The tool targets any LTI 1.3 platform; Canvas is the primary target
but nothing outside the Canvas-specific custom variable substitutions
depends on it. User-facing strings and the README now say "LMS"
except where a mechanism genuinely is Canvas's (the $Canvas.* /
com.instructure.* substitutions and their handling), and the README
states plainly that the tool has so far been exercised only against
the saltire emulator, not yet a production LMS.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Instructor home shows a setup banner while a course has no class
schedule, since that is the one gap that degrades the student
experience (free-date opt-outs, no withdrawal locking); it points at
the prefilled schedule form and settings and disappears once any
class day exists.
Calls gain assessment_entered_at, stamped when an outcome is recorded
or actually changed (bulk saves that change nothing don't restamp),
null for pending calls and legacy imports, and included in the calls
export — distinguishing assessed-in-class from assessed-later.
The README setup section now gives both install routes: pure
pip/venv, and Debian system packages with the apt line. Fixing this
uncovered that numpy was missing from the project dependencies (the
grading engine needs it; the system-site-packages venv had masked
the omission).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The README now documents the daily workflow (working date, live mode
vs printed lists, regenerate semantics, full day-editor editability),
describes cycle mode's rolling rotation accurately, repairs the
custom-parameters section, and points to the saltire testing guide,
which gains resume-after-a-break instructions.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Ports the timing-neutral foregone-participation grading scheme from
participation_grades.R: answer quality (per-course level points) minus
a deduction for participation foregone through unavailability,
estimated by Monte Carlo simulation of the actual weighted draw and
averaged over when absences fall, plus a small form-filing incentive
for drawn-while-absent-without-opt-out days. Reason-blind and
luck-protected; zero-answer students floor to 0. Parameters
(allowance in SD units, passing line, form penalty, simulation
size/seed) are course settings. Verified against the R engine's
rendered 2026q2 reports via the new import-legacy command: quality and
availability match exactly, finals within Monte Carlo noise; dropped
students import as inactive enrollments and are excluded identically.
Grades are computed on demand into stored GradeRun snapshots and
reviewed on a grades page with CSV export and per-student reports
(also served to students via a publish toggle). Display scales map
points to UW 4.0, a threshold table (one-click import of the Canvas
course grading scheme), or raw points. Gradebook passback via AGS
sits behind a settings toggle with a review-then-push flow.
Opt-out withdrawals are now soft-deletes with a withdrawn_at audit
trail, and close when class begins (class days gained optional start
times), so availability records cannot be rewritten after the fact.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Instructor participation report: per-student histograms, outcome mix
by class day, and a sortable table including the fairness ratio
(answered calls over questions present for, with opt-out days out of
the denominator), plus CSV exports of students, calls, and opt-outs.
Assessment scales are now per-course data: ordered levels with labels
and points out of 100 (defaults carry the old R grading values), with
calls referencing levels by id so renames follow through to history.
Renaming, re-pointing, reordering, and adding levels are always
allowed; deleting a level in use by recorded calls is blocked.
Pronouns and course term dates come from Canvas custom variable
substitutions, at launch and roster-wide via rlid-scoped NRPS; the
student page notes that names/pronouns are Canvas-sourced. Rosters
can also be refreshed outside launches: a "Sync roster now" button
and a sync-rosters CLI command for an hourly cron job, skipping ended
courses. Alembic now runs SQLite-compatible batch migrations with a
constraint naming convention.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
OIDC login, launch, and JWKS endpoints built on pylti1p3next's Flask
adapter, with launch-claim processing split into a testable module.
Instructor launches refresh the roster through NRPS; the sync code is
source-agnostic and also drives the dev-mode fake roster. Dev mode
(COLDCALL_DEV_MODE=1) provides fake instructor and student personas
that set up the same session state as a real launch, so the rest of
the app can be developed before a Canvas Developer Key exists.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Flask application skeleton with SQLAlchemy models for courses (including
per-course settings), students, enrollments, opt-outs, and calls; the
weighted and cycle selection logic ported from the manual coldcall
scripts; alembic migrations; and a pytest suite covering selection
behavior and the model query helpers.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
