01
Break large material into retrievable units
A candidate should not mentally traverse the manual to practise one concept.
In the build: chapter and section level card types
A study layer over the Manual of Patent Examining Procedure, shaped around how one Patent Bar candidate actually reviewed and recalled the material.
My friend Josh was preparing for the Patent Bar, working through the Manual of Patent Examining Procedure. The MPEP is the reference examiners use, and it is enormous and written to be authoritative rather than learnable.
I did not start with "I want to build a flashcard website." I started by asking Josh how he was actually studying, because the obvious framing, that studying is hard because the content is dense, does not tell you what to build.
What he described was a loop rather than a reading task. Find the relevant concept. Understand the rule. Remember it. Come back to it later. Test whether it stuck.
Patent Bar prep asks you to move constantly between dense reference material and fast recall.
Stated that way, the intervention has a job: reduce the effort of getting back to the right information, while keeping enough legal context attached that the material still means something. That is a more useful brief than "make studying easier."
This was a focused intervention built around one person's workflow, and I would rather say that plainly than dress it up as representative research. Josh is a real person with a real exam date, not a persona.
Through contextual conversations I was trying to understand a few specific things: which material he kept returning to, which rules were hard to recall, how he moved between chapters and concepts, how he tested himself, what slowed him down while reviewing, how much surrounding context he needed for an answer to make sense, and what made his existing methods tedious.
The goal was narrow: understand how a candidate moves between lookup and recall. I am not claiming this was a representative study of Patent Bar candidates.
These numbers are in the product, not just the case study. The tool surfaces them as short insight cards with their sources attached, because a candidate deciding how to spend an evening benefits from knowing that retrieval practice measurably beats rereading. It also justifies why the whole tool is a retrieval loop rather than a reader.
The loop has to support four different things: recognition, retrieval, verification, and repetition. Most study tools are good at one and treat the others as navigation problems.
The interesting design question is step four. Once you have compressed a rule into a prompt and an answer, how much of the surrounding legal context has to come with it before the answer stops being a fact and starts being law you could apply?
The intervention is not a replacement for the MPEP. It is a lighter layer over it.
The source stays authoritative. Content is drawn from the Ninth Edition, Revision 01.2024, and the interface only makes repeated review cheaper.
PATENT BAR STUDY TOOL
│
├── MPEP CONTENT chapter · topic · rule / concept
│
├── STUDY PROMPT question / cue · recall attempt · answer / explanation
│
├── SOURCE CONTEXT MPEP chapter · relevant section · surrounding legal context
│
└── STUDY SESSION current item · next item · repeated review
relationship CHAPTER → TOPIC → PROMPT → RULE → SOURCE
A candidate should not have to navigate the whole manual to practise one concept. So the content model breaks down to the level of a single rule, and every prompt keeps a path back up to the chapter and section it came from.
In the tool this shows up as two entry points: study a whole chapter straight away, or build a deck by choosing card types from any chapter. Card types can be a whole chapter or individual sections, and mixing across chapters is the normal case rather than an advanced feature.
Progress and saved decks live in the browser, so nothing depends on an account.
A candidate should not mentally traverse the manual to practise one concept.
In the build: chapter and section level card types
A study answer should not drift away from the MPEP context that gives it authority.
In the build: chapter and revision carried on the content
Moving from one recall attempt to the next should cost nothing.
In the build: tap to flip, got it or missed it, straight to the next card
Dense legal content already carries the cognitive load. The interface should not compete.
In the build: one card in view, timer optional, no decoration
The exam gives 3:36 per question, so pacing is part of the skill. But a candidate testing with extended time is training against the wrong clock if the tool assumes standard conditions.
Onboarding asks whether the candidate will use testing accommodations: standard, time and a half, double time, or another arrangement. That choice recalibrates the card timer and the pace statistics so practice matches real conditions. It is stored on the device and editable at any time.
Josh did not study one chapter at a time. He returned to clusters of related rules that happened to live in different parts of the manual.
The builder lets you drag card types from any chapter into one deck, deduplicates automatically, shuffles by default, and saves the result by name. The chapter remains the source of record without becoming the unit of study.
A card you eventually get right after thirty seconds is not learned. On a timed exam it is a card that will cost you a question. Marking it correct hides that.
The session records time per card and the summary surfaces the ones you lingered longest on, with a direct action to practise those slow cards. The dashboard carries the same entry point, so the tool has an opinion about what to do next.
The moment right after a session is when a candidate knows exactly which rules did not stick, and it is also the moment they are most likely to close the tab.
The session summary offers retrying only the missed cards, or saving them into a practice set for later. Neither requires rebuilding a deck. The loop closes where the information is freshest.
The two gold states are the ones I care about most. Everything before them is table stakes for a flashcard tool; those two are where the tool responds to how the session actually went.
I designed the tool and built it with Claude. It runs as a single page against MPEP Ninth Edition, Revision 01.2024, and keeps progress in the browser.
Rather than mock up screens here, the useful thing is the build itself: onboarding, the dashboard with its insight cards, the deck builder, a study session with the timer on, and the summary with slow cards and missed card retry.
Josh used the tool during his Patent Bar preparation. He later passed the Patent Bar.
Real use also changed my view of the work. A concept can be internally coherent and still be annoying on a Tuesday night at the eleventh hour of studying. Watching the tool sit inside someone's actual routine is the only way I found out which parts were carrying weight.
The accommodation question is the clearest example. It looks like a settings detail. In use it changes whether every pace number the tool shows you is meaningful or misleading, which makes it a correctness issue rather than a preference.
Designing around one person's real workflow gave me a better problem than starting from a generic idea for a study app. The constraint that mattered was never how much study content to add.
It was deciding how much of the original legal context had to stay attached to a much smaller recall interaction. Cut too much and you are memorising sentences. Keep too much and you have rebuilt the manual, which already exists and is free.