Tollywood Analysis

A private, dashboard-first benchmarking app for Telugu-cinema heroes: named leaderboards, release-cadence and audience charts, a labelled performance scatter, a sortable table, side-by-side comparison, and a separate Annexure with film-level evidence, methodology, sources, coverage, change log and a correction workflow.

Real data is on by default. The app ships a snapshot built from Wikipedia (filmographies, release dates, budgets, worldwide grosses, box-office verdict sentences) and Wikidata (identifiers, runtimes, languages, recorded X follower counts). Only Telugu heroes and only films released from 1 January 2000 count. Every number links back to its source. Lead roles are auto-classified from billing order and are marked pending editorial check; use the correction form if one is wrong.

Set NEXT_PUBLIC_DATA_MODE=demo for the old synthetic dataset (marked DEMO DATA), or live to read Postgres.

---

1. See it running (no setup)

The quickest path, and it works from a phone:

1. Go to vercel.com/new and import this GitHub repository.
2. Keep all defaults and press Deploy. No environment variables are needed: the real-data snapshot is bundled.
3. Open the URL Vercel gives you.

To run locally instead (Node 20.9+):

npm install
npm run dev          # http://localhost:3000

2. What is in the app

A short top navigation (Overview, Heroes, Rankings, Compare, Trends, Inside the data) and one scrolling overview page: cinema banner with the top three heroes → filters → headline numbers → photo leaderboard (Star Score by default; success ratio, hits, blockbusters, total box office, plus "More measures") with tie-aware ranks → top-5 lists → career atlas → "Go deeper": two Pareto (best trade-off) charts with changeable axes and a "find a hero" list, plus collapsible box-office, release-pace, recent-form and career-reach sections → "Every hero, every metric" table with CSV → “How we calculate everything” and "About this snapshot".

• Bars are coloured by debut era or by film family (Mega, Nandamuri, Akkineni, Daggubati, Ghattamaneni, Manchu, other); switch with Colour by, and tap a legend item to hide a group.
• Every bar, dot and table row shows the hero's photo (free-licensed, from Wikimedia Commons; initials when none exists).
• Tap a hero to highlight him everywhere; double-tap / double-click to open his own page (/hero/<name>): banner with his headline numbers, the Star Score part by part, highlights, career milestones, results bar, year-by-year chart, comparison against the median hero, every statistic with his rank, and a large sortable table of every film (result with its source, worldwide gross, budget, gross ÷ budget, director, music, runtime, genre, language, billing) where each row opens its evidence; plus the appearances that were not counted and why. /heroes lists every hero as a card.
• Ratios and averages only rank heroes with at least 5 judged films, so a 3-for-3 newcomer can't top the chart.

Page	What it shows
/	The single-page dashboard described above
/hero/<slug>	One hero's page
/compare, /trends, /rankings	Extra views, linked from the footer and hero pages
/methodology	Every formula, weight and threshold, plus what is deliberately excluded
/annexure/…	Film-level evidence: methodology, registry, per-hero evidence, sources, coverage, change log, corrections, photo credits
/admin	Editorial admin (sign-in with ADMIN_TOKEN)

Fans' ranking (off by default)

Visitors pick their top three heroes (3/2/1 points), one ballot per device (httpOnly cookie; voting again replaces the ballot). To switch it on, set NEXT_PUBLIC_FEATURE_USER_RANKING=true and redeploy. Votes are stored in the user_rankings table when DATABASE_URL is set; without a database they live in server memory and reset on restart.

Hero photos

npm run photos:fetch downloads the lead portrait of each hero's English Wikipedia article only if it is hosted on Wikimedia Commons (free licence), saves it to public/heroes/, and records author, licence and file page in lib/data/hero-photos.json (shown on /annexure/photo-credits). Non-free images are skipped.

3. Where the real data comes from

npm run data:import rebuilds lib/data/real/snapshot.json (cached under .cache/import, gitignored):

1. Exports each hero's Wikipedia article and "… filmography" page via Special:Export (bulk, polite, no HTML scraping) and parses the film tables. TV, web series, songs, cameos, voice roles, re-releases, unreleased and non-Telugu titles are dropped.
2. Exports every film article and reads the infobox (release date, language, director, music, runtime, budget, gross).
3. Enriches from Wikidata SPARQL: release dates (P577), languages, genres, runtime, IMDb ids (links only), hero portraits (P18) and follower counts (P8687, split by qualifier into X / Instagram / YouTube).
4. Keeps a credit only if the hero is billed first, or second/third right after another roster hero (co-lead). Character actors use strict billing. Supporting, antagonist and extended-cameo parts that slip through (e.g. Allari Naresh in Maharshi) are removed in lib/constants/editorial.ts, with the reason shown on the hero page.

Each film's result comes from the first source that states one (lib/import/verdict.ts, lib/import/trade-blog.ts):

This is the beginning of the README. Read the rest at the original. Read the full README ↗