GATE 2027 Tracker
A concept tracker with spaced review, test-score log, timer and to-do list. It is a PWA (installable on your phone) that runs on
Vercel with data in Supabase Postgres, so your phone and laptop always show the same data.
Accounts: sign up with email + password and choose your branch (DA or ME); you stay signed in on that device for a year. Everyone's data — syllabus, mistakes, tests, schedule, settings, screenshots — is completely separate (each account gets its own copy of the syllabus for its branch, and every query is scoped to the signed-in user). You can add the other branch to your account later.
- Put it online and install it: see DEPLOY.md.
- Try it locally first:
node dev-server.js → http://localhost:5050 (an in-process Postgres in data/pglite; create a local account on first visit). Node 22+.
Pages
| Page | What it does |
|---|
| Today | Days left, time studied, concepts due for review, today's to-dos, latest score per tier |
| To-do | Today / This week task lists (▶ starts the timer on a task, repeat weekly, missed tasks roll over), plus the syllabus checklist |
| Timer | Pomodoro or stopwatch; time is logged automatically (optionally against a to-do). Study log, weekly totals, consistency calendar. A running timer shows in a bar on every page |
| Concepts | Write a concept instead of a question: a title, a tag (🔴 struggled with / 🟡 must remember / 🟢 new in an exam), subject, optional note. Concepts are resurfaced on the review schedule below |
| Tests | Enter marks only. Tiers: weekly quiz → topic-wise → subject-wise → full mock, with an optional platform and link. Latest / best / last-3 average per tier and a progress chart |
| More | Branch switch, theme, settings (exam date, review intervals, syllabus editor) |
How the algorithms work
Review scheduling (shared/srs.js). Default intervals 1, 3, 7, 14, 30 days, each measured from the actual date of the previous review (so overdue reviews don't compress the spacing). A concept's state is always re-derived by replaying its review events, which is why editing dates, undoing a review or changing intervals can never leave it inconsistent.
- Solved independently → next interval; streak +1. Mastered after N consecutive independent solves (default 4) — then only maintenance spot-checks every 60 days.
- Partial → repeat the same interval, streak reset. Hint → one step back, streak reset.
- Failed / couldn't remember concept → back to step 0, retry after 1 day. A mastered item becomes Reopened.
- "Got it wrong again" in a later test is recorded as a recurrence (same effect, counted separately).
- Several reviews on one day never advance the schedule twice.
Priority score (shared/priority.js) is a plain sum of named parts: recurrences, failed reviews, severity, marks lost, days overdue, days since last success, your own importance rating (neutral at 3), and confidently-wrong answers. Mastered items are scaled down. Hover a score or open the detail view to see every line. Weights are in Settings. Nothing assumes a topic is "high weightage" unless you set its importance.
Today's plan (shared/planner.js): overdue → repeated/high-impact → due today → optional maintenance, then packed into your time budget (items that don't fit are skipped, smaller ones still fill the gap).
Accuracy is reported two ways and only when the data exists: among attempted (correct ÷ correct+incorrect) and among all (correct ÷ total questions). Error-rate trends use attempted questions as the denominator so "more practice" isn't mistaken for "getting worse".
Database
Postgres (Supabase in production; PGlite locally and in tests, so the same SQL runs everywhere). Schema changes are ordered migrations in src/db.js, applied automatically and tracked in schema_migrations. Tables: subjects, topics (self-referencing: subtopics), streams, causes, tests, mistakes, reviews (event log), tasks, study_log, topic_progress, settings, attachments (screenshots, resized in the browser before upload).
Security
Passwords are hashed with scrypt (salted). The API requires a signed session cookie (HttpOnly, SameSite=Strict, 1 year, renewed on use). Every table has a user_id, every query filters on it, and every id from a request is looked up together with the user id, so another user's id behaves like a missing one — covered by an automated cross-user test. Cross-site requests are refused. Optional INVITE_CODE limits who can create accounts.
Tests
npm test
Covers scheduling, priority, planner, analytics, API validation, bulk operations, tests↔mistakes linking, schedule tracker, restart persistence, login/session behaviour, attachments and path-traversal/cross-origin protections.
Layout
api/index.js Vercel serverless entry (all /api/* rewritten here)
src/ db (Postgres + migrations), store, validation, HTTP
public/ the app (no build step); public/shared/ = pure logic used by browser + server
test/ node:test suites
scripts/ icons, reset-password
Known limitations
- Needs internet (the service worker only keeps the app shell for faster loads; data is always live).
- Dates use each device's local day; the browser always sends its date (the server's own date is only a fallback).
- Mastery-progression chart shows what is currently mastered over time (a later reopening removes the earlier point).
- The Postgres code is tested on PGlite; the thin
pg connection adapter is only exercised on a real deployment.
