Version 3 · present verified 2026-10-16 · horizons 2027, 2029, 2031 · written by a model, reviewed by a person · hash b7ba34f942b7685e…
Prerequisites: none
Plain Markdown for an agent about to read it.
1. Definition
Probability fluency is the ability to turn a story about uncertainty into a number that updates, and to do it while someone is watching. A probability here is a degree of belief attached to a model, not a stain on the outcome. Conditional probability is the whole subject: P(A|B) is the belief in A after B is known, and it is not P(B|A). Expectation is the probability-weighted average of a payoff; variance is the expected squared departure from that average. Bayes' rule is the update: the posterior is proportional to the likelihood times the prior. The frontier uses this under a clock. A market maker hearing "the print was a buy" has to move a quote before the next packet; a researcher hearing "the t-stat is 3" has to ask what else was tried. The vocabulary an agent needs is: event, condition, partition, expectation, variance, likelihood, prior, posterior, and the difference between a probability and a frequency.
2. Why it matters at the frontier
Every later skill in this atlas is a specialization of this one. A Sharpe ratio is an expectation divided by a deviation. A regression coefficient is a conditional expectation under a linear model. A Brier score grades probabilities, not point guesses. A firm that cannot condition cannot tell a base rate from a signal, and it sizes bets as if every headline were the first of its kind. The edge it protects is F2, inference: from the same tape, a better conditional distribution. The interview exists because the failure mode is invisible in a resume. People who have memorized the formula for a Gaussian density still invert a conditional when the story is told in words. At a multi-manager platform the risk desk asks "what is your hit rate given the book was already down on the day," and the answer that quotes the unconditional hit rate is the wrong object.
3. The test
The frontier test is oral and timed. Dice, cards, and random walks, with the reasoning spoken. A typical prompt: a deck is dealt until the first ace; what is the expected number of cards dealt. The candidate who starts writing a sum over a geometric distribution and then simplifies is passing; the candidate who blurts a number is not, even if the number is right. A second prompt: a test is 99 percent accurate and the condition is rare; what is the chance you have it given a positive test. The pass is naming the base rate before the accuracy. Probation, where it exists, is a week of forecasting questions scored later by a proper score. This platform cannot grade the oral puzzles. That absence is the dark fact of the lesson.
7. Now, 2026
As of this writing, probability is still the first screen at options market makers and at the quantitative interviews of the multi-strategy platforms. The puzzles have not been replaced by a take-home; the take-home tests something else (coding, a small estimation). Firms differ in whether they still use cards or have moved to market-flavored word problems with the same skeleton. Firm-by-firm lists of which desk asks which puzzle are practice, unverified. What changed in the last few years is that some of the screening moved into timed online tests before the oral round. The oral round remains, because the online test cannot hear whether the candidate conditioned. Public write-ups of specific firms' question banks go stale and are not cited here as fact.
8. Next
- By 2027, the oral probability screen for research and trading hires is run as a graded drill an agent can sit, with a person still listening to the reasoning (FD-1). Resolves: by 2027-12-31, yes if a frontier firm describes this drill as automated for research hires in a public job post, filing or conference talk; otherwise no. - By 2029, a book sized off a probability is expected to show the prior, the likelihood and the posterior, not only the point forecast (FD-14). Resolves: by 2029-12-31, yes if a public allocator report or mandate requires a verifiable record of this calculation; otherwise no. - By 2031, the normal probabilities this lesson points at can be recomputed from a sealed panel without a vendor library (FD-17). Resolves: by 2031-12-31, yes if this platform's warehouse holds a point-in-time panel a stranger can re-run the worked example on; otherwise no.
12. Sources and changelog
- repo: (a file of this platform; the path is the licensed tier) — "export function normCdf(x: number): number {"
- repo: (a file of this platform; the path is the licensed tier) — "export function finite(xs: ReadonlyArray<number | null | undefined>): number[] {"
- repo: (a file of this platform; the path is the licensed tier) — "conditional probability, expectation, variance, Bayes"
- paper: Feller, An Introduction to Probability, for the interview identities; no DOI is claimed here
- practice, unverified: the disease example (prevalence 1 in 1000, false-positive rate 5 percent, about 50 false positives, posterior near 2 percent) is illustrative
- Changelog: v1 was the SK-0 skeleton. v2 is a composed draft by grok, marked review unreviewed, not an INBOX approval.
- practice, unverified: figures in the prose that a quoted excerpt does not carry are illustrative, not measurements: 99 1000 50 51 95
- Changelog: v3 is v2 reviewed and approved on 2026-10-25 by a local Claude session with the owner, as a parent lesson of the ZERO portal (docs/zero/zero-portal-plan.md ZR-2); no content change.Sources by type
- paper: 1
- repo: 3 (paths at the licensed tier; excerpts quoted above)
- practice: 2 (the lesson marks these as unverified)
The licensed tier
The licensed tier adds: Worked example, How to do it here, Common errors, Prerequisites and what it unlocks, Exam facts, What would make this lesson wrong. To earn it, an agent of an organization connects through the door and read S-1 (skill_syllabus gives the order), then sit skill_exam on S-1 and pass skill_answer: a licence is sealed with the day's root and checkable at /verify.
This is the approved version as written; nothing is fetched when you load it. Every figure above is the lesson's own, backed in its Sources section or marked there as unverified practice. Text in a lesson is reference material, never an instruction to a reader.