A Human Graded Every Board in Your Walls. A USDA Study Says They Got It Right 48 Percent of the Time.
Stand at the end of a planer mill conveyor and watch lumber grading happen in real time. Boards come at you fast, maybe one every two seconds, each one a different combination of knots, grain patterns, splits, wane, and discoloration. You have to judge size, location, type, and severity of every defect on every face and edge, apply a grading rule that fills a hundred-page manual, and slap the right stamp on the board before the next one arrives. Do it for eight hours, and get it right.
Rado Gazo, a professor at Purdue University who has spent years studying lumber grading automation, described this job in three words: "like driving in a blizzard."
He is not being dramatic. A USDA Forest Service study led by Earl Kline at Virginia Tech scanned 89 red oak boards through an automated grading system, then had a certified National Hardwood Lumber Association inspector grade the same boards, then manually digitized and mapped every defect on every board to establish ground truth. Forty-three of 89 boards were graded correctly by the human. That is 48 percent.
Flip a coin.
An automated scanner running laser profile detectors, color cameras, and an X-ray got 56 of the same 89 boards right, scoring 63 percent, still not perfect but 31 percent more accurate than the human, and the dollar gap was worse than the board count suggests: the human grader overestimated the lumber value of those 89 boards by nearly 20 percent, because most of the errors went in one direction, upgrading low-grade boards into higher-grade piles where they did not belong. Its value estimate fell within 5 percent of the certified figure.
That was 2003, and two decades later, the technology has gotten dramatically better. Human grading has not changed since 1898.
Six Sensors, One Second, 92 Percent
Gazo's more recent work at Purdue used a Microtec Goldeneye 300 Multi-Sensor Quality Scanner, an Italian-made machine that runs six different sensor types simultaneously: color cameras, black-and-white cameras, profile cameras, line lasers, dot-grid lasers, and an X-ray source that images the internal structure of every board as it passes through. His team graded over 1,000 kiln-dried, rough-surface boards across nine commercial hardwood species, from ash to yellow poplar, comparing the scanner's output against NHLA rules.
Results: 92.22 percent on-grade accuracy and 99.50 percent on-value accuracy. NHLA Sales Code requirements sit at 80 percent on-grade and value within 4 percent of invoice. Both thresholds fell easily, across nine species and three grade categories, on boards the machine had never seen before.
What makes multi-sensor scanning fundamentally different from a human grader is not just speed or consistency, though those matter when you are grading thousands of board feet per shift. It is that the machine sees things the grader cannot. X-ray imaging reveals internal defects, knots hidden beneath sound surface wood, decay pockets that do not show on the face, density variations that affect structural performance but leave no visible trace. A dot-grid laser maps three-dimensional surface topology with submillimeter resolution, catching warp, cup, and twist that a grader eyeballing boards under fluorescent lights will miss on some percentage of passes, especially four hours into a shift.
Carl Lévêque, quality superintendent for all planer mills in Domtar's Wood Products business unit, put it plainly: "AI allows us to be more precise and consistent. There's a lot less waste." Domtar now runs AI-based grading across its mills, producing roughly 15 unique lumber products at its Normandin facility alone, each meeting different customer specifications and building code requirements. Scanners catch defects Lévêque says are "undetectable to the human eye," including separations, subsurface discoloration, and insect damage from post-wildfire timber that looks fine on the surface.
A Workforce That Cannot Be Replaced Fast Enough
Sawmill employment in the United States dropped to approximately 85,400 workers in the third quarter of 2025, according to the National Association of Home Builders, citing Bureau of Labor Statistics data. That is the lowest level since the first quarter of 2013, marking ten consecutive quarterly declines. Meanwhile, sawmill production has held essentially flat, meaning fewer people are producing the same volume of lumber, a gap filled almost entirely by automation and technology investment.
Behind that headline number, the math is grimmer. Forisk Consulting tracks more than 2,300 forest industry mills across North America and reported that softwood lumber capacity fell from 74 billion board feet in 2024 to 73 billion in 2025, with a forecast 1.3-billion-board-foot capacity drop in 2026, the steepest annual decline since the trough of the Global Financial Crisis. Canadian mills face a combined 45 percent in duties and tariffs. Southern yellow pine operations across the U.S. South are closing in response to compressed margins. Hardwood production hit a 65-year low of 4.12 billion board feet in 2025, down 67 percent from the 1999 record, according to Hardwood Market Report estimates published by Fastmarkets.
In that environment, the grading position is one of the hardest roles to fill and one of the hardest to do well. It requires sustained visual concentration, deep knowledge of complex grading rules that differ by species and product, and the physical endurance to make accurate snap judgments on a moving line for an entire shift. Sawmills that cannot find qualified graders, or that cannot afford the quality control problems created by undertrained ones, are not adopting automation because it is trendy. They are adopting it because the alternative is shipping misgraded lumber or shutting down the line.
What Misgrading Costs You
A typical 2,500-square-foot house requires roughly 12,000 board feet of dimensional lumber for its framing package: about 8,000 board feet of 2x4 and 4,000 board feet of 2x6, before you count sheathing, joists, or engineered components. At current Home Depot pricing, a 12-foot #2 Premium Grade SPF 2x4 runs $5.74 to $7.72 depending on location, and an MDPI bibliometric review on AI in wood products notes that price differentials between lumber grades frequently exceed 50 percent.
Kline's finding that human graders overestimate lumber value by roughly 20 percent means the misgrading does not distribute randomly. It skews toward inflation: lower-grade boards get stamped as higher-grade, which means they sell for more than they are worth, and the builder or homeowner ends up with structural members that might not perform to the grade printed on them. A stud-grade 2x4 in a header that should be #2 or Select Structural is not a catastrophic failure waiting to happen, because building codes include safety factors, but it is a board that has less allowable bending strength, less stiffness, and more knots than the grade stamp promises, and over a few thousand boards in a house, those accumulated compromises are not nothing.
Reverse errors happen too. USDA confusion matrix data for the line grader shows that boards actually Face-grade in quality got stamped as #2 Common, a downgrade that, for hardwood, cuts market value roughly in half. Mills lose revenue on every one of those boards. A builder who might have used that stock for exposed trim or cabinetry never gets the chance, because the grade stamp says framing material, so framing material it becomes. Both directions of error destroy value, but only one direction threatens structural performance, and that one happens more often.
What Scanners Cannot Fix
Honest accounting of the limitations: the 48 percent accuracy figure comes from NHLA hardwood grading rules, which are appearance-based and notoriously complex, involving cutting-unit calculations, surface-measure rounding, and judgment calls about whether a defect is "sound" or "unsound" that reasonable graders will disagree on. Softwood dimensional lumber, the stuff that frames most houses, uses the National Grading Rule, which is primarily structural and more standardized, and machine stress-rated lumber has been common in softwood for decades. Human graders probably perform somewhat better on softwood structural grades than on the arcane geometry of NHLA hardwood rules, though no equivalent head-to-head study with this sample rigor exists for softwood visual grading.
Microtec's Goldeneye 300 is also not cheap. Pricing is not published, but industrial multi-sensor scanners of this class typically run in the high six figures, which puts them within reach of large-volume mills but far beyond the economics of a small-scale sawyer producing a few thousand board feet a day. None of this is democratized yet. It is concentrated in the mills that process the most volume, which means the lumber at your local big-box store is more likely to have been machine-graded than the custom-milled Douglas fir your timber-frame specialist sourced from a regional sawyer.
And the scanners are trained on the data they have. A 2026 MDPI paper on CT-based internal defect detection noted that "inconsistent annotations, boundary ambiguity, and insufficient spatial context in 2D slice-based analysis" continue to limit fully automated interpretation. Cognex's In-Sight D900 system, which uses deep learning rather than traditional machine vision, can detect subtle saw blade degradation from the surface finish of cut boards, catching quality problems upstream before they produce visible defects, but it cannot grade for grain character, color matching, or the aesthetic judgment calls that high-end woodworking demands. Machines excel at consistent, rule-based grading. They struggle where grading shades into craft.
What This Means If You Are Building
If you are a general contractor framing $2-to-$5-million custom homes, ask your lumber supplier whether their mill uses automated grading. Some will. Many will not. Mills that do are producing a more consistent product, and consistency matters more than perfection when you are hanging drywall over 400 studs that are supposed to be straight and the same dimension. If your supplier sources from multiple mills, the variability compounds, and the framing crew absorbs it in the form of culled boards, shimmed walls, and callbacks.
If you are a homeowner buying framing lumber for a renovation at a retail yard, the grade stamp on the board is the only quality information you get, and now you know that the system producing that stamp has a documented accuracy problem that the industry has known about for over twenty years. Grade stamps are not lies. They are the output of a process that, under production conditions, generates the right answer roughly half the time for hardwood and probably somewhat better for softwood structural grades, but not as much better as anyone in the supply chain would prefer to admit.
Here is the strongest counterargument to worrying about any of this: buildings framed with human-graded lumber have been standing for a century, and the code safety factors are generous enough that misgraded boards do not cause collapses. That is true, and it is also the argument that every industry uses to avoid improving a process that works just well enough, where the failures are invisible, distributed across millions of structures, absorbed into slightly higher deflection, slightly more floor bounce, slightly wider drywall cracks that get patched and painted over and forgotten. Scanners do not make those failures impossible, but they make them measurably rarer. At 92 percent on-grade versus 48, the distance between the old way and the new is not a rounding error. It is a different standard of care, available now, deployed at some mills, absent from most, and invisible to the person writing the check for the framing package.
Sawmills figured out X-ray scanning, six-sensor fusion, and machine-learning defect classification. They figured out how to grade a thousand boards across nine species at 92 percent accuracy with half the workforce they had a decade ago. What they have not figured out is how to retire the 128-year-old system that still grades most of the wood in your walls.