Input: the case (state/cases/N.json), the answers, and the queued comment (state/queue/N.json, final_comment).
Output: one line, VERIFY #N ok or VERIFY #N fix: <check number, what to change>. Answer each check yes/no from the
text in front of you, not from memory. post --approve on a REVIEW item needs this pass to say ok.
- Environment. Does the comment ask for a version, OS, graphics card or driver that the case's
environmentor the body already gives? (Then fix.) - Existence claims. Does it say a feature exists, is missing, or "can already be done"? Was it confirmed by
appcheckanddocs/ui-features.md? A command list cannot prove a UI feature absent. - Duplicate. Does the linked issue match the SYMPTOM and the AREA (same command, same error, same visible failure), not just a generic error ("operation failed", "boolean Union: no result")? Is its live state what the comment says?
- Restating. Does the comment repeat the report or the bot? (Start with the question or the answer.)
- Already given. Does it ask for something already attached or answered in a comment?
- Zip. Does it ask for a design? Only for a modelling failure where the design matters, always zipped, never for a crash, GUI or usage report, never when a file is attached. A kernel assertion in a "Crash:" report is a crash.
- Claims. Does it say "reproduced", "I confirmed", "workaround", "fixed in", "we will", "thanks", or promise anything for the maintainers? Only a run on the current release may say what was run, and what was and was not tested.
- Kernel. A kernel-area report gets the same treatment as any bug and is named as kernel work (Solidcore), with a matching tangency-labs/solidcraft issue or PR linked if one exists. It never says "upstream" or "tangency-labs/solidcore" (Solidcore is in SolidCraft since 2026-11-08). Nothing says kernel work is paused (Gabe, 2026-11-04).
- Privacy. No private or work file, no local path, nothing from
state/private-names.txt. - Thread state. Is the issue closed, labelled
in progress/claimed, linked to a PR, or did a person (or another triage account) comment after the case was built? (Then do not post; rebuild the case.) - One question. One specific question or none. No template question, no "more details".
- Wording. Plain, short, no jargon the reporter did not use, no fragments left from an edit, the reporter's language as a second paragraph when they did not write English.
- Index numbers. Profile, sketch, body and face numbers match the saved command and say their base.
profiles: [0]is "profile 0" (0-based), never "profile 1". - Capability limits ("limited to", "only works on") match the command's source, not only its list entry. (#7727:
Emboss was wrongly called planar-only;
crates/kernel/src/emboss.rsalso handles a tangent exterior cylinder.) - Cited issues. A linked issue cited for the same symptom, card, setting or area really says that: open it. (#7822 cited #6948 for a corruption "at a Quality setting"; #6948 names no Quality setting.)
- Panel and menu claims ("only toggles", "has no option for") match the option list in source. (#7861: the Sketch
Palette has ten options, built in
crates/engine/src/cmd/sketch.rsoptions_json, not one.) - "Nothing does X" claims hold for every command family that could do it,
construct.*included. (#7832: theconstruct.point.*commands make standalone points.) - Kernel framing only when the failing step runs in the kernel: a kernel error, or code under
crates/kernel. Anarea:kernellabel is not enough. (#7177: Point to Point move is computed in the Move dialog.) - "Not something SolidCraft controls" only after checking SolidCraft's own preferences
(
crates/ui-egui/src/prefs/*.rs, labels inassets/locales/en.json). (#7965: Preferences has a slicer-reuse option.) - Quoted project rules say what the document says (CONTRIBUTING.md, docs/), not what someone said it says. (#7967)
- Retest version is the current release from
gh release list -R tangency-labs/solidcraft -L 1(lint V01). - Crash records: what a record holds (a replay or not, which commands) is read from THIS report's own sections, not copied from a similar report. (#6830 copied #6839.)
- Preview vs OK: a hang that the live preview did not have is not in the kernel step the preview already ran; the
preview runs the same commands as OK (
crates/ui-egui/src/dialogs.rspreview_commands). (#8028) - Sibling already asked: the same failing step asked about on a sibling issue gets a pointer to it, not a second request. (#6699 and #6690)
- Signature. The comment holds no "Triage by", "Verified by" or other signature line: the kit adds the one header, as the posting account (lint S01). (Another account copied a dudedagger42 header on #5394, #3532 and #6260.)
- Claim (MULTI). You hold the claim (
claim Nsaid so), and no other account has a triage comment on the thread.
Any "fix": change the answers or the comment, submit again, re-run this pass once. Two failed rounds: do not post; write
the reason in notes and leave it for review.