influence-sensor

▶ Try it live: https://endika.github.io/influence-sensor/

Find out how captured your social feed is — which accounts actually own your attention, not just who you follow. Drop your Instagram data export and get a health score, an interactive graph, and the raw numbers behind it. (A limited, relationships-only mode also supports TikTok exports — TikTok hides who you engage with.)

Everything runs in your browser. Your export is never uploaded or stored — a strict Content-Security-Policy (connect-src 'none') makes that enforceable, not just a promise.

Use it

1. In Instagram, go to Settings → Accounts Center → Your information and permissions → Download your information, choose "Some of your information", and tick: Likes · Comments · Saved · Story interactions · Followers and following.
2. Set Format: JSON (not HTML) and Date range: All time. (Liked posts can't be analysed — Instagram no longer exports the post author — so Story interactions is what gives the richest result.)
3. Open the app and drop — or pick — the .zip you received. Read your report. Nothing leaves your machine.

TikTok (limited): Profile → ☰ → Settings and privacy → Account → Download your data, request your Activity data in JSON format, then drop that .zip. TikTok hides the creator of liked/watched videos, so only your relationships and activity volume can be shown.

How the health score works

The verdict is a 0–100 score (banded Captured / Moderate / Healthy), combining two opposing signals from how your interactions spread across accounts:

• Diversity — normalized entropy of your attention, in [0, 1]. Higher = more evenly spread.
• Concentration — the share of your attention held by your top 10 accounts. Higher = riskier.

health = round(100 * (0.5 * diversity + 0.5 * (1 - top10Share)))
band   = health >= 66 ? "Healthy" : health >= 33 ? "Moderate" : "Captured"

It is a transparent heuristic, not a scientific measurement — the app always shows the components and the raw ranked data next to it, so the number is never a black box. It measures attention spent (likes, comments, saves), because that — not your follow list — is what reflects who actually influences you.

Privacy

• No backend, no upload, no storage. Your export is parsed in memory and discarded when you close the tab — nothing is written to localStorage, IndexedDB, or cookies.
• connect-src 'none' in the page's Content-Security-Policy means the browser blocks any network request the app could try to make. It is open source and statically hosted, so you can verify all of this yourself.

Develop

npm install
npm run dev      # local dev server
npm test         # run the test suite
npm run build    # production build to dist/

Tech

Vanilla TypeScript + Vite, JSZip (in-browser unzip), d3-force (the ego graph). No backend, no UI framework. A normalized schema decouples per-platform adapters from the metrics engine, so new networks can be added behind the same analysis.