Repository navigation
Description
What
Add a classical-CV print-quality tier to 3d printer vision and 3d printer watch --job: white stringing debris (flecks and strands), pattern loss (stripes, roughness, under/over extrusion), uneven first layer, with notice/warning/severe levels judged against the job's own first layers, a debris trend, one tg --photo message in Russian per issue and level (never a pause or cancel), and a per-job monitoring report (counts per severity, worst frame) saved next to the print/runs evidence.
Why (motivation)
2026-11-05: a 10 h sheet print on the Snapmaker U1 (270x270, PETG Wooden 232 C, many islands) stringed from layer 2 on, white thread debris all over the part; printer vision --once returned findings [] with p 0.0 q 0.0 on all 12 frames, the job watch stayed silent, and the owner saw it only on the real part (everything in stripes). The Obico spaghetti model has no stringing class and the job watch only checks that the part is still where it was.
User impact
The owner is told within minutes, with the frame and what to change (temperature, retraction, drying), instead of finding a ruined 10 h print; a human can read what the monitor saw after every job.
Cost of inaction
Every long multi-island print that strings or under-extrudes is discovered only at the end; the monitoring stack stays blind to the most common PETG defect.
Acceptance criteria
- On the 12 saved U1 frames of the 2026-11-05 SIMBOL sheet job f1-f5 raise nothing and f7-f12 raise at least WARNING for stringing (table frame -> verdict before/after in the PR). — forced: tests/test_print_quality.py, tests/test_printer_quality.py on the 12 saved U1 frames (tests/data/quality): f1-f5 raise nothing, f7-f12 raise at least WARNING for stringing; PR feat(printer): print-quality monitor — stringing, pattern loss, first layer in printer vision and the job watch (Refs #733) #734 CI green
- A quality finding is never CRITICAL: it sends tg --photo FRAME with a short Russian text once per issue and level and never pauses or cancels; part gone, spaghetti and the printer's noodle verdict keep acting as before. — forced: tests/test_printer_quality_watch.py: quality finding never CRITICAL, tg --photo once per issue and level in Russian, never pauses/cancels; part-gone/spaghetti paths unchanged (tests/test_printer_jobwatch.py); CI green
- printer watch --job and printer vision both use the tier; the job ends with report.json, report.md and worst-frame.jpg in a monitor-report folder next to the run evidence; --help and docs describe it. — forced: tests/test_printer_quality_watch.py + tests/test_printer_vision_cycle.py: job watch and printer vision both use the tier; report.json/report.md/worst-frame.jpg written; docs/commands/printer.md, printer-vision.md, docs/research/print-quality-monitor.md; CI green
- Tests use real downscaled frames (under 2 MB) plus synthetic cases; the pre-commit gate is green. — forced: real downscaled frames under 2 MB in tests/data/quality plus synthetic cases; commit hook (ruff, mypy, full pytest) green on every commit; CI green
Screenshots
(none)
Deploy
Deployed: no
Links
- PR: feat(printer): print-quality monitor — stringing, pattern loss, first layer in printer vision and the job watch (Refs #733) #734
- Merge commit: 338563d
Skipped gates
- links: false positives: severity names and a repo test-data path, not entities to link
Activity
fix(printer): per-model quality defaults come from a data file, not a…
merge: feat/bed-watch-tool into feat/print-quality-monitor (Refs #733)
merge: origin/main (squashed #727) into feat/print-quality-monitor; t…
feat(printer): print-quality monitor — stringing, pattern loss, first…
alex-mextner commentedon Nov 5, 2026
Shipped: merged via #734.
Merge commit: 338563d
This does NOT close the ticket — acceptance still needs proof. Run
task read #733for the outstanding criteria, thentask check/task done.feat(printer): macro passthrough and per-model alias commands (#721)