How the map is built, and where it stops.
We predict the shape of the loop and rank what it tests. We don't pretend to know the questions.
Know, infer, predict, prepare, cannot know
- What we know
- Two things: the job description you paste, and the recruiter notes you supply. An explicit, unambiguous round statement in those notes is treated as confirmed. We do not independently authenticate the notes.
- What we infer
- Round structure that follows from the role, the level, and the way the job is written. Inference is labelled as inference, every time, with the reason attached.
- What we predict
- The likely shape of the loop and what each round tests, ranked by confidence. A prediction is a ranking, not a promise.
- What you prepare
- One ordered list: study now, review, skip. The point of ranking is to let you skip things.
- What we cannot know
- The questions. The interviewer. The internal rubric. Anything confidential. If a product claims to have the real questions, it is either guessing or it should not have them.
Confidence states
Every round carries a state. There are no percentages, because a percentage here would be invented precision. State is carried by shape and by words, not by colour alone.
Recruiter notes
If the notes you provide explicitly name a round, that round is CONFIRMED. Ambiguous notes do not confirm a format. Conflicting statements remain visible and should prompt a question to the recruiter rather than an arbitrary choice by the map.
Sources and recency
Perplexity searches public sources once when company and role title are provided. We use relevant search excerpts, not a claim to have read every full page. Source links and available dates appear on your saved map. External reports remain inferred; they never override explicit recruiter notes. Very likely requires at least two relevant, independent, dated public reports from the last 12 months that agree about that specific round. If research is unavailable, empty, or irrelevant, the map uses the job description and recruiter notes only and says so. Private notes and background are not sent to search. We never invent dates or infer company ownership from a page title. Job links still require pasted job text; search snippets are not reliable full job descriptions. Exact citations, dates and publisher counts are deterministic safeguards. They cannot guarantee that the model correctly judged the reports as semantically agreeing.
Illustrative inference examples
These examples show the reasoning rules. They are not candidate results or evidence that a prediction was correct.
- Evaluation responsibility
- A job description that emphasizes evaluation makes evaluation a preparation priority. It does not establish a standalone evaluation round.
- Ambiguous recruiter wording
- “Technical conversation” does not confirm live coding, system design, or a take-home. The format stays unknown until clarified.
- Missing background evidence
- A résumé that does not describe evaluation work is not proof the candidate lacks that skill. The map should ask for an example or recommend preparing one.
Validation status
We have not published candidate-level prediction validation or an accuracy percentage. The examples and internal synthetic fixtures illustrate reasoning and test rules; they do not establish candidate or hiring outcomes.
Limits
- A thin job description produces a thin map. We would rather say unknown than fill the gap.
- A map is generated using the information available at that time. Interview processes can change, and inferred rounds may be wrong.
- The map is about structure and priorities, not answers.
- We do not and will not sit in your live interview.