▶ Watch the 32-second launch video — every feature, from chapter weights to seat prediction. (Click the still, or the link.)
Every chapter, ranked by what actually gets asked — one PYQ topic-analysis report per chapter, built from 16,649 past questions — plus percentile-to-rank conversion and a JoSAA college predictor on seven years (2020–2026) of opening and closing ranks, every round of every year.
React 19 + Vite. The production build is a single self-contained
dist/index.html — fonts, CSS and JS are inlined, so it opens straight
from disk with no server.
npm install
npm run dev # http://localhost:5173npm run build # writes dist/index.htmlFonts are subset to latin + latin-ext only. The full round-by-round record costs about 1.3 MB of that: ~3.3 MB raw, ~1.6 MB gzipped, still a single file that works offline.
npm test # node --test, no browser needed
npm run typecheck # tsc --noEmitEvery chapter in the app is one markdown report in report/. That folder is
the source of truth: it holds the counts, the topic split, the sub-topics, the
quoted questions and the reading order is derived from it. Nothing about a
chapter is written twice.
npm run reports # report/*.md -> src/reports.generated.jsThe generator refuses to emit data that does not add up — every table in a
report must sum to the same headline count — so a bad report fails the build
instead of rendering a chart that contradicts itself. test/reports.test.mjs
re-checks the compiled output, and test/sequence.test.mjs fails if a chapter
is added to the archive but not to the reading order in src/lib/syllabus.ts.
Reading order is authored, not computed. Weight says what pays; sequence says what to learn first, and sorting by weight inside a unit puts Heat and Thermodynamics before Units and Measurements, which is backwards.
The rank data is packed from the JoSAA CSVs in the parent directory
(../josaa_2020.csv … ../josaa_2026.csv) into base-36 strings:
npm run data # regenerates src/data.generated.jsSeats are delta-encoded across rounds. Each record stores the seat identity, its round-1 opening and closing rank, then a three-character bitmask of two bits per round (absent / unchanged / changed) followed only by the rounds where a rank actually moved. Six rounds fit in three base-36 characters.
Presence is explicit because a seat can drop out of the allocation for a round and come back; a run of rounds is not always contiguous. Round 1 to the final round is not a rounding error — later rounds routinely admit thousands of ranks further down the list, so the round selector changes which seats a reader is told are reachable.
The decoder lives in src/core.ts (decodeSeats), and test/codec.test.mjs
asserts an exact round-trip against the CSVs at build time.
| Path | What it is |
|---|---|
src/core.ts |
All data and query logic. DOM-free; shared by tests and the app. |
src/lib/reports.ts |
Chapter model: priorities, rates, routes, search. |
src/lib/syllabus.ts |
Reading order, unit grouping, chapter scope lines. |
report/ |
One PYQ topic-analysis report per chapter (source of truth). |
src/reports.generated.js |
Those reports as structured data (generated). |
src/data.generated.js |
Packed rank data (generated by npm run data). |
src/lib/ |
Stores (theme, favourites, progress), formatting, routing hooks. |
src/components/ |
UI primitives, icons, chart components, app chrome. |
src/views/ |
One module per route. |
src/styles/ |
tokens.css (the design contract), base.css, app.css. |
tools/build-data.mjs |
CSV → packed strings. |
tools/build-reports.mjs |
Report markdown → structured chapter data. |
public/ |
PWA manifest, service worker, icon set. |
test/ |
node --test suites for the codec, prediction, institutes, routes and charts. |
Every colour, radius, shadow and easing curve resolves to a custom property
declared in src/styles/tokens.css. Nothing in a component hard-codes a hex
value, so light and dark stay in sync by construction.
- Each year ships its final round only (round 6 where a sixth round ran, round 5 in 2024 and 2026). Comparing rounds across years would be misleading.
- Quota
Anycollapses duplicate rows for the same seat to the best closing rank across quota pools, and names which quotas contributed. - A missing year for a branch means JoSAA allocated no seat for that combination that year — not that the branch did not exist.
- Ranks are JoSAA All-India ranks within a category, not JEE percentiles.
- NTA publishes no chapter weightage; syllabus priorities are estimates from past-paper trends.
