Building an exam platform alone, where a new exam is data and not code.

The mistake I wanted to avoid is common in single-exam sites: the exam's rules, subjects and prices are written into the code, so every new exam needs a developer. In Score Delta an exam is a record with a pattern (sections, timing, marking, negative marking) and a syllabus mapped to topics. The scoring code only ever reads the pattern. It never asks which exam it is scoring.

That made the rule testable on its own: scoring is covered by unit tests for each question type against each marking rule, including partial marks and optional questions. It also made the content side scriptable, because a whole exam bank, with its mocks, sectional tests and topic tests, can be built from one command that reads the exam's pattern. The same data is what the website, the app and the admin panel show, so there is no second copy to keep in sync.

The first version had periods of 1, 3, 6 and 12 months, a combo pass for every Himachal exam and several packs. On paper it was flexible. In practice I bought a plan myself while testing and could not find where it had gone: it appeared in My Plans, but I could not tell what it unlocked or where to use it. If the person who built it gets lost, a student will too.

So I cut it to two plans per exam, and removed everything else: a Mock Pack at Rs 259, one time, valid for 6 months, that opens the full mocks, and an Exam Pass at Rs 499 a month that opens everything for that exam. The tests list now says what is free, what is locked and what your plan opened, and the free and paid tests are mixed on the first page so a student can see what paid contains before deciding. The same change had to be made in the website, the saved FAQ and guide text, and the Android app.

I measured the live site with Lighthouse on a simulated phone. The home page scored 73 for performance, with first paint at 4.5 seconds. The page was small, only about 43 KB, but it waited on an external font stylesheet that pulled many font files from a second host, and my own security policy blocked those files, so they were requested, refused and logged as console errors.

The fix was to host the three fonts on the site itself as variable files, loaded only for the scripts a page uses, and to drop the two external hosts from the policy. I also fixed two contrast failures and a link whose label did not match its visible text. On 11 November 2026 the pages I measured scored 100 on mobile for performance, accessibility, best practices and SEO, and the desktop home page scored 99 for performance. Part of that depends on the server's response time and the network, so I treat it as a measurement, not a guarantee.

There is no evidence yet that students do better with it: it is new and I have no outcome data. The question banks are built by scripts and checked by automated tests, and I have not had each of the 14,000+ questions reviewed by a subject teacher. Only ten Himachal Pradesh exams are live. The Android app runs on my own phone but is not on Google Play yet. And the Lighthouse scores are one day's measurement of public pages, not of the logged-in screens.

In about three weeks, one person took Score Delta from a scope document to a live site with ten exams, thousands of mock tests, two plans per exam, an admin panel and an Android app. A new exam can be added from the admin panel and loaded with one command, and a student can go from a Google sign-in to a first free test and a weak-topic report without help.

What it isn't yet is a business with a track record. The next steps are the ones a startup has to do and no commit can do for it: get the Android app onto Google Play, add the next exams, and get real students to use it and tell me what is wrong.