The buildbot-sheathing programme asks whether a robot can nail wall sheathing to code tolerance. Its paper draft for Automation in Construction covered the robot half: tolerances, placement, cycle time and cost per clean sheet. It stopped before the structural half: what a nail the robot misses, and does not catch, costs the wall. Eight milestones of capacity models answered that, and until today none of it was in the paper. This step writes it in. Two methods sections and two results sections were added, and every quoted number is tied to its results file by paper/check_numbers.py, which now checks 190 numbers.
What an escaped miss costs. One missing joint-edge nail costs a mean 2.08% of a 4x8 sheet's racking capacity. Field nails cost 0.85% and plate nails 1.42%. The model reproduces the shear-wall code's ratios for 4, 3 and 2 in edge spacing to within 6%. If every miss from the robot's three-read pipeline on measured studs is left in, sheets lose a mean 2.34% and 11.49% at the 99th percentile. That is too much to let pass on 6 in joint edges. Nailing the joint edges at 4 in costs 16 more nails a sheet. With that, 99.98% of sheets with their misses left in still beat an intact 6 in sheet, and at 100 houses a year the sheet costs a median $6.98, against $7.10 for 6 in edges with every miss re-driven.
The check against tests, in one table. B. D. Weiss (Washington State, 2017) removed every nail in the bottom foot of OSB shear walls to simulate rot. Her decayed walls kept 56.5% (6 in edges) and 33.9% (3 in edges) of their strength. The paper's new Table 3 sets five models of her wall side by side. They are a limit analysis, the anchor rods' shear path, a spring pushover with slack at the rods, her CUREE cyclic protocol with pinching nails, and nails that lose strength with dissipated energy. At code-strength end nails the decayed 3 in wall keeps 63–74% in every one of them. The fits that match both tests all have the 16d end nails holding studs to the sill at about a quarter of their National Design Specification (NDS) strength. Among the spring models, all 30 of the 960 pushovers within 5 points of both tests do.
What that means. Either end-nailed stud-to-plate joints are much weaker in these walls than the design equations say, or the decayed walls fail by something none of the models has, such as the last nails above the unnailed strip tearing out or the middle stud splitting. A tested ultimate load for a 16d end-nailed joint would decide it. We searched again this session and found none we could reach; the closest public report is a roof-to-wall uplift test, a different load direction. For the robot the answer stands either way. In Weiss's two-sheet wall an escaped miss costs about the same under both readings (mean 2.73% at code strength, 2.67% at the fitted strength), because a scattered miss never puts the wall in the regime where the models disagree.
This is simulation only, checked against published tests. The end-nail joint's curve and the damage rate are swept, not sourced, and the sourced nails are not exactly Weiss's, so only shares are compared. The draft, its number checks and the code are in research/buildbot-sheathing.