--- title: "Drawing Check: Engineering Drawing Review (Concept) | Veldarium" canonical: https://www.veldarium.com/drawing-check html: https://www.veldarium.com/drawing-check description: "Drawing Check, the engineering drawing review Veldarium intends to build: compare each revision with the last, check it against written rules, mark every finding on the sheet; an engineer decides. A concept: nothing reads a drawing today." last_updated: 2026-10-12 representation: registry-derived Markdown --- # Drawing Check > Drawing Check, the engineering drawing review Veldarium intends to build: compare each revision with the last, check it against written rules, mark every finding on the sheet; an engineer decides. A concept: nothing reads a drawing today. ## Drawing Check (Concept) Plate is cut, material ordered and joints welded from drawings. A stale reference or a wrong parts-list quantity goes straight into the hull. Drawing Check is the engineering drawing review Veldarium intends to build, to catch such errors at the sheet. It would compare each issued revision with the one before it, check it against written rules and mark every finding where it sits. A named checker or engineer would decide each one. It would read drawings; VeldBuild would hold the recorded review as evidence against the revision; VeldMind would reason over that record. None of the three releases a drawing or the work it governs. Boundary: Concept · no Veldarium software reads a drawing today. ## The intended review, illustrated (synthetic) Two revisions, every finding marked on the sheet, and who decides each. The sheet is synthetic drawing SYN-STR-2201; each finding is outlined on both revisions in its zone, and given a status only on the revision under review. - Revision 02, the prior revision, of synthetic drawing SYN-STR-2201: a stiffened deck panel with 4 stiffeners dimensioned 5 × 480 = 2400, its section A-A with an unsized fillet weld, notes, a revision table with one row, none approved, a parts list giving 4 stiffeners and item 3 in AH36, a title block, and a zone grid of 8 numbered columns and 6 lettered rows on its border. - Revision 03, the revision under review, of synthetic drawing SYN-STR-2201: a stiffened deck panel with 5 stiffeners dimensioned 6 × 400 = 2400, the dimension clouded and tagged 3, its section A-A with an unsized fillet weld, notes, a revision table with 2 rows, none approved, a parts list giving 4 stiffeners and item 3 in S355, a title block, and a zone grid of 8 numbered columns and 6 lettered rows on its border. ### What the comparison assumes - **Illustration, not a demonstration.** Both revisions were drawn for this page and every finding placed by hand; no software read either sheet. - **Assumed rules, not a published standard.** C: a revision is clouded, tagged and entered in the revision table. 1: a parts-list quantity equals the count drawn. 2: a departure from note 1 is entered in the revision table. 3: a cited drawing is at its current revision. 4: a fillet weld carries a size, or a note gives one. - **The governing revision is not on the sheet.** Revision 03 is newer, but its approval cell is empty: every finding against it is provisional. - **Only the two sheets are read.** No model, drawing register or welding procedure. Stiffeners are matched in order from the plan’s upper edge; dimensions are in millimeters. ### What the marks say - **Recorded change:** the sheet records it. - **Verified inconsistency:** the sheet contradicts itself. - **Potential issue:** intent needs a person. - **Cannot establish:** the sheet alone cannot answer. - **Revision 02:** the same places, outlined without a status. ### What a comparison of the two revisions would report - **C · Geometry · Recorded change · Zones A1 to C1, D5 to D8, F6.** Rev 02 to rev 03, stiffeners: 4 at 480 to 5 at 400. The stiffeners at 480, 960, 1440 and 1920 move to 400, 800, 1200 and 1600; one is added at 2000. Revision 03 is recorded; its approval is not. Decides: Responsible engineer, then document control. - Requirement or definition: The stiffening arrangement in plan, set by its chain dimension. - Downstream work, by production stage (the sheet names no work package): Fabrication: nest release, cutting. Assembly: fit-up, welding. - Evidence lineage: The revision 02 plan, 4 stiffeners dimensioned 5 × 480 = 2400; the revision 03 plan, 5 dimensioned 6 × 400 = 2400, clouded and tagged 3; revision table row 03, whose approval cell is empty. - Cannot be established from the sheet: Whether anyone approved revision 03. Until someone does, every other finding is read against it provisionally. - Disposition: Undecided. The responsible engineer approves or rejects revision 03, and document control records the approval against the sheet. - **1 · Parts list · Verified inconsistency · Zones F1 to F2, A1 to C5.** Rev 02 to rev 03, item 2 quantity: 4, unchanged. The revision 03 plan draws 5. Decides: Drawing checker, then responsible engineer. - Requirement or definition: Parts list item 2, the stiffener quantity. - Downstream work, by production stage (the sheet names no work package): Fabrication: nest release, cutting. - Evidence lineage: Item 2 quantity 4 in revision 02 and 4 in revision 03; stiffeners counted in each plan view, 4 then 5. - Cannot be established from the sheet: Which of the two is right while revision 03 is undecided. That the sheet contradicts itself is established. - Disposition: The drawing checker accepts or rejects the finding; the responsible engineer approves the proposed markup on a copy, item 2 quantity to 5. The sheet itself is never edited. - **2 · Parts list · Potential issue · Zones F4 to F5, D1 to D4.** Rev 02 to rev 03, item 3 material: AH36 to S355. No cloud, no revision table entry, and note 1 calls for AH36 unless noted. Decides: Responsible engineer. - Requirement or definition: Note 1, the steel grade, against the item 3 material. - Downstream work, by production stage (the sheet names no work package): Material procurement: receiving inspection, heat and grade traceability. - Evidence lineage: Item 3 material AH36 in revision 02 and S355 in revision 03, unclouded; revision row 03 records only the spacing change. - Cannot be established from the sheet: Whether S355 is an intended exception or an error, and whether it belongs to revision 03 at all. Intent is not on the sheet. - Disposition: The responsible engineer states which material was intended. If S355 stands, the drawing owner records it in a revision. - **3 · Note · Cannot establish · Zones D1 to D4.** Rev 02 to rev 03, note 3 reference: SYN-STR-2190 REV 01, unchanged. Unchanged is not current. Decides: Document control. - Requirement or definition: Note 3, the drawing this sheet depends on and its revision. - Downstream work, by production stage (the sheet names no work package): None that can be read from this sheet. - Evidence lineage: Note 3, identical in revisions 02 and 03. - Cannot be established from the sheet: Whether revision 01 of SYN-STR-2190 is current. Answering needs the drawing register, which no sheet carries. - Disposition: Document control confirms the current revision; the responsible engineer updates the reference if it moved. - **4 · Annotation · Verified inconsistency · Zone B8.** Rev 02 to rev 03, section A-A weld size: none, unchanged. No note gives a size either. Decides: Welding engineer, then responsible engineer. - Requirement or definition: Note 2, welding to the governing procedure, and the weld symbol it governs. - Downstream work, by production stage (the sheet names no work package): Assembly: welding. - Evidence lineage: The section A-A weld symbol in revisions 02 and 03; notes 1 to 3, none of which sizes a weld. - Cannot be established from the sheet: The size itself. It comes from the governing welding procedure, which is not on the sheet. - Disposition: The welding engineer confirms the size; the responsible engineer approves the markup on a copy that adds it. ## The review as designed, step by step From two issued revisions to an engineer’s decision. Each step opens to show what it reads and where it stops. ### 1. What it reads Two revisions of one sheet, exactly as the drawing office issued them. A review would start from the revision under review and the revision before it, each exactly as issued. Each file is identified on arrival, and every later statement cites it. Before anything is compared, the review confirms that both files are the same drawing and sheet and that one follows the other. A failed check stops the review: two sheets that may not be the same drawing are never compared as if they were. A scan, a pasted detail or a stamp is reported as not comparable until a reader for it has been measured on labeled sheets. ### 2. What it reports Every change, with both values. Nothing unreadable reads as unchanged. Each element is reported as unchanged, changed (with both values), added, removed or not comparable. Not comparable is never folded into unchanged, because a region nobody can read is exactly where an unrecorded change would hide. Every change is then checked against what the revision says about itself. A change the revision records is a recorded change. A change it does not record goes to a person: finding 2 on the sheet above is that case. ### 3. What it checks Three classes of rule, each stating what it must never conclude. Comparison says what changed. Rules say what is wrong with the sheet as it stands, changed or not. Every rule is written down with the one conclusion it may reach, the conclusion it must never be read as reaching, and when it cannot be evaluated. - **Internal consistency.** Does the sheet agree with itself? - A parts-list quantity equals the count drawn (finding 1) - A fillet weld carries a size, or a note gives one (finding 4) - Reads: The sheet alone. - Never concludes: That the sheet is right. A sheet can agree with itself and still be wrong. - Cannot be evaluated: When a value it needs sits in a region it cannot read. - **Reference integrity.** Is everything the sheet cites at the revision that governs? - A cited drawing is at its current revision (finding 3) - A cited procedure or standard detail exists under the number cited - Reads: The yard’s drawing register, read at review time. - Never concludes: That the cited drawing is correct, or released for this work. Current in the register is not released for production. - Cannot be evaluated: Without the register. Then every reference is cannot establish, as finding 3 shows on a sheet read alone. - **Rules the yard adopts.** Does the sheet follow the drafting and design rules the yard has written down? - Every title block field the yard requires is filled - Plate thicknesses come from the yard’s stock list - Reads: Rules the yard writes, owns and versions, each naming the clause of its own document it encodes. - Never concludes: Conformance with any published standard or classification rule. A rule encodes the yard’s reading of its own document; whether that reading matches a class society’s rule is for the yard and the society to settle. - Cannot be evaluated: When the rule’s owner has not confirmed it for this drawing type. ### 4. Four statuses, and who decides each Why the fourth, cannot establish, has to exist. Every finding carries its rule, its place on both sheets, the values read and what it cannot establish. A finding a reviewer cannot check against the drawing is not raised. A checker with two answers has to turn every missing input into one of them. If it raises a fault, engineers learn to dismiss it; if it stays silent, silence reads as a pass. Either way the cost falls on the faults a sheet read alone cannot catch, and those are the ones that reach the floor. Cannot establish names the missing input and who holds it, so the gap is closed rather than guessed. It is the rule VeldBuild’s readiness already follows: unknown is a third answer, never a pass. - **Recorded change.** The revision records this change. Not a fault: it is the list of what a checker has to check. - Decides: The responsible engineer approves or rejects the revision; document control records the approval. - Closes: When the revision is approved or withdrawn. Recording a change is not approving it. - **Verified inconsistency.** The sheet contradicts itself between two things read directly from it: a count and a quantity, a dimension chain and its total. - Decides: The drawing checker accepts or rejects the finding. The responsible engineer, a different person, approves the correction. - Closes: When an issued revision makes the rule pass. A markup that passes is a proposal, not a closure. - **Potential issue.** Something a rule can see and cannot judge: a change nobody recorded, a departure from a general note. Whether it is an error depends on intent, which is not on the sheet. - Decides: The responsible engineer states the intent. - Closes: When the intent is recorded. If the departure stands, the drawing owner records it in a revision. - **Cannot establish.** The rule needs an input the review does not have: the drawing register, a welding procedure, a region it could not read. - Decides: Whoever holds the missing input supplies it (document control for the register, the welding engineer for a procedure), and the rule runs again. - Closes: When the rule has been evaluated, or the responsible engineer records with a reason that it does not apply. Never by being dismissed. ### 5. After a decision Every decision on the record; every correction on a copy. Each decision is recorded against its finding: accepted or rejected, the reason, the person, their role and the time. A rejection stays on the record, and the same question is not asked again until the element it was about changes. Where an accepted finding’s evidence determines the correction (finding 1: item 2 quantity to 5), the software proposes it as a markup on a copy and checks it again. The responsible engineer approves or declines it. The drawing owner then issues a new revision through configuration control, and that revision is reviewed in turn. A finding closes when its rule passes on an issued revision and a person confirms it. ### 6. Where it sits in the build An open review would hold VeldBuild’s configuration gate. A closed one releases nothing. Intended integration, described and not built. VeldBuild’s configuration gate asks which released revision governs a work package. A recorded review would be evidence on that gate. An undecided finding would hold the gate open: no work starts against a drawing still under review. A review whose basis has moved, because a cited drawing was re-issued, would make the gate unknown until the review runs again. A fully decided review would satisfy one condition and nothing more: the revision still has to be released, and the package’s other gates (material, predecessor work, access, quality, resources) still have to hold. A review could keep work from starting. It could never start it. [The configuration gate on one work package, in VeldBuild](https://www.veldarium.com/veldbuild) ## The authority boundary The software would read, compare and propose. Only a named person decides. ### The software may - Read an issued revision without altering it - Compare it with the revision before it, field by field - Check it against written rules, and say when a rule cannot be evaluated - Mark each finding on the place on the sheet that caused it - Propose a correction as a markup on a copy, and check it again before it counts ### Only a named person - Accept or reject a finding, with a reason. Held by: Drawing checker or responsible engineer. - Approve or decline a correction. Held by: Responsible engineer, never the one who accepted the finding. - Close a finding and record the review. Held by: Responsible engineer. ### Never, whoever asks - Edit the issued drawing. A correction is a markup on a copy; a new revision is the drawing owner’s - Treat a region it cannot read as unchanged - Accept, reject, approve or close anything - Release a drawing revision, or the work it governs ## The review, stage by stage, as designed 1. **Read** (Drawing Check, deterministic): Take an issued revision and read each sheet, separating what it can read from what it cannot. The source is never modified. 2. **Compare** (Drawing Check, deterministic): Report every change from the revision before it. A region it cannot read is reported as not comparable, never as unchanged. 3. **Check** (Drawing Check, deterministic): Run written rules, each stating the conclusion it may never be read as reaching. A rule that cannot be evaluated says so. 4. **Find** (Drawing Check, deterministic): Mark each finding on the sheet with its evidence and one status: recorded change, verified inconsistency, potential issue or cannot establish. 5. **Propose** (Drawing Check, deterministic): Where the evidence determines a correction, propose it as a markup on a copy and check it again. 6. **Review** (Engineer, person): A named reviewer accepts or rejects each finding with a reason; only the responsible engineer approves a correction. 7. **Record** (VeldBuild, record): The recorded review becomes evidence against the revision. It releases nothing. ## Where it could fail Six ways a drawing review goes wrong, and what the design does about each. ### Geometry read wrongly - What goes wrong: A line read at the wrong position, or a dimension’s overridden text taken as its value. A misread raises a finding that is wrong, or hides one that is right. - What the design does: Every finding prints the values it read on both sheets beside its place, so a reviewer checks the reading against the drawing at a glance. A dimension whose printed text disagrees with what it measures is itself a finding. A region not read to a measured accuracy is cannot establish, never a finding. - What remains: A confident misread in a file format the reader was not tested on. Each format would be qualified on a fixed set of sheets before its files are accepted. ### A rule that is wrong - What goes wrong: A rule encodes a requirement wrongly, or one that does not apply to this drawing type. It raises findings engineers reject one by one, or misses the case it was written for. - What the design does: Every rule has an owner at the yard and is tested before release on sheets it must flag and sheets it must pass. A rejection that says the rule is wrong goes to its owner. A review records the rules it ran, so changing a rule never quietly changes an old answer. - What remains: A rule wrong in a way no test sheet exercises. Only adjudicated use finds those. ### A stale reference - What goes wrong: The sheet cites a drawing at a revision since superseded. Nothing on the sheet changed, so a comparison reports nothing. - What the design does: References are never judged from the sheet. The review reads the drawing register and records what it read and when. If a cited drawing is re-issued later, the answer becomes cannot establish until the review runs again. The earlier answer survives in the history; its authority does not. - What remains: A register that is itself wrong. The review can be no more current than the register document control keeps. ### A reviewer who rubber-stamps - What goes wrong: A checker decides findings without opening them, because there are many and most are right. The record then says a person decided when nobody did. - What the design does: There is no accept-all. Each decision needs a reason, and the record shows whether the finding’s evidence was opened before it was decided. The person who accepts a finding cannot approve its correction. A yard can place a known fault in a copy under review to test whether its process finds it. - What remains: Diligence belongs to a person. The review can make its absence harder to hide; it cannot supply it. ### A drawing built to mislead - What goes wrong: A file crafted to crash or exploit the reader, text hidden on a sheet to steer a model, a value under a fill, a layer hidden on screen that still prints. - What the design does: Every file is untrusted input, read in isolation. A file that cannot be read yields cannot establish for the whole sheet, never an empty review. Text on a drawing is data, never an instruction. Content the printed sheet does not show is itself a finding, because the floor works from the print. - What remains: A reader fault nobody has found yet. Isolation limits what one can reach; it does not prevent one. ### Drawings that must not leave the yard - What goes wrong: Drawings carry a yard’s designs, and often its customer’s. A review that copies them to the wrong place is a breach, whatever it finds. - What the design does: A deployment holding real drawings would have to meet these requirements before reading one. None is claimed. (Concept: requirements, none of them claimed) - Each customer’s drawings stored apart, under keys only that customer’s deployment uses - No drawing ever reachable by a public link - Access granted per drawing package, at the least a role needs - No drawing content in any model’s context outside that customer’s own processing - Deleting a source deletes everything derived from it - Whether a deployment may hold export-controlled drawings is not established, and is decided per customer before any drawing is read - What remains: Each requirement would have to be shown to the customer’s own security reviewers. This page can only state them. ## What a yard would need before relying on it No precision or recall figure for Drawing Check exists: no Veldarium software has read a real drawing. A yard should rely on it only after measurements on its own drawings, judged by its own engineers. - **Concept (today).** What would exist: This page, the design behind it and the illustrated sheet. Nothing reads a drawing. What would be measured: Nothing. There is nothing to measure. - **Synthetic.** What would exist: A separate system that reads synthetic sheets, against a labeled set in which every rule has seeded faults and clean twins. What would be measured: Each rule fires on every seeded fault and stays silent on every clean twin, and every unreadable region is reported as not comparable. That shows the rules work as written. It says nothing about real drawings. - **Building.** What would exist: Real issued revisions from one drawing office, under agreement, read in shadow: the yard’s checkers finish their normal check before they see any finding. What would be measured: Every finding, and every fault the checkers found, adjudicated by the yard’s engineers. Per rule class: the share of raised findings they uphold (precision), the share of upheld faults the review raised (recall), how often it answered cannot establish, and what each side found that the other missed. - **Working.** What would exist: Use on live revisions at a named yard with a named owner, findings recorded against the revision and the configuration gate reading them. What would be measured: The same measures, run again whenever a rule, a reader or a file format changes. The threshold each rule class must meet is set by the yard’s engineering lead, and no figure is published before it is measured. ## Related - [Why the newest drawing is not the governing one](https://www.veldarium.com/resources/blog/newest-drawing-is-not-the-governing-drawing) - [Which drawing governs when copies disagree](https://www.veldarium.com/use-cases/configuration-authority) ## Next: VeldBuild Established here: how a drawing review would run from two issued revisions to an engineer’s decision, and how it could fail. What would a recorded review change about the build? Whether work can start is VeldBuild’s answer, computed today on synthetic records. [VeldBuild](https://www.veldarium.com/veldbuild) ## Contact To discuss a first VeldBuild deployment, or the company: [chris@veldarium.com](mailto:chris@veldarium.com). Name the release decision your team rebuilds by hand. Do not send confidential information in the first message. ## Sitemap - [Full site map](https://www.veldarium.com/sitemap.md) - [Machine context](https://www.veldarium.com/llms.txt)