Every prep system answers one question: what should this student do next. Most answer it with the syllabus order, or with a menu — which hands the decision back to the student, who does not have the information to make it. Rhovecs derives it. The derivation is the product, so this page documents it end to end.
What follows is the system as it currently runs: the evidence it uses, the order the checks fire in, and the parts it deliberately does not do.
The evidence
Every problem on the main practice track is a real previous-year JEE question. Practice therefore happens at the exam's own difficulty, on the exam's own items, which removes the most common source of inflated confidence — a practice set that is easier than the paper.
Each attempt updates one number per concept: mastery. Mastery is not percentage correct. It is computed across the concept's structural patterns — the distinct question forms the exam actually uses for that concept — and each pattern has a satiation point past which more of the same form earns nothing. Clearing every pattern is the only route to a high score. Grinding one easy pattern is not.
Four rules govern what a single attempt is worth.
- Evidence decays. Each new attempt is added to a running total that first shrinks the old total by ten percent, so recent, sustained correctness outweighs a good week from two months ago.
- Harder questions count for more. Credit for a correct answer is weighted by the question's intrinsic difficulty, on a 1-to-5 scale set by the paper it came from.
- Hints halve the credit. A correct answer after a hint is worth half of an unassisted one.
- Viewing the solution scores nothing. That attempt is excluded from mastery entirely.
A wrong answer subtracts, and the size of the subtraction depends on the kind of mistake — covered further down.
Explanation always arrives after the evidence is recorded, never before. The stepped solution walkthrough, and the per-step "I don't follow this" explanation inside it, are post-attempt surfaces by construction. You cannot consult them and then be measured on the same attempt.
Step 1 — what you are allowed to practise
Concepts carry explicit prerequisite edges, and the system works on the transitive closure of those edges, not just the direct parent. A concept is executable only when every concept in its full prerequisite closure sits at or above the mastery gate of 0.65.
If it does not, the concept is locked, the specific blocking concepts are named, and the system points at the deepest one you can actually start today. None of this is advisory. A locked concept cannot be practised until the foundation underneath it clears.
Step 2 — which mode you are in
Mode is resolved per subject, across the concepts you have marked ready. Four states, checked in this order:
- Structural learning — some ready concept is below 0.65. Acquisition is the job.
- Strengthen — everything is at 0.65 or above, but something is still below 0.80.
- Review — everything is at 0.80 or above, and at least one concept is due for retrieval.
- Maintain — everything is at 0.80 or above and nothing is due.
You never choose the mode; the mastery distribution determines it. Adding a new chapter drops the subject back to structural learning by construction, because a new concept enters at zero.
Step 3 — which concept
Each mode has its own selector, and each returns exactly one concept.
Structural learning takes the next concept in the learning-path order. That order is banded by exam weight first, then ranked by a priority score weighing exam yield most heavily (0.35), how many later concepts it unlocks (0.25), measured pressure and impact (0.15 each), and graph depth (0.10). A foundation that unlocks a lot of the syllabus comes before a high-weight dead end.
Strengthen takes the concept with the largest gap to 0.80, breaking ties toward the one with the thinnest evidence behind it.
Review takes the concept whose predicted recall has fallen furthest below the retention target, weighted by exam weight, with a fairness term so a long-overdue concept is not starved by a heavier one.
Maintain takes the concept you have the least evidence on.
Step 4 — which problems
Inside the chosen concept, problems are ordered by what they can prove, not by chapter order.
Patterns you have not yet cleared come first, interleaved round-robin, so a session spreads across the concept's question forms instead of drilling one of them. Patterns already at satiation are filler and trail, because attempts on them cannot move mastery. Within a pattern, problems ascend in difficulty and the longest-untouched problem comes first.
Two holdouts protect the signal: a problem you attempted in the last 24 hours is kept out of the pool, and one you solved cleanly in the last 7 days is not re-served. What reaches you is a question you have not seen recently, in a form you have not yet proved.
Retention runs alongside, not afterwards
Revision is not a separate mode you have to remember to enter. Every session reserves a floor of problems for the learning concept and fans the remainder across concepts that are due, in even chunks with a minimum size. A due concept that cannot be given a meaningful chunk is skipped rather than tokenised into a single question.
Due dates come from a power-law retention model. For a concept with stability , measured in days, predicted recall days after the last successful retrieval is
Stability grows in proportion to . A success after a long gap — when recall had genuinely decayed — buys a large increase in durability. A success minutes after the last one buys almost nothing. Cramming cannot inflate the schedule; the model is shaped so that it can't.
One asymmetry is deliberate. A concept only enters review after it has already crossed 0.65, so a correct review is recall of a skill you already hold, and it adds no mastery. A failed review applies the full penalty — and if that drags the concept back below 0.65, it returns to the learning queue and is taught again. Mastery is built by learning, and only confirmed or decayed by review.
What a wrong answer changes
Not every wrong answer means the same thing. Three cases are separated:
- Knowledge gap — the concept itself is missing or wrong.
- Execution error — the setup was right and the algebra, arithmetic, or procedure broke.
- Behavioural — carelessness, misreading, or time pressure.
Two independent mechanisms do the classification, and both run for everyone. When the option you picked is a known trap tied to a specific misconception, that mapping records the exact error. Separately, your record on structurally similar problems is checked: a strong record across that neighbourhood points at carelessness, a weak one at a real gap.
On the Ultra plan there is a third input. Working the problem by hand with a stylus, the written steps are checked one by one, so the break can be located in the derivation rather than inferred from the final answer.
The classification is not cosmetic, and it drives two different things.
It sets the retention penalty: a knowledge error cuts the concept's stability more deeply than an execution slip does, so the two mistakes produce genuinely different revision schedules.
It also selects the repair set. The failure mode resolves to a repair strategy — formula recall, algebraic execution, problem setup, or a weak structural pattern — and three to five problems are assembled to exercise that specific strategy. Not more of the chapter. The technique that broke.
The second track: speed
Being right is necessary and not sufficient. A three-hour paper is a time budget, and a method you can execute correctly but slowly costs marks exactly like a method you don't know.
Speed is therefore measured on its own track, against generated variants of a pattern rather than previous-year questions, because measuring pace needs more repetitions of one form than the finite past-paper bank can supply. That track is deliberately walled off from the mastery track in both directions:
- It unlocks only after competence is proven — the concept must be at 0.80 or above and the pattern already cleared on the mastery track. You are never asked to go fast at something you cannot yet do.
- A fluency score combines accuracy and pace — 0.6 accuracy, 0.4 speed — against a target time set per difficulty band, over a rolling window of recent repetitions, and needs a minimum number of observations before it reports at all.
- The drill queue is ordered by what the gap costs: how often the pattern appears in past papers, multiplied by how far off pace you are.
- Timed drills never write mastery and never touch the review schedule. Racing through variants cannot manufacture mastery, because the two tracks do not share a ledger.
Where the exam itself comes in
Full-length mock papers, high-yield sets, and tests you assemble yourself are all available, and they are the one place the student sets the scope rather than the system. That is intentional: this is evaluation, not the learning path, and evaluating yourself against a scope you chose is a legitimate thing to want.
Completed full-length mocks feed back in one specific way. They anchor the rank projection: each finished full mock is treated as a direct observation of your score, weighted by how recent it is, and the projection is corrected toward the observed result in proportion to how much evidence there is. The correction is symmetric — it raises the projection as readily as it lowers it — and one unusually good or bad paper cannot dominate, because all of them count and stale ones fade out on their own. High-yield and chapter papers never anchor, because they are not built to the exam's blueprint and would bias the estimate.
What this method does not do
Published method is only worth publishing if it is also honest about its edges.
Difficulty is never calibrated against other students. A question's difficulty is a property of the question by the exam's standard. Calibrating it against a cohort would make mastery circular — you would be measured against whoever else happened to be practising — and would let a strong batch deflate the scale for everyone.
The rank projection does not choose your practice. It reads the same signals and reports where your marks are leaking, but it is a readout. Concept selection runs on mastery, prerequisites and retention exactly as described above, and not on the projection.
Speed is measured but not yet charged. The time-risk a fluency gap represents is computed and shown to you per pattern, and it does not currently move the projected score. The shape is settled; the magnitude is not calibrated, and until it is, a number that is only approximately right should not be silently changing a rank on your screen.
Nothing here has been validated against a real JEE result. Rhovecs is new; no student has yet sat the exam with this system behind them. The mock anchor above is the system checking itself against the closest thing to ground truth that exists today — a full, timed, blueprint-shaped paper. Actual exam outcomes are the real test, and they are still ahead of us. Until they arrive, the models are a principled prior grounded in established learning science, specified precisely — which is why they are written down here rather than described as proven.
The claim of this page is not that the method is validated. It is that the method is stated, and stated precisely enough to be argued with.