Layer 1 — Curriculum phase
The OFFERGOBLIN curriculum is a tree: phase, then topic, then section. Every candidate has an active phase based on their progress.
- The active phase has weight 100x.
- Future phases that have not yet been unlocked have weight 5x.
- Already-mastered review material has weight 1x.
This means roughly 95 percent of any practice session draws from the candidate's current learning frontier, with light bleed-over into future material and rare review of mastered topics to prevent decay.
Layer 2 — Question health
Every question the candidate has seen has a state. GOBLINMODE weights within whatever tag the engine is pulling from:
- Unseen — never shown to this candidate — weight 50x
- Don't Know — last marked wrong — weight 20x
- No-response — shown but skipped without grading — weight 5x
- Know — marked as mastered — weight 1x
The two layers multiply. A candidate practicing in GOBLINMODE primarily sees: brand new questions in the current phase, then questions in the current phase that they got wrong before, then bleed-over from adjacent future phases, with occasional review of mastered material.
What the candidate experiences
From the candidate's seat, GOBLINMODE feels like a deck that knows what to give them next. The first time they open the app, they see fundamentals from the active phase. As they grade themselves, the deck adapts: the questions they got wrong come back, and the ones they got right fade out.
A candidate cannot game the engine by always marking Know. The unseen pool inside the current phase still has weight 50x, so new questions keep arriving until that phase is exhausted. And the engine tracks what the candidate marked, not how confident they were — there is no payoff to lying.
Why phase weighting exists
A pure spaced-repetition engine that treats every question equally would mostly serve random questions across the whole curriculum. That is wrong for banking interviews because most candidates have a real, identifiable level: a sophomore should not be fighting LBO modeling questions before they have accounting cold.
The phase layer enforces a learning order. Within that order, the health layer picks the next card.
How GOBLINMODE compares to GOBLIN100
GOBLIN100 is a fixed-sequence ordered ramp of 100 questions for candidates who do not yet know what to study first. There is no adaptation in GOBLIN100 — every candidate sees the same 100 questions in the same order.
GOBLINMODE is the opposite: pure adaptation against the candidate's history. Most candidates run GOBLIN100 first to build a baseline, then graduate to GOBLINMODE for indefinite reps. The engine's 20x Don't Know boost naturally resurfaces anything they got wrong during the ramp, so the ramp questions stay alive in the active rotation.
When candidates use Bank & Round instead
GOBLINMODE is the daily mode. Bank & Round is the situation mode — used when a known interview is on the calendar and the candidate wants to drill the specific firm and round.
The two are designed to be used together. Most accelerated users open GOBLINMODE every day and switch into Bank & Round for the week leading up to a Superday.
What GOBLINMODE is not
GOBLINMODE is not an AI feature. It is a deterministic weighted selection algorithm. The same inputs (curriculum phase, question health, tag pool) always produce the same probability distribution. There is no model, no embeddings, no inference. It is a database query with weights.
This matters because candidates can trust the engine to do exactly what it says: surface unseen and weak questions in the current phase. There are no surprises in production.
Pricing
GOBLINMODE is available on both the Free tier (7 questions per day) and the Accelerated tier ($40 a month, 80 questions per day). The engine is identical across tiers. The only difference is the daily volume cap. See Free vs Accelerated for the side-by-side.
