Changelog

[1.7.0] - 2026-06-18

Changed

• Complete/incomplete (self-study) assignments are now auto-excluded from feedback analytics by a grade-type heuristic, replacing the purely manual flag. Scale-graded assignments (assign.grade < 0) typically never receive written feedback, so they are treated as not feedback-relevant by default instead of dragging coverage down. The programme coordinator stays in control: the drill-down now shows every assignment (excluded ones greyed with an Auto-excluded / Excluded badge) rather than hiding them, and each row carries an inline override — a fa-ban icon (tooltip "Exclude") to drop a feedback-relevant assignment, and a fa-plus-square icon (tooltip "Include") to count an auto-excluded one anyway. Overrides are stored minimally as two config keys (feedback_excluded_cmids = forced out, feedback_included_cmids = forced back in); a choice matching the heuristic stores nothing. feedback_exclusions now exposes is_scale_graded(), get_included_cmids(), and apply_state() (returning the effective state), and is_excluded() reflects heuristic-plus-override. The "How is this determined?" methodology card documents the heuristic.


• The drill-down web service annotates instead of filters. get_assignment_feedback returns each row with scalegraded, excluded, and excludedauto flags (sourced from the sibling gradereport_coifish breakdown's new scalegraded field) so the coordinator can see and reverse what the heuristic decided; toggle_feedback_exclusion returns the effective excluded/excludedauto state. The AMD module re-renders the toggled row in place from its client-side cache.

[1.6.0] - 2026-06-17

Added

• Coordinators can mark an assignment as "not feedback-relevant" directly from the lecturer-profile "Feedback by course" drill-down — for complete/incomplete self-study activities that never receive written feedback and would otherwise drag feedback coverage down. Each assignment row gains an inline "Not feedback-relevant" action (coordinators only, gated on local/coifish:viewlecturerprofile); clicking it calls the new AJAX web service local_coifish_toggle_feedback_exclusion(cmid, excluded) and removes the row. The exclusion is global per assignment and backed by local_coifish\feedback_exclusions (config key local_coifish/feedback_excluded_cmids); get_assignment_feedback now filters excluded cmids out of the drill-down before returning, and the sibling gradereport_coifish change drops them from the feedback-coverage denominator.

[1.5.1] - 2026-06-17

Fixed

• Assignment drill-down rows now line up with the course-table columns. The per-assignment breakdown was rendered as a separate nested <table> inside a full-width cell, so its columns auto-sized independently and drifted out of alignment with the course header row. The rows are now rendered as <tr>s in the same table (each course's detail is a collapsible <tbody>), so every column aligns with its heading. Assignment names are indented to keep the nested-under-course cue.

[1.5.0] - 2026-06-17

Added

• Feedback-by-course drill-down on the lecturer profile so a coordinator can see which courses make up a lecturer's composite feedback-quality score, rather than the number being opaque. A collapsible "Feedback by course" card lists each course (worst composite first) with its coverage/depth/quality/personalisation, composite, and with-feedback/graded counts, via lecturer_api::get_feedback_breakdown() — one query over the gradereport_coifish_feedback cache, scoped to the same in-scope courses (and course filter) as the rest of the profile, and consistent with the dilution fix (only courses with totalgraded > 0). Read-only.


• Lazy per-assignment expand. Each course row expands on demand to its per-assignment feedback breakdown, fetched once over a new AJAX web service local_coifish_get_assignment_feedback(userid, courseid) and cached client-side per course (never eager-loaded). The web service guards \gradereport_coifish\report::get_assignment_feedback_breakdown() with class_exists/method_exists and returns [] when the optional grade report is absent. New AMD module local_coifish/feedback_breakdown.

Changed

• Feedback-quality scores are no longer diluted by ungraded courses. A lecturer's avgfeedbackquality (and the coverage/depth/personalisation averages) now average only over courses where they actually graded something (totalgraded > 0), instead of being dragged toward zero by courses they hold a teaching role in but do not grade. Applied consistently across the cached profile, the filtered/range view, and the weekly trend snapshot.


• Removed an N+1 query from the daily profile build. build_lecturer_profiles previously issued one forum-post count query per course inside the per-lecturer loop; it now uses a single grouped query per lecturer.

[1.4.3] - 2026-06-12

Changed

• Active-snapshot task is much lighter on the database. build_active_snapshots now skips students whose snapshot is unchanged — it compares each student's last logstore activity and last gradebook change (two per-course aggregate queries) against the snapshot's compute time, and only recomputes the ones that actually moved. A jittered 7-day TTL forces a periodic rebuild so course-structure drift still self-heals, spread across days to avoid a synchronised spike. The run logs refreshed N, skipped M via mtrace.


• Per-course-invariant work hoisted out of the per-student loop. The expected-activity count and the forum-discussion list (previously re-queried for every student inside capture_student_metrics) are now fetched once per course and passed in, removing two queries per student. capture_student_metrics gained optional $totalactivities / $discussions parameters (callers that omit them are unaffected), and the discussion query is exposed as metrics_helper::get_course_discussions().

[1.4.2] - 2026-06-12

Fixed

• The course_category setting now applies to lecturer profiles. Previously the category limit was honoured only on the student/risk-overview side, so the lecturer profile (cached and filtered), the profile course dropdown, the weekly trend snapshots, and the per-course time report counted courses in every category (including dev/test courses outside the configured tree). Every lecturer-side course-selection query now applies filter_helper::get_category_scope_sql(), consistent with the risk overview. The time report additionally gained the lecturer_excluded_courses filter it was missing.


• Student-outcome averaging no longer silently drops course rows. build_lecturer_profiles() grouped the per-course outcome query by (courseid, courseenddate) while keying the result on courseid, so a course with more than one end-date value lost rows (and skewed the average). Now grouped by courseid alone with MAX(courseenddate) for trend ordering.


• Filtered lecturer profiles no longer emit a fullname() debugging notice. compute_lecturer_profile_for_range() now fetches the full set of user name fields, matching the other two profile paths.

[1.4.1] - 2026-06-12

Changed

• Grading turnaround now clocks from the first grade, not the last edit. All three turnaround sites (lecturer_api::compute_lecturer_profile_for_range(), lecturer_api::build_lecturer_profiles(), and lecturer_period_snapshot::upsert()) now measure submission ' assign_grades.timecreated instead of assign_grades.timemodified, so a grade correction made months later no longer retroactively inflates a lecturer's turnaround. The weekly snapshot's period bound is likewise on first-grade time. The shared expression matches the gradereport_coifish plugin for consistency.


• Integrity referrals pause the turnaround clock. Where the optional local_unifiedgrader_referral table is present (Unified Grader installed), an item escalated for an academic-integrity review stops the lecturer's clock at the moment of referral (1-stamp model: submission ' escalation), so delays outside the lecturer's control are not counted against them. Every query that touches the table is guarded by table_exists() and degrades gracefully to the first-grade base when it is absent.

Added

• Per-assignment turnaround drill-down via lecturer_api::get_turnaround_breakdown() and a collapsible "Turnaround by assignment" card on the lecturer profile. Each row shows the raw (first-grade) vs adjusted (referral-paused) average days, the number graded, and a badge counting items held for integrity review. Assignments averaging more than 7 days are highlighted. Read-only.

[1.4.0] - 2026-06-09

Added

• Historical lecturer-activity trends with weekly per-lecturer caching. A new local_coifish_lecturer_period_snapshot table holds one row per lecturer per ISO week. The existing daily task writes the current week's row; a new hourly backfill_lecturer_snapshots task fills missing past weeks at a bounded rate per run (default 50/run) until the configured horizon is met. Past weeks are written exactly once and never recomputed — read path on the lecturer profile is a single indexed SELECT, no log scans.


• "Historical trends" card on the lecturer profile rendering inline SVG sparklines (no chart-library dependency) for feedback quality, grading turnaround, forum posts per week, intervention effectiveness, hours total, and average student grade. Each sparkline shows the latest value with a Δ% vs the window start.


• Admin setting lecturer_backfill_years (default 3, range 1–10) controls how far back the historical backfill goes.


• Dynamic paginated tables on Student Risk Overview and Lecturer list using Moodle's core_table\dynamic API. AJAX pagination, sortable columns, and configurable page size. Summary count cards (high/moderate/total) move to a separate aggregate query so the numbers reflect the full filtered population rather than just the visible page.


• Defensive hard cap (MAX_RISK_OVERVIEW_ROWS = 1000) on api::get_risk_overview() to guard programmatic / web-service callers against accidentally pulling unbounded result sets.

Changed

• Risk Overview filter dropdowns preserved as URL state but the table itself now refreshes via AJAX. Bookmarks with existing ?categoryid=…&risklevel=…&enrolstatus=… continue to work.


• metrics_helper::capture_student_metrics() signature gains a $starttime parameter applied to the three logstore_standard_log queries (engagement, feedback view, self-regulation) plus the forum_posts and assign_grades time-bounded supporting queries. Callers populate from the course start date.


• lecturer_api::estimate_activity_hours() signature changed from (uid, courseids, $now) to (uid, courseids, $timefrom, $timeto) with both bounds applied to the marking and communication logstore_standard_log queries. Callers pass either the user-selected date range or a 120-day cron lookback. Each UNION branch in the local_unifiedgrader fallback now uses unique :tfromN/:ttoN placeholders.


• Cohort programme pattern resolution factored into filter_helper::get_course_ids_matching_pattern(). Simple anchored patterns (^THE, ^BIB2) translate to SQL LIKE 'THE%' and run server-side. Complex patterns fall back to a category-scoped PHP regex pass — never the whole course table. Per-request memoisation cache included.


• Renderables transitioned to the Output API. The CSV export form's PHP-echoed HTML moved to templates/export_form.mustache with a new \local_coifish\output\export_form renderable. The cohort-programme admin setting's inline <script> and HTML moved to templates/cohort_patterns_setting.mustache + amd/src/cohort_patterns_setting.js loaded via $PAGE->requires->js_call_amd(). No remaining PHP-built HTML strings or inline JS in the plugin.

Fixed

• N+1 query patterns eliminated in seven hot paths by bulk-loading users and existing snapshot rows ahead of loops. refreshcurrent.php pre-loads the student's existing active-snapshot rows; build_active_snapshots::refresh_course() does the same per-course; build_profiles::snapshot_course() pre-loads existing userids via get_fieldset_select; api::get_early_warnings(), lecturer_api::get_all_lecturer_profiles(), local_coifish_generate_export_data(), and external\get_lecturer_time_report all bulk-load user records using \core_user\fields::for_name(). The two lecturer-time call sites also bulk-load (lecturer, course) role assignments in one query. Query counts now constant per page render.


• logstore_standard_log queries no longer scan unbounded history. All five queries in the plugin apply strict time bounds aligned with the table's shipped course-time and user-module indexes (both ending in timecreated), turning range scans into index seeks. Addresses peer-reviewer concern about production log-table size.


• get_lecturer_profile() no longer triggers Moodle 5.x "missing name fields" debugging notices. User record fetch now uses \core_user\fields::for_name() so all fullname-relevant columns (including phonetic and alternate names) are present on the object passed to fullname().

[1.3.4] - 2026-05-20

Changed

• Longitudinal feedback-review metric now counts Unified Grader feedback views alongside the native mod_assign events, via the shared \gradereport_coifish\report::get_feedback_view_event_sql() helper. Snapshots taken against courses that use UG for assignments / quizzes / forums / BBB no longer under-report student engagement with feedback.

Fixed

• Peer-review blockers: added GPL v3 LICENSE file in the plugin root (was missing), and frankenstyle-prefixed the only global helper generate_export_data() ' local_coifish_generate_export_data() in export.php to remove the namespace-collision risk flagged by the Moodle plugins-DB reviewer.

[1.3.3] - 2026-05-19

Fixed

• False "persistent / across courses" recommendations on single-course students. social_isolation and feedback_neglect risk factors fire on the per-course average, so they correctly flag a one-course student with very low forum/feedback activity — but the recommendation text claimed a cross-course pattern that didn't exist yet. api::generate_recommendations() now selects a _single wording variant ("In their recent course, …, watch whether the pattern repeats") when coursescompleted < 2. Same treatment for intervention_positive / intervention_unresponsive based on totalinterventions < 2 so a single intervention isn't described in the plural. Badge labels risk_factor_social_isolation and risk_factor_feedback_neglect are also reworded to drop the "persistent / across courses" framing.


• Snapshots for concluded courses no longer skew longitudinal data with post-course activity. metrics_helper::capture_student_metrics() now accepts an $endtime parameter; the build_profiles task passes $course->enddate so engagement views, forum posts, feedback views, assignment-grade windows, and grade-checks are clamped at course end when the snapshot is written. build_active_snapshots is unchanged — it already excludes concluded courses.


• Engagement metric in longitudinal snapshots now honours grade-category drop/keep rules via the shared \gradereport_coifish\report::get_expected_activity_count() helper. Optional assignments/quizzes in "best N of M" or "drop lowest N" categories no longer inflate the engagement denominator.


• install.xml now declares the XMLDB schema namespace (resolves core\db\plugin_checks_test::test_db_install_file failure).

[1.3.2] - 2026-05-13

Added

• Enrolment-status filter on Student Risk Overview — new dropdown with three options: "Currently enrolled" (default), "All", and "No current enrolment". Reduces noise from students on a study break, and surfaces re-engagement opportunities. Sourced directly from user_enrolments so it works even before the active-snapshot task has run.

Fixed

• Empty Current Enrolments card for new-term students — capture_student_metrics now returns a row with grade = null (instead of bailing out) when a student has no grade_grades entry yet or when the course has no grade_items row of type course. The active-snapshot task and the on-demand refresh both write the row in this case. Post-course build_profiles continues to skip students without a final grade.


• Non-academic courses leaking into current-enrolment queries — the existing local_coifish/course_category "Limit course matching to category" admin setting is now honoured by the active-snapshot task, the on-demand refresh, the Current Enrolments display, and the "Currently enrolled" filter on the Risk Overview. A new helper filter_helper::get_category_scope_sql() centralises the include logic.


• Stale active-snapsh