14 pieces.
How well does AI grade logs, according to the research?Off-the-shelf image models in a 2026 study: 92.6% on species with look-alike oaks merged, 40 to 55% on value, where guessing the middle band scores 50%.Can harvester data predict productivity? What the studies foundYes, to about R² 0.7 in two studies. One built its model on 1.12% of 144,973 records without saying why. Tree size drives most of it.Does machine learning forecast lumber prices better than the market?The four recent studies we read don't test it in their abstracts. They compare models with models, not with a no-change forecast or futures.What has computer vision achieved in log yards?About 1% diameter and volume error where capture is controlled, on log ends or a sorting line. From ordinary photos, one model missed four logs in ten.Are digital twins in sawmills delivering, according to the research?In one documented case, yes: a 2025 study reports a sawmill twin with commercial success. The published evidence is still one mill deep.How accurate is LiDAR forest inventory in published trials?At stand level, about as accurate as a field cruise when both are checked against mill scale. At tree and diameter-class level, much less so.What does the research say about optimising log truck routes?One large 2026 preprint reports a 35% distance cut against executed plans. It points to pooling contractors' trucks, but does not test that.Why don't forest supply chains share data? What the studies showOne study of harvester data found no clear owner and no accepted way to price it. A consumer-goods study found sharing pays in plan alignment.How do small manufacturers adopt AI, according to the research?Slowly, and mostly at the edges. Where manufacturers take AI into production, a 2025 study finds a short-run dip, deepest in older firms.How accurately does document AI read handwritten forms?Close to a careful person on free text, weaker on handwritten tables, and wrong in plausible ways. What 2025 and 2026 benchmarks found.What works in demand forecasting for wood products?Time series for the short run, structural models and scenarios for the long run, and drivers only if they are easier to project than demand.What does the research say about decision support in forest planning?Strong on calculation, thin on reported proof. A 2025 review of 23 harvest planning tools found none reporting independent validation or open code.What is data integration worth to a manufacturer? What studies measuredNo independent study has measured integration alone. The closest, of 30,000+ plants, found data use paid 1 to 3% only with the right complements.Does AI improve yield in secondary wood processing?Against fixed-length cutting, yes: a 2024 trial lifted yield from 69.2% to 81.5%. Against a good optimiser, untested, and false detections cost wood.