Jig previews: upload the renders and fail on jig warnings; record the structural known issues in the TODO

[dorkmo]

Summary

This PR has two independent parts:

1. Fix the jig-preview workflow. It has failed on all three of its runs on main. It now uploads the renders instead of committing them, frames each view so the whole jig shows, and fails when a jig doesn't evaluate or render cleanly.
2. Record the structural known issues in LifeTrac-v25/TODO.md, so the design work they need can be picked up later.

1. Jig-preview workflow

The Generate Jig Previews workflow has failed on all three of its runs on main, most recently when #134 merged. On push events, its "Commit and push renders" step ran git add renders/*.png renders/*.jpg. The root .gitignore ignores *.png and *.jpg, so git add failed. The renders were never committed, and the six images the jig README embeds have always been broken. The OpenSCAD review in #133 flagged this, along with the hidden render failures below (its CI table, generate-jig-previews.yml row).

Changes:

• Artifact instead of commit.
  • The jig-renders artifact is uploaded on every run, not only on pull requests, and kept for 90 days.
  • The commit step is removed, and so is the README-timestamp step, which only existed for the commit.
  • The job now only needs read access.
• Whole jigs in every view. Each render uses --viewall --autocenter, so the view fits the whole jig. Three of the six views used to run off the edges of the image: the pivot-mount jig's two views and the tube jig's front view. --camera still sets the view direction.
• Fail on OpenSCAD warnings. OpenSCAD 2021.01 exits 0 on warnings, even with --hardwarnings for PNG output, and writes a blank or partial image.
  • A new step evaluates every jig before anything is rendered and fails on any warning or error. When a jig fails to parse, it prints OpenSCAD's message.
  • The render step fails on any warning OpenSCAD prints while building the geometry, such as mixing 2D and 3D objects, which the evaluation can't see. The three render steps are now one step with a helper, so every view gets this check.
  • The || echo "Warning…" after the pivot-jig renders hid real failures too. It's removed.
• When it runs. It also runs when lifetrac_v25_params.scad changes, because every jig includes it, and on pull requests that change the workflow, as pushes already did.
• README. The jig README now says where to get the renders instead of embedding images that were never committed. Downloading an artifact needs a GitHub login, and the link goes to the latest successful run on main.
• Review. The 2026-09-25 review's follow-up status table records this PR.

Verification. I ran the workflow's steps locally with OpenSCAD 2021.01, the version CI installs.

• Current jigs: the checks pass with no warnings, and all 6 views render with the whole jig in frame, plus their JPG copies. The upload paths match the 12 images and leave out renders/.gitkeep.
• Unknown variable planted in a jig: the evaluation step fails and prints the warning.
• Include pointed at a missing file: the evaluation step fails and prints the warnings.
• Parse error planted in a jig: the evaluation step fails and prints the parser error with its line number.
• 2D and 3D objects mixed in a jig: the evaluation passes, as expected, and the render step fails with the warning.

The old workflow passed in every one of these broken cases.

The alternative. If you'd rather see the images in the README on GitHub, the renders could be committed instead. That needs a .gitignore exception for renders/, like the existing one for assembly.png, and keeping a fixed commit step. The artifact approach avoids bot commits of binary files to main.

2. Structural known issues in the TODO

The rebuilt structural analysis (#136) fails 8 checks, which it lists in STRUCT_KNOWN_ISSUES. A new ## Structural known issues section in LifeTrac-v25/TODO.md records them as ST-0 to ST-11, and section J of the pre-field-deployment checklist points to it:

• ST-0: choose the bucket cylinders. They set every bucket-case load. The model's 3" cylinders and the BOM's 2" × 1" ones disagree, and a run of the model with the 2" ones still fails the arm, the arm tip, the U-lug and all three lug-bolt checks.
• ST-1 to ST-8: the 8 failing checks, each with its numbers, cause and the review's suggested fix (B4, P14, M8). It also notes that the model cuts every U-lug to 76.2 mm, so the bucket lugs have no holes for the bolts that ST-6 and ST-8 check.
• ST-9: the 92 kg rated operating capacity. It explains why the capacity is low (the load's reach ahead of the front axle), what would raise it, and what's known about the mass estimates.
• ST-10 and ST-11: two part-drawing issues found while merging v25 OpenSCAD review, plus fixes for broken part exports and silent CI jobs #133 and Rebuild the structural and stability analysis for the current v25 design #136:
  • Revision letters flip between CI runners, because a fitted hole centre sits on a rounding tie.
  • The part drawings don't carry the "Not yet ready for fabrication" warning.

Verification. Independent checks compared every figure, dimension, link and citation against main's analysis log, the model and the review, and a second pass re-checked every correction. A later review re-ran the model to confirm the figures added after that, including the 285 kg tipping load with the front axle 100 mm forward. The section renders with GitHub-flavoured Markdown, and all 14 links point at files on main.

Branch history

main was merged into this branch so the review's follow-up table, which #136 added, could record this PR. The analysis bot then rewrote the log's header on this branch, and the last commit puts the log back to main's copy. So the PR's diff against main is exactly these five files: the workflow, the jig README, renders/.gitkeep, TODO.md and the review.

After merging

The jig README change sits under openscad/, so the merge re-runs the part-drawings workflow on main. Its bot may commit refreshed drawings with P5's revision letter going up. That's the revision-flip issue recorded as ST-10, not a change from this PR.

🤖 Generated with Claude Code

https://claude.ai/code/session_01A4ZY2kr3UU7D2vK8aHjmkS