The title block is the metadata layer of a sheet set
In short
A title block carries the sheet number, sheet title, discipline, project and client identity, scale, the revision history and the drawn, checked and approved initials. Sheet number and revision number are reliable enough to index a set on. Scale, dates and initials are not, and a system that treats every field as equal will index a set wrongly and confidently.
Key takeaways
- Sheet number and revision are the two fields dependable enough to key a document system on.
- Scale is close to useless as data: 'as shown' is common and page scaling breaks the ratio anyway.
- Drawn and checked initials are template residue as often as they are a record of who did the work.
- Extraction is a per-consultant, per-sheet-size problem — a 480-sheet set is roughly 6 templates.
- Completeness is asserted against the transmittal's sheet list, never against what happened to arrive.
The title block is the structured index a sheet set carries on every page, and every document system built over drawings is really built over it. Sheet number, sheet title, discipline, project and client identity, scale, the revision history and the drawn, checked and approved initials — those are the fields, and they are not equally trustworthy. About half are consistent enough across consultants to build on.
Being precise about which half matters, because the title block is the join key for everything else. An RFI raised against a sheet, a field markup, a submittal referencing a detail, a comparison between two issues: all of them resolve through a sheet identity read off the block.
Field by field, and how far each one can be trusted
| Field | What it holds | Reliability across consultants | What to do about it |
|---|---|---|---|
| Sheet number | The set's own identifier, such as A-101 | High. Everything else references it | Never renumber. Keep a parallel index instead. |
| Discipline designator | A letter or pair identifying the discipline | High within one consultant, medium across them | Map per consultant, not per project |
| Sheet title | Free text describing the sheet | Medium. Edited between issues, so not a stable key | Index it for search; never join on it |
| Revision number | The issue identifier, numeric or alphabetic | High where a revision block exists; formats vary | Store the raw string as well as the parsed value |
| Revision date | The date of that issue | Medium. Often the plot date, not the issue date | Reconcile against the transmittal |
| Revision description | Free text saying what changed | Low as data, high as a hint | Route to a person when there is no purpose field |
| Scale | A ratio, a graphic bar, or 'as shown' | Low. 'NTS' is common and page scaling breaks the ratio | Measure from the graphic bar or a known dimension |
| Drawn, checked, approved | Initials | Low. Frequently blank or template residue | Do not drive approval workflow from it |
| Project number | The consultant's own job number | Medium, and different per consultant | Never use as the project key |
The extraction is a per-consultant problem, not a per-sheet one
It is also why failures cluster. When a field starts coming back wrong it is rarely one bad sheet; it is one template that changed, and every sheet from that consultant since. Fix the template and re-run rather than correcting sheets one at a time.
The revision block decides everything downstream
Two things make the revision rows harder than they look. Formats differ — numeric, alphabetic, or a letter-and-number pairing tied to the purpose of issue — so parse defensively and keep the original string. And the date in the row is often when the sheet was plotted rather than when it was issued, which matters the moment anyone asks what was current on a given day.
The purpose of issue is the field that decides whether a sheet is buildable, and it is frequently buried in the free-text description rather than held in a box of its own. That is the same failure as status living in two places, which the submittal log says approved and the inbox says otherwise takes apart: when the authoritative value has no home, everyone reads a different copy of it.
Where the manual fallback goes
- Extract per template and score confidence per field, so the system knows which values it is unsure about rather than presenting them all identically.
- Route failures by consultant and sheet size rather than by sheet. One person fixing 40 sheets that share a pattern is a task; 40 unrelated corrections is a shift.
- Correct the template and re-run. A per-sheet correction is lost the moment a new issue arrives.
- Assert completeness against the transmittal's own sheet list before publishing anything. A set that arrived with 118 of 120 sheets looks complete if you only count what you received — the same trap as a document signed in full with half of it missing.
- Keep the raw crop of every block beside the parsed values, so a disputed field is checked in seconds without reopening the drawing.
What the title block does not tell you
- Which sheets actually changed. A bumped revision says something changed on that sheet, not what — that is a comparison job, and why a revision comparison flags hundreds of meaningless changes explains why the naive version is unusable.
- Whether this is the set being built to. Currency is a distribution property, not a document property, and it is held in keeping one current set across office and site.
- What was actually built. The as-built record lives in field markups, and what one has to carry to be promotable is set out in what a field markup has to carry to become an as-built.
- Which document governs. Precedence is a contract term, covered in order of precedence when the drawing and the spec conflict.
Extraction with a confidence score and a human queue on the exceptions is the shape that survives contact with a live project — the argument we make about operational work generally in AI in logistics operations and build under AI agents and automation. The rest of this silo sits under drawings, specs and document intelligence, inside our construction and contracting work.
Extraction with a confidence score and a human queue on the exceptions is the shape that survives contact with a live project, which is the argument we make about operational work generally in AI in logistics operations and build under AI agents and automation. The rest of this silo sits under drawings, specs and construction document intelligence, inside our construction and contracting work.
Frequently asked questions
Short answers to the follow-ups this page tends to raise.
What information is in a drawing title block?
Sheet number and title, the discipline, project and client identity, the consultant's name and job number, scale, sheet size, the revision history with dates and descriptions, the drawn, checked and approved initials, and space for a seal. Which of those are filled reliably varies by office.
Which title block fields can a document system rely on?
Sheet number and revision number, and after that not much. Sheet title is edited between issues, revision dates are often plot dates, scale is frequently 'as shown', and drawn and checked initials are commonly template residue. Index on the reliable fields, store the rest as attributes, and never join records on free text.
How do you extract title block data from a PDF drawing set?
Identify the consultant and sheet size first, then read fixed regions per template rather than searching each sheet. Vector PDFs carry a text layer that reads directly; scans need optical character recognition and behave far worse. Score confidence per field and route failures by template, because they arrive in batches.
Should we renumber a consultant's sheets to match our own convention?
No. Every RFI, submittal, markup and transmittal already raised refers to the consultant's numbering, and renumbering breaks all of those links at once. Keep a parallel index mapping their sheet numbers to your internal convention, and leave the sheets alone.
- drawings
- document ai
- metadata
- sheet sets
The work behind this page
Builds from our portfolio that this page draws on.
GroundUp
A construction project-management command centre for general contractors that keeps schedule, RFIs, budget and the field log in one place — and maps the critical-path recovery the moment a job slips.
Real EstateAskVault
An AI internal knowledge-search platform that answers employee questions from your own docs — grounded in citations, with knowledge gaps surfaced and deflection tracked.
Productivity AIRead next
- The drawing comparison marks every sheet as changedWhen a comparison returns differences on every sheet, it is usually telling you it never managed to line the two files up. Three sheets you know did not change will settle it in about ten minutes.diagnostic
- When a field markup becomes an as-built, and what it must carryA red line is not a record until it says which issue it was drawn on, when, by whom, why, and under what authority. Most sets are missing at least three of those.definition
- Deferred submittals: the items the register has to hold openA deferred item is an obligation with a later trigger and a reviewer outside your contract. A register that files it under 'not started' has already lost it.definition
- Extraction is clean on the issued PDFs and useless on the scanned onesMeasure effective resolution at the smallest annotation, the skew angle across the sheet, and the contrast between linework and background. Those three explain almost every scan that will not read.diagnostic
- Order of precedence when the drawing and the specification conflictPrecedence is a contract term, not an industry constant. Until somebody reads the clause on this project, no conflict-detection tool can rank what it finds.definition
- The extracted equipment schedule is one column out of alignmentValidate three known rows against the sheet by tag number rather than by position, then run the four post-extraction rules that catch a shift before anyone prices the schedule.diagnostic
Working on something in this space?
Tell us where you are in a sentence or two. We'll tell you honestly whether we're the right team, and what a sensible first slice of the work looks like.
Start the conversation