A drawing is not a description of a shape. It is a list of requirements, and each one costs something to satisfy. The moment two dimensions describe the same feature, the drawing has stopped being a list of requirements and become a set of competing instructions, because the part has to satisfy both, and it cannot always do so if the two disagree at the edges of tolerance.
Dimension types exist to manage that. A dimension that governs a feature, a dimension that merely reports a value, and a dimension that establishes a theoretically exact position are three different things, and they are drawn differently so a reader can tell them apart at a glance. Learning to distinguish them is what stops a drawing from being read as a flat series of numbers.
The distinction matters most in fabrication, where the sequence of operations already introduces variation. Every unnecessary requirement on the sheet is another way for a part to be judged non-conforming without being wrong.
A reference dimension is a value provided for information. It is normally a duplicate of a dimension given elsewhere, or a useful overall figure, and it does not set a requirement. It tells the reader what the result should approximately be without making that number a condition of acceptance.
The reason drawings need them is practical. A fabricator wants to know the overall length without adding up a chain of dimensions. An inspector wants to see the resultant size at a glance. Providing those figures helps, as long as everyone understands that the governing dimensions are elsewhere and that the reference value is allowed to fall where the tolerances put it.
The common failure is a reference dimension treated as a requirement. A shop that makes the overall length exact by adjusting a flange has now made every dimension in the chain wrong in order to satisfy a number that was never binding. The part fits the reference and misses the function.
A basic dimension is the opposite kind of thing. It is theoretically exact, it is drawn in a frame to show that it carries no tolerance of its own, and its purpose is to establish a position or a profile from which a separate control applies. The tolerance does not vanish. It moves into the control that accompanies the basic dimension, where it describes a zone rather than a pair of limits.
This is a genuinely useful arrangement for fabricated parts, because it separates the question of where a feature is meant to be from the question of how far it may drift. It also makes inspection clearer, since the inspector is checking a zone rather than two independent limits that might each be met while the feature is still in the wrong place.
What it demands is that the accompanying control is present and understood. A basic dimension with no control is an untoleranced requirement, which is worse than an over-toleranced one, because nobody knows what to build to.
Chain dimensioning places each dimension from the previous feature, so tolerances accumulate along the chain. Baseline dimensioning places every dimension from a single origin, so each tolerance stands alone. Neither is wrong. They answer different questions and they produce different risks.
A chain is fast to draw and matches how a part is often laid out on a bench, one feature after another. Its weakness is that the last feature in the chain carries the sum of every tolerance before it, which is frequently the feature that has to line up with a hole in another part. A baseline removes that accumulation but can produce a drawing where every dimension starts somewhere far from the feature it describes, which is harder to work from.
The decision should follow the function. Features that must relate to each other should be dimensioned from a common reference. Features that are independent can be chained without consequence. Mixing the two on one part, without thinking about which features need to agree, is how a drawing ends up with an accumulation problem in exactly the place it hurts.
Tolerance stack-up is often discussed as a calculation, but the important part is the map. Any feature that is positioned through a series of other features inherits their variation. On a welded assembly that series usually runs through the fabrication, the fixture, and the weld itself, and each of those contributes something.
Seeing where the stack accumulates changes what a designer does about it. If a hole pattern has to match another part, dimensioning it from a common datum removes the accumulation entirely, at no cost. If a length has to be exact, making the last operation a machining cut rather than a sequence of cuts lets the tolerance sit where it can be held. The arithmetic is secondary to noticing which features are in series.
Two dimensions of the same feature is the most common drafting fault in fabricated work, and it usually arrives by good intention. A dimension is added for clarity, then another is added from a different reference, and nobody removes the first. Now the feature has two requirements, and the tolerances on them are almost never compatible at the extremes.
The result is a part that cannot be fully conforming. It can satisfy one dimension or the other near the edge of tolerance, and which one it satisfies depends on how it was made. Inspectors then disagree, and the disagreement gets recorded as a quality problem rather than a drawing problem.
The fix is to make the intent explicit. If one dimension governs, the other becomes a reference. If both genuinely govern, then the design has a conflict that no amount of tolerance adjustment will resolve, and the drawing is telling you something about the part rather than about the process.
Some callouts on a fabricated part are not measurements at all. A note that a seam must be continuous, that a surface must be free of undercut, or that a component must be fitted after welding is an instruction about process, and it belongs on the drawing in words rather than as a dimension. Trying to express those requirements as numbers produces dimensions that no instrument can verify and that a shop will learn to ignore.
The distinction is worth keeping clean because it protects the dimensions that do matter. A drawing where every callout is a real, checkable requirement is one a shop can work to. A drawing padded with unverifiable numbers trains everyone to treat all of it loosely.
These checks take very little time and remove most of the arguments that follow a first article.
Read that way, a drawing becomes a statement of what the part owes, and each line of it has exactly one owner.
When a fabricator opens a drawing, the first pass is not about the shape. It is about which numbers are requirements and which are information. That reading determines where the effort goes: which features need a fixture, which can be allowed to fall where the process puts them, and which callouts will be checked at goods-in. A shop that cannot make that distinction has to treat every number as binding, which is safe and expensive.
It also determines what a buyer should ask about. A quoted price that seems low on a tightly dimensioned part is usually a sign that the shop read some requirements as advisory. A price that seems high on a loosely dimensioned part usually means the shop treated information as binding. Either way, the conversation is about dimension types rather than about rates.
You can see the result of that discipline in the TIG titanium welding parts work, where the joint geometry is dimensioned from the features that have to agree, and in the titanium fabrication components range, where the same approach keeps assemblies interchangeable. The exhaust pipe and other parts pages show it on simpler components, where a single reference value is often all the drawing needs.
How those drawings are agreed before a project starts is set out under customization, and the material differences that decide how much tolerance a process can realistically hold sit under shop by material. Dimension types are a small piece of drafting vocabulary with a large effect on how a part gets made.
