There is no such job title

We scraped 909 job postings from heavy civil contractors looking for the person who builds the machine control model. That person has no job title. Across 596 distinct titles, not one contains model, modeler, modeller, machine control, grade control, 3D, or digital. Not one. The closest thing in the entire set is a posting titled Sr Take Off Technician.

Read that as narrowly as it deserves. These are 909 postings pulled on one day, 2026-08-07, from one list of firms we assembled ourselves. It is a snapshot of what one set of contractors happened to be hiring for that morning. It is not a labour market study. It cannot tell you how many people do this work, what they earn, or whether the title exists at firms we did not scrape. What it can tell you is what a large sample of heavy civil contractors call the job when they sit down to write the ad.

What we searched for, and what came back

The one posting in 909 that contains the word "modeler" is Kokosing's Lead Design Engineer, and the word appears there as "CAD modelers." That is a design office drawing plans. It is not the person cutting a surface for a dozer.

The gap between the two columns is the finding. The work is discussed in the body and never in the title.

AGTEK sits with the estimators

AGTEK appears 41 times across the 909 postings. Every one of those mentions sits in a posting titled something else, and the titles cluster: Estimator I, Estimator II, Estimator III, Senior Estimator, Chief Estimator.

What the 41 mentions establish is where the takeoff tool is named, not who builds the file. Nothing in 909 job ads can tell you who cuts the surface a dozer reads. What they can tell you is that the desk where dirt gets quantified for a bid is the one desk in these ads that is named as already having the tool open.

Every GPS title is an equipment operator

Four postings out of 909 put GPS in the title, across three distinct titles: GPS DOZER OPERATOR, GPS Excavator Operator, EXCAVATOR OPERATOR W/ GPS. All four are seats in a machine. The GPS in those titles belongs to the excavator, not to the model. No title in this sample names the work of making the file the machine reads. The eleven postings that mention machine control in the body all carry some other title.

Why the title does not exist

Our reading is that no party in the contract owns the model, so no party staffs for it. That is an inference from the documents below, not something 909 job postings can measure.

FHWA’s technical brief "Utilizing 3D Digital Data in Highway Construction" (FHWA-HIF-17-031, April 2017) makes an engineering point rather than a contractual one: "The data is often not sufficient for construction due to a variety of reasons. The most notable is that the original ground basis for the design differs to field conditions." The same brief recommends design practices "that prioritize the 3D model as the source of the contract plans," which cuts toward elevating the model, not subordinating it. TxDOT’s PS&E Preparation Manual posts cross sections and 3D models under a section headed "For Information only," with a mandated disclaimer that the data "is for non-construction purposes, only". Caltrans ranks "supplemental project information" last of six contract parts in Standard Specifications 5-1.02 (2025 Edition) and describes electronic design files in the subsection of that name — though whether a given project’s model carries that status is set by its special provisions, not by the Standard Specifications. WisDOT's Construction Data Packet requires the contractor to build to the plans. SUDAS Standard Specifications section 1040 says electronic support files "are for information only," and that "Should there be a discrepancy between an electronic support file and a contract document, the contract documents shall govern".

Line those up and the pattern mostly runs one way: the model is furnished for information and the plan set carries the contract. It is not universal. PennDOT has defined a delivery mode in which a model "comprises the primary construction contract document," which is the question we read six agency documents to answer. We are describing what these documents say about which governs, the plan sheets or the model, not offering a reading of anyone's contract. An owner will not create a job title for a document it formally disclaims, and a contractor will not create one for a task the specification already hands it whole. So the file-work lands wherever there is already a takeoff open on a screen.

What that person is actually up against

They are digitizing cross-sections off a plan set, and the sheets do not tell them which line is which.

Take one of our five sets, FL-003. Every one of its 111 cross-sections carries between 8 and 15 distinct line styles, counting a style as one setting of the drafter's pen: width, colour, dash pattern. On that set, zero of 111 sections are unambiguous. Our Oregon set (an FHWA Federal Lands project on the Historic Columbia River Highway, not Oregon DOT) runs lighter at 1 to 7 distinct lines per section, and even there only 19 of 86 sections are unambiguous.

The legend does not rescue this. On our corpus, legends do not map line styles to surfaces, and the layer names live in the PDF's layer metadata rather than printed anywhere on the sheet.

So the decision "this dashed hairline is existing ground and that solid one is finished grade" is a judgement call made by a person, silently, once for every line style on every section — between 8 and 15 of them on each of FL-003's 111 sections — with nothing written down about having made it. When a grade busts and the rework starts, that is the decision nobody can go back and check.

The report our own software writes declares cross-sheet agreement NOT CLAIMED: the honest count of scoreable station comparisons is zero, and the most favourable ground-line ruling reaches seven. We cannot yet show you a station disagreeing with itself across sheets. We can only show you the ambiguity that would make it possible, and the fact that nobody wrote the decision down.

Where our own numbers stand

We build software that digitizes cross-sections, so it is fair to ask what it can actually do. The honest answer is that Mathyra has never shipped a surface.

All 26 output sets we have produced in development carry zero points and zero faces, and each one names the same three reasons it would not hand over a surface: frame handedness unconfirmed — meaning we cannot yet prove which way the offsets run — alignment absent, and alignment curvature unknown. Rather than guess, the software declines and says why. One caveat travels with every count in this section: none of those 26 was produced by the engine we run today. They came off two earlier engines, so read them as a record of what those runs produced, not as a spec sheet for current code.

What it does do, across five sets: 434 source pages, 332 stationed cross-sections, 6,481 polylines exported — polylines, not breaklines, because the software will not tell you what any of them represent — plus a counted list of the polylines it excluded and on what grounds: 4,661 short fragments, 64 over the export cap, 994 with no horizontal extent on Oregon.

Not everything is accounted for that cleanly. There is no denominator of sections that exist on those sheets, so 332 is the number of sections exported, not a coverage rate and not a recall figure. The same report also declares denominator completeness NOT CLAIMED: 56 of 104 cross-sheet comparison units are absent across four sets with no recorded reason, and we do not yet know why.

On the one measurement that touches reality, a person froze 60 elevation points read by hand off the sheets before the reader software was written, blind to what the software would say. Fifty-nine were scored. What changed after that first scoring was the measuring tool, not the engine — it had been misreading closed pavement boxes — and the exported data did not change. Measured again after that correction, on the same bytes, the same 59 hand readings score p95 0.126 ft and 58 of 59 within half a foot, on one set. The canonical artifact for this check still reads FAIL, and it sits deliberately outside our v1 gate roster. Two more caveats ride with the number: the reader recorded to 0.1 ft, so roughly 0.05 of that 0.126 is the reader's own rounding, and the same points were never read twice, so repeatability was never tested.

Against a designer's LandXML on one Oregon corridor, our recovered grade line ran median 0.325 ft and p95 0.833 ft over 86 sections. That set also carries a defect we have made visible but not fixed: 100 of its 211 offset rulers are lost, and the candidate fix is deferred in code with a test pinning the deferral. Trying to name which family of lines is a surface went worse: the solid family scored within half a foot on 19 of 80 sections, median RMS 1.503 ft over 80 scored sections on one corridor. Every semantic class abstained. The software would not tell you which line is finished grade, and it was right not to.

We also treat that LandXML comparison as limited on purpose. Cross-section sheets are plotted from the same corridor model the bid-package LandXML is exported from, a workflow dependency the agencies themselves document. We have adopted that as a binding limit on what our numbers can mean: scoring extraction against that model measures how well we recover a plot, not accuracy against the ground. We have not measured accuracy against a surveyed value, and the report we generate says so on its face. What we will and will not assert is sorted by how well it is evidenced, not by how good it sounds.

The point

The absence of the title is not a hiring gap waiting to be filled. It is where the risk got put down. The model that runs your machines is being built somewhere, by somebody whose title says something else, on a tool sold for pricing dirt, from sheets that do not label their own lines, under a contract that says the plans govern and the model is for information only.

If that person works at your firm, they already know all of this. What we are trying to build is a way for the decisions they make to be written down, named, and checked, instead of living in one person's head between bids.