We built KitCat because CAT prep in India has become theatre. Two-year subscriptions with recorded lecture libraries you will never finish. Weekend mocks with analytics that tell you what you already knew. Chatbots that call themselves “AI coaches” and paste the third Google result. Testimonials from selectors filmed with too much lighting.
You do not need any of that. You need a map of where you stand. A plan that shortens as your accuracy climbs. A coach that has read yourprep, not the internet's. Mocks that respect the actual CAT blueprint — four RC passages, four DILR caselets, mixed-difficulty QA, one hundred and twenty minutes, no upsells.
Take the diagnostic. It costs nothing and takes two hours. If you would rather not sign up after, we will show you your weakness map and let you leave with it. If you do sign up, it is one plan, nine hundred and ninety-nine rupees a month, cancelled from your account settings in two clicks. We do not run trials — the diagnostic is our try-before-you-buy. Everything after the paywall is what it says on the tin.
A partial catalogue.
Names withheld out of professional courtesy.
Two hundred hours of talking heads on video. Sold by the year. Watched by nobody past the first week of enrolment.
One paper. One percentile. One weekend analytics dashboard that boils down to 'attempt more questions'. Repeated for thirty-six months.
Rebadged as 'AI coach'. When asked why an inference option fails, it paraphrases the passage back at you.
Testimonials filmed under studio lighting. Screenshots with company watermarks removed. Selectors credited to platforms that never taught them.
None of the above teaches you the exam. All of it fills a subscription pipeline. The gap between the two is where a serious aspirant loses eight to twelve months of preparation, sometimes twice.
What follows is the entire product. There is no separate ‘premium’ tier, no add-on packs, no coach-hour credits. You get everything, or the diagnostic.
A complete CAT-shaped paper — VARC, DILR and QA, forty minutes each — spanning the subtopics in our taxonomy and calibrated to the same difficulty band as a live paper. The output is a weakness map, not a percentile.
3 sections · 120 minutes · anonymous
Four phases keyed to your exam date, with a specific drill each day derived from your weakest live subtopic. The plan compresses as accuracy climbs — the point is to end before the exam, not fill the calendar.
Foundation · Sectional · Full-test · Last-mile
Claude Sonnet, briefed at every turn with your weakness profile plus your last ten messages, grounded in retrieval over 49,711 chunks indexed from real CAT prep material. Every claim points back to the note it came from.
49,711 note chunks · citations attached · image-aware
Four full mocks and a custom builder, each following the exact blueprint — four RC passages of 350–500 words with four questions apiece, four DILR caselets with real data, twenty-two QA with a thirty-percent TITA ratio. One hundred and twenty minutes.
68 questions · 4 RC · 4 caselets · 30% TITA
This is not a lightened sample. It is one of the questions in our diagnostic bank, typeset here in full.
Passage · The Phillips Curve Mirage
When economists first noticed a stable trade-off between unemployment and inflation in the 1950s, they treated it as a discovery — a durable regularity, akin to a physical law. By the 1970s the pattern had inverted; stagflation reigned, and the profession's response was not to abandon the framework but to add expectations to it. When those refinements failed to predict the quiet inflation of the 1990s, the curve was pronounced dead — again — before returning, in slightly modified form, in the 2010s. Each generation of economists appears to hold two views simultaneously: that their particular version of the curve is the definitive one, and that the field's previous confidence in a definitive version was misplaced. This is not so much a sign of intellectual failure as of a discipline that mistakes recurrence for endurance.
Question
The author's central point about the Phillips Curve is:
Every reply is built from three sources. Your weakness profile — updated after each attempt. Your last ten messages in the thread. The three most relevant chunks from a 49,711-item note library indexed from real CAT prep material. Answers arrive with the source note attached, quoted verbatim, page anchor preserved.
I keep missing RC inference. What's the actual pattern?
Coach
Looking at your last twelve RC attempts, the miss is not on the passage — it's on the choice. You anchor on options that restatea fact and tweak one quantifier (‘most’ → ‘some’, ‘caused’ → ‘contributed to’).
Try the negation test on inference: swap the option's claim with its opposite. If the passage still permits the opposite, the option is a restatement, not an inference.
Reconstructed from a live thread · student consented · names removed
Every mock is generated against the real CAT blueprint. Four RC passages of 350–500 words with four inference-heavy questions each. Four DILR caselets with real data — arrangements, caselets, table interpretations. Twenty-two QA items across arithmetic, algebra, geometry and number theory, with a thirty-percent TITA share. One hundred and twenty minutes on the clock.
every one labelled. every one drilled.
The CAT really tests this many distinct ideas — we built a question bank that respects that. Every one of the 2,621 questions is labelled by subtopic and difficulty, so your practice can adapt in something closer to one-question-resolution.
Extracted, OCR'd, deduplicated, and chunked into 49,711 note fragments that the coach cites verbatim. The corpus is the product.
Recent entries from the founder's notebook,
in reverse chronological order.
Audited every number on this page against the system it describes. The worst was here in plain sight: "every one of the 4,725 questions", written when the bank held 4,725 and left there after we withdrew everything we could not stand behind. The real figure was 2,567. Elsewhere a coach note told you that "roughly eighteen percent of test-takers who reach 90+%ile" spot a particular trap — a statistic we had never measured and had no way to. That one is gone rather than corrected. The counts you now see are read from the database at request time, so the page cannot drift from the product again.
— Ram
A generated explanation is only as good as the key it is explaining, and our keys came out of OCR'd PDFs. So every disputed key was re-solved from scratch by models that could not see the stored answer or each other's work. Where two independent solvers agreed against the key, 307 keys were corrected. A third model settled 116 more. Where all three disagreed, we did not pick a winner — those questions were withdrawn. That is the uncomfortable part: the bank got smaller every time we checked it, from 4,866 down to 2,567. What is left is the part where a worked solution and an independently confirmed answer both exist.
— Ram
Found while auditing something else. TITA answers are typed on a numeric keypad, and grading compared the typed string to the stored key character by character. Any key that carried a unit — "7.1%", "1,200", "250 km" — could never be matched by a student who was completely right. They would type 7.1, see a red cross, and reasonably conclude they had misunderstood the question. Grading now compares the numbers. Twenty-one items whose stored answer was not a number at all, like "Bottle 105 cents, Cork 5 cents", turned out to be multiple-choice questions that had lost their options in extraction; they are withdrawn.
— Ram
Most DILR questions had been unpublished for asking about a table the student could not see. The recovery script only understood documents with "SET n" headings, which meant it skipped the real CAT past papers entirely — they carry no such heading. What they carry is better: the extraction had fenced the shared data explicitly. Reading those fences, plus noticing that each set's opening question had the whole caselet glued onto its stem, brought back the sets. Separately, a lot of DILR needs no caselet at all — a seating puzzle states its own scenario — and we had been throwing those away with the orphans.
— Ram
Checkout depended entirely on a Razorpay webhook to activate a subscription. Webhooks cannot reach localhost, so activation never happened in development, and in production it needed an event we had not enabled. The client handler redirected on success without ever telling our server the payment had gone through. A customer would have paid and stayed on the free tier. Activation is now confirmed synchronously against Razorpay before the redirect, writing the same row the webhook writes so the two paths agree.
— Ram
Posted here when there is something worth saying. No newsletter, no drip sequence, nothing to unsubscribe from.
No trial period, no promo tier, no annual discount, no coach-hour packs. The diagnostic is our try-before-you-buy. If it does not tell you something you didn't already know, do not sign up.