I had my paywall backwards.

Somewhere between 190,000 and 220,000 people sit the LET every year. Preparation is dominated by review centers: expensive, location-locked, and run at one pace for everyone in the room.

People pay for months and still walk into the exam not knowing which competencies they are weak in. The bottleneck is not access to material. It is that the cheapest prep available is usually the worst prep available. The addressable market sits at roughly ₱1B to ₱1.6B.

I went in on a civic thesis rather than a market one: better teachers produce better citizens.

My paywall was backwards. Full mock exams were free, and Gabby, the AI coach, was gated. Gabby is a chibi Philippine eagle, named from gabay. Conversion was zero.

There were two obvious moves. Keep everything free and chase a vanity user count. Or paywall the mock exams, which are the thing everyone actually shows up for. I did neither.

I flipped it. Unlimited mock exams are free permanently. What you pay for is the coaching layer, weekly LET simulations, and unlimited practice and review sessions.

The harder part was positioning, not code. I said publicly that a mission-driven product being sustainable is not a betrayal of the mission, and I paired the paywall with a sponsorship program: paying users and partners fund premium access for teachers who cannot afford it. I take no personal income from LETPass.

Daily sales rose 400% in the month that shipped. The diagnosis was pricing and packaging. The product did not change.

The fast version of this product is piping a prompt into an LLM and shipping thousands of questions overnight. I refused.

Instead I built a validation pipeline with layered context grounded in CBRC reference books, and wrote items against Bloom's Taxonomy rubrics using a structured item-writing process. I am adding a human validation layer across every question and recruiting LET takers as validators.

It is much slower, on purpose. A wrong answer key on a licensure exam app is not a bug. It is someone's teaching career.

The most instructive outage was Supabase Disk IO throttling in production. Symptoms first, then the diagnosis, then the fix. Read queries degraded under load in a pattern that looked like application slowness rather than an infrastructure ceiling, which is what made it worth writing down.

Subscription billing runs through PayMongo, which is what turned the pricing decision above into something I could actually measure.

Growth ran zero in early 2026, to 2,100+, to 3,000+ by late July, to 7,000+ in August 2026. Subscription revenue is live and averaged around three sales a day in late July.