STACK q-type Analytics Docs / Chapter 7

Architecture & Design

Moodle (PHP)
     |
     | reads quiz_attempts/grades/log events via the question engine,
     | gradelib, and logstore_standard_log directly
     v
  Moodle DB
     |
     +--> classes/quiz/analytics/*.php    (Quiz Analytics + Question
     |     Analytics: STACK/Maxima response parsing, statistics, chart
     |     JSON, PDF layout)
     |
     +--> classes/stack/analytics/*.php   (Model Analytics + Diagnostics
           Analytics: indicators, targets, report builders, PDF layout)
     |
     v
Plotly.js / KaTeX (client-side rendering) or TCPDF (server-side PDF)
  • No CSV round-trip, no external service. classes/quiz/data_fetcher.php reads finished attempts straight out of quiz_attempts via Moodle's question engine; classes/stack/local/stack_attempt_reader.php does the same for Model Analytics/Diagnostics Analytics.
  • STACK question text is rendered through STACK's own CAS engine (castext2_qa_processor), never read as raw stored questiontext.
  • Every computation is pure PHP — no Python, no external service, no Composer dependencies at runtime.
  • Caching (Quiz Analytics/Question Analytics only): every fetch/compute path is backed by a Moodle MUC cache area, keyed on a cheap SQL fingerprint (attempt count + latest timefinish + summed grades) rather than a fixed TTL — an entry is only ever served while that fingerprint still matches. Model Analytics/Diagnostics Analytics have no result cache; every dashboard view recomputes from scratch.
  • Analytics API integration (Model Analytics): db/analytics.php registers both prediction models via \core_analytics\manager::update_default_models_for_component(), consumed automatically by core on install/upgrade.

See How Everything Is Calculated for the actual math behind every number this produces, and Privacy & Security for what data this touches and where it stays. The proxy-label circularity caveat referenced throughout How Everything Is Calculated is documented in full there. This plugin is itself the merger of two previously separate, independently installed plugins (local_quizanalytics and local_stackanalytics) — see CHANGELOG.md for the full history of that merge, including the renames along the way.

Known, tracked gaps

Documented rather than hidden — see README.md for the current list, currently:

  • Two Model Analytics indicators are documented simplifications of the architecture doc's literal spec — question_difficulty_irt's classical- test-theory proxy instead of a jointly-fitted 2PL IRT model, and feedback_ineffectiveness's aggregate log-odds effect size instead of a per-branch paired McNemar's test — both because the fuller version needs a batch step the Analytics API's per-sample indicator model doesn't provide. Detail: How Everything Is Calculated.
  • concept_dependency_report (concept-dependency mapping across questions) is an intentional placeholder, not implemented — the architecture doc frames it as offline sequence-mining work better suited outside a live dashboard page.
← Previous6. Privacy & Security