later to be answered from evidence rather than recollection.
Institutional memory stops being personal. On most projects the reasoning behind old decisions lives in three or four heads per party and leaves when they do. Here it survives staff churn on every side of the table, including ours.
| Class | Checked against | Produces |
|---|---|---|
| Submittals, shop drawings | Specification, approved drawings, brand standard | Deviation flags, drift entries |
| Invoices, applications | Contract, order, verified progress | Three-way match, holds |
| Variations, notices | Clauses, time bars, precedent | Claim-risk scores, deadlines |
| Correspondence, minutes | Instructions, commitments, decisions | Decision log, evidence links |
This is the discipline the other seven stand on. Cost control can only match an invoice that has been read. Claims defense can only produce evidence that was captured contemporaneously. Design governance can only catch a substitution somebody opened. Reading the record in full is not one of eight services — it is the condition that makes the other seven possible.
Each document is checked against the references that govern it on arrival — a submittal against its specification section and the approved drawings, an invoice against contract terms and verified progress, a notice against the clause and the clock it starts. What reaches a person is the exceptions, with the governing references attached. Nobody is reading faster; the searching has been removed.
A CDE stores and controls documents. It does not read them. It will tell you the current revision and who has it; it will not tell you that this submittal contradicts the specification, or that this letter started a 28-day clock. Storage is a filing question. This is a reading question.
Nothing. Institutional memory on a project usually lives in the heads of three or four people on each side, and it walks out with them. When every document is indexed and connected, the reasoning behind a decision made two years ago is recoverable by anyone, including a new team.
No — that is the point of building it continuously. The complete, dated, cross-referenced record exists as a by-product of daily governance rather than as a scramble assembled once a dispute is already live.