# Read geometric waiting in Book Isle: first success on trial three differs from any success by three

Language: en
Canonical: [Read geometric waiting in Book Isle: first success on trial three differs from any success by three](<https://allluckly.cn/en/guides/nasbook-geometric-first-success-trial-failures-memoryless-controls/>)
中文：[中文 HTML](<https://allluckly.cn/guides/nasbook-geometric-first-success-trial-failures-memoryless-controls/>)
English: [English HTML](<https://allluckly.cn/en/guides/nasbook-geometric-first-success-trial-failures-memoryless-controls/>)
Published: 2026-11-19T00:00:00+08:00
Modified: 2026-11-19T10:18:47+08:00
Author: Bison

Check first-success positions, cumulative tails and conditional waiting at success probability one-quarter, distinguishing trial and failure counts.

English translation of the [Chinese original](<https://allluckly.cn/guides/nasbook-geometric-first-success-trial-failures-memoryless-controls/>), translated on 2026-11-19. Original publication dates, attribution, code and version notes are retained.

Counting successes in three fixed trials differs from stopping at the first success. Read fictional independent-trial cards and waiting/tail probabilities in Book Isle, without turning mathematical success into a real guarantee. Calculations are external, with no actual card draws or behavioural predictions.

## Declare stopping and the variable’s starting value

Trials have only S, F, are independent and identically distributed, with P(S)=1/4 and P(F)=3/4, stopping at first S. T counts trials including that success, so T=1, 2, 3,…, never zero. Put rules/patterns on P1, probabilities on P2 and conditional/nonreplacement controls on P3; export PDF and retain source work. Check T and bookmark P2. No preset cap does not mean infinitely many trials were executed, and page-card count is not an extra stopping limit.

## First success on trial three permits only FFS

T=3 requires FFS, with probability (3/4)²(1/4)=9/64. SFF stops at one; FSF stops at two, so neither belongs. Generally P(T=n)=(3/4)^(n−1)(1/4). Do not multiply by three as in a fixed-three-trials exactly-one-success question: the first-success location already fixes the order. The incorrect 27/64 combines different stopping positions.

Here p is per-trial success probability, not a probability that accumulates after failure. After two failures the next trial still succeeds with probability 1/4 in the independent constant-p model, not 1/2 because success is “due.” Separate a one-trial conditional probability from cumulative success-by-time probability, which increases without changing per-trial p.

## Use the complement for success by trial three

T≤3 includes S, FS, FFS with probabilities 1/4, 3/16, 9/64, totaling 37/64. The complement FFF has probability 27/64, giving the same 1−27/64. T\>3 means still no success, not success on three. Keep point, cumulative, tail columns and check 37+27=64, not 9+37. Complementary events, not matching denominators, justify a sum of one.

```text
independent repeated trials; P(S)=1/4; P(F)=3/4
T=number of trial containing first success, T=1,2,...
P(T=3)=P(FFS)=9/64
P(T<=3)=37/64; P(T>3)=27/64
P(T>5 | T>2)=(3/4)^3=27/64
E[T]=4; failures Y=T-1 have E[Y]=3
P(T<=4)=175/256, not1
```

## Failure count differs from T by one

Let Y=T−1 count failures before first success, allowing zero. Y=2 and T=3 are the same event, probability 9/64. Y=0 means success on trial one, probability 1/4, not success without a trial. P(Y=k)=(3/4)^k/4 for nonnegative integer k differs by one exponent because it counts another object. Read whether a “geometric” variable counts trials through success or failures before success before copying its formula. Different textbook conventions are not contradictory probability theorems.

## Conditional waiting counts three further failures

Given T\>2, asking T\>5 requires failures on trials three through five. Independence and fixed probability give (3/4)³=27/64, also obtained from P(T\>5)/P(T\>2). Previous failures do not make success due. Distinguish total five trials from three additional trials and a tail event from first success exactly on five; conditioning removes the already-known two failures.

Check tail inequalities: T\>3 means three initial failures and has probability 27/64, while T≥3 requires only two initial failures and has probability 9/16. For the failure count, Y≥3 corresponds to T\>3. Derive events before exponents instead of always using q³ whenever a three appears.

## Expected four trials is not a four-trial guarantee

The tail sum E\[T\]=Σ from n=0 of P(T\>n)=Σ(3/4)^n=4 gives E\[Y\]=3. Expectation is a model average, not a deadline. P(T≤4)=1−(3/4)^4=175/256\<1, so five or more remains possible; do not force S on card four. Three displayed sample cards do not make the mean three. The convergent formula needs a success probability with 0\<p≤1; p=0 never succeeds and is outside this finite-mean conclusion.

## A nonreplacement control changes the assumptions

A uniformly shuffled four-card set with one S and three F, read without replacement until S, has T=1,…, 4 with probability 1/4 each. In particular P(T=3)=(3/4)(2/3)(1/2)=1/4, not 9/64. Failure changes the remaining proportions, so iid assumptions fail. This is a paper control, not observed shuffling. Diagnose stopping, replacement and variable convention before exponents/complements, not alleged App simulation or display randomness.

In the without-replacement control P(T=1)=1/4, but after an initial F the next success probability becomes 1/3. This locates the failed constant-p/independence premise, beyond merely comparing 9/64 with 1/4. Four cards also make T\>4 impossible, unlike a geometric variable with arbitrarily large positive-integer support.

## Reopen from patterns to events and limits

At reopened P2 recover FFS=9/64, success by three=37/64 and still no success=27/64, then explain why two failures create no compensation. Check Y and nonreplacement controls and distinguish mean four from a deadline. Deliver rules, support, patterns, point/cumulative/tail probabilities and conditions, marking unperformed bookmarks, highlighting and reopening pending. The model supplies no real-attempt prediction, App independence test, success improvement or completed random experiment.

## Related tutorials and sources

[Read binomial probability in Book Isle: check repeated independence before exactly two successes](<https://allluckly.cn/en/guides/nasbook-binomial-four-trials-exact-two-conditions/>); [Read expected values in Book Isle: 1.75 need not be a possible single score](<https://allluckly.cn/en/guides/nasbook-discrete-expectation-weighted-value-not-single-outcome/>).

Method references: [OpenStax geometric distribution](<https://openstax.org/books/introductory-statistics-2e/pages/4-4-geometric-distribution>). Diagrams, report data, counterexamples and ledgers are original exercises.

Version checked on 19 November 2026: the [Apple public lookup](<https://itunes.apple.com/lookup?id=6766198898&country=cn&entity=software&timestamp=20261119>) lists Book Isle 1.1.4, requiring iOS / iPadOS 15.6 or later; see the [official App Store](<https://apps.apple.com/cn/app/id6766198898>) for published capabilities. Use self-made PDFs, EPUBs or clear text materials; bookmarks, highlights and notes depend on format and visible controls, without automatic solving or literary analysis. External preparation and paper predictions do not constitute device observations. Leave unperformed checks blank.

