Skip to content

feat(gre): observed Performance on the dashboard (pin bump + docs) - #54

Merged
f15cubing merged 1 commit into
mainfrom
agent/gre-dashboard-performance
Jul 5, 2026
Merged

f15cubing merged 1 commit into
mainfrom
agent/gre-dashboard-performance

Conversation

@f15cubing

Copy link
Copy Markdown
Owner

Intent (fast lane)

Wire the GRE dashboard's Performance slot to a real score instead of the stale "Arrives Thursday" placeholder. Pure Qt-UI + a read-only view-model that pools the exam attempts Exam Mode already persists to a profile side-file and renders them as a Wilson range. No collection mutation, no scheduler/undo/store, three scores stay separate — so it rides the fast lane.

Summary

  • Bumps the anki submodule pin d11b424 → b7cf7c2 (feat(gre): surface observed Performance on the dashboard anki#3), which:
    • replaces the placeholder with observed rights-only accuracy as a Wilson range + n via the existing CalibrationStrip
    • adds pure load_exam_attempts() + observed_performance() to dashboard_data.py; greDashboardData reads the side-file best-effort (read-only)
    • ScoreSlot gains an observed state; the no-fabrication guard still collapses any range-less state to not_available
  • Docs (same change): qt.md (§ Observed Performance + footer SHA), models/performance.md (§ Live surface — observed vs the calibrated eval-only model), INDEX.md (dashboard row), STATUS.md (in-flight).

Honesty ceilings

Always a range, never a bare point; a give-up state (never a fabricated 0) with no attempts; scores never blended; deliberately the observed low-n surface, not the calibrated scoring/performance.py model.

Test plan

  • ./ninja check_svelte green (anki PR)
  • ./ninja check_pytest_aqt green; dashboard test files 23/23 (9 new)
  • Read-only invariant: endpoint only reads the mastery RPC + the side-file; no OpChanges
  • Live GUI smoke of the Performance slot with a real timed mock (human-gated)

Notes / follow-ups

  • Stacked on the current pin: anki PR bases on agent/exam-preset-fix (= d11b424); retarget to anki main once that merges.
  • Follow-up: carry observed Performance into the synced gre_scorecard (Android panel still not_available).
  • Pre-existing ruff-format debt in dashboard_data.py/mediasrv.py at the base is left untouched (additions are clean).

Made with Cursor

…oard

Bumps the anki submodule pin to f15cubing/anki@b7cf7c2 (observed Performance on
the GRE dashboard) and updates the codebase docs in the same change:

- qt.md: new "Observed Performance" section (Exam Mode attempts side-file ->
  pooled Wilson range via CalibrationStrip), updated dashboard_data/ScoreSlot/
  mediasrv notes, tests count 11->20, footer SHA bump; scoring-adapter honesty
  note clarified (synced card Performance still not_available -> follow-up)
- models/performance.md: "Live surface" section (observed accuracy vs the
  calibrated, eval-only logistic+Platt model)
- INDEX.md: dashboard row bumped + observed_performance code paths
- STATUS.md: in-flight entry

Pairs with f15cubing/anki#3. Read-only; three scores stay separate.
@f15cubing
f15cubing force-pushed the agent/gre-dashboard-performance branch from 7683740 to 46e3f3f Compare July 5, 2026 18:40
@f15cubing
f15cubing merged commit fb6a736 into main Jul 5, 2026
@f15cubing
f15cubing deleted the agent/gre-dashboard-performance branch July 5, 2026 18:40
f15cubing added a commit that referenced this pull request Jul 5, 2026
…c demo items (#58)

* docs(demo): refresh the demo plan to cover every merged feature + all spec demo items

The Sunday-cut demo plan had drifted behind main. Made it a comprehensive superset
(marked the spec-required 3-5 min core with a star; the rest is optional showcase):

- Added the missing SPEC-REQUIRED "review session" scene (basic + interactive MCQ,
  FSRS-graded) -- it was only implied in pre-flight before.
- Added the study feature (15%): the live interleaving toggle + the interactive
  "How this differs from FSRS" explainer, with the honest "ablation pre-registered,
  not yet run" caveat.
- Added interactive MCQ cards and the timed Exam Mode shell (with honest capacity
  gating -- only the Mini preset the held-out bank can fill).
- Updated the dashboard scene: Performance now shows an OBSERVED accuracy range from
  in-app Exam-Mode answers (PR #54), not the stale "not available".
- Refreshed proofs: make bench now reports next-card p95 0.33ms alongside the mastery
  650ms; added the re-runnable gates + interleave-report/ablation-analysis.
- Fixed stale state (engine SHA ea3acae -> 4c991c9, "PRs #32-#36") and expanded the
  vs-base-Anki table + pre-flight + honesty framing + runnable-proofs list.

The Milestone-1 appendix is preserved verbatim (its 2026-07-01 snapshot). Docs only.

* docs(status): record demo-plan refresh (#58)

---------

Co-authored-by: Felipe Caicedo <felipecaicedo@Felipes-MacBook-Pro.local>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant