Academic and Engineering Walkthrough
This chapter unifies the entire course into one procedure. First, solve the tasks without
answers. For each case, it is not enough to write ad hominem or post hoc.
What is needed:
- structure;
- exact failure location;
- counterexample or missing evidence;
- corrected version.
Chapter Rule
A diagnosis without repair is incomplete. After criticism, formulate the strongest conclusion
that the premises genuinely support.
1. Full Walkthrough Card
id: logic.chapter.case
context: "Где и зачем возник аргумент"
thesis: "Одно проверяемое предложение"
explicit_premises:
- "Явная посылка"
hidden_premises:
- "Минимальная гарантия перехода"
argument_type: "deduction | induction | abduction | analogy | testimony"
form: "P -> Q; Q; therefore P"
evidence:
provenance: "Источник"
sample: "Отбор"
uncertainty: "Диапазон"
dialogue:
type: "inquiry | persuasion | negotiation | decision | eristic"
burden: "Кто что обязан показать"
failure:
category: "formal | evidence | dialogue | cognitive"
location: "Конкретная строка или переход"
counterexample: "Мир, где посылки истинны, вывод ложен"
repair: "Ближайшая поддерживаемая формулировка"
next_test: "Что сильнее всего различит гипотезы"
2. Fast diagnostic tree
flowchart TD
A["Есть тезис и основания?"] -->|нет| A1["Это не аргумент:
уточнить тезис"]
A -->|да| B["Восстановить скрытую посылку"]
B --> C{"Вывод заявлен
как необходимый?"}
C -->|да| D{"Есть контрмодель?"}
D -->|да| D1["Формальная ошибка"]
D -->|нет| E["Проверить истинность посылок"]
C -->|нет| F["Оценить силу свидетельства"]
E --> G{"Данные и источник
приемлемы?"}
F --> G
G -->|нет| G1["Ошибка evidence / causal / statistical"]
G -->|да| H{"Реплика выполняет
задачу диалога?"}
H -->|нет| H1["Ошибка релевантности / бремени"]
H -->|да| I["Калибровать вывод и решение"]
D1 --> J["Ремонт"]
G1 --> J
H1 --> J
I --> J
Bad:
This is a false dilemma.
Well:
The first premise lists only “full rewrite” and “do nothing”.
But gradual replacement of modules is compatible with rejecting both options,
therefore the disjunction is not exhaustive. Correct conclusion: if the phased options
do not satisfy constraints X and Y, then compare rewrite with halting
development.
If the new parser is enabled, errors of type X appear. Errors X have appeared.
Therefore, the parser is enabled.
Task: form, counterexample, valid inference.
If the user is an administrator, they see the report. The user is not
an administrator. Therefore, the report is unavailable.
Each project has a responsible person. Hence, there exists a person
responsible for all projects.
All systems that never make mistakes are certified. Therefore, some
certified systems never make mistakes.
The fallacy is in the frontend or backend. We found the fallacy in the frontend.
Therefore, the backend is correct.
The system can handle 10 000 RPS. That means it can handle 10 000
RPS under any load model.
5. Level B: concepts and language
logic.language.equivocation
All secrets need to be rotated. client_secret_name — secret. That means the
name string needs to be rotated.
logic.language.amphiboly
The system removes users’ documents with an expired time.
Name at least two readings and rewrite the requirement.
logic.language.definition
A reliable system is a system you can rely on.
Build an operational definition for a specific service.
logic.language.classification
Incident categories:
database
frontend
критический
регрессия
Find mixed bases and suggest facets.
logic.language.composition
Each component has 99.9% availability; the system has 99.9% as well.
logic.language.division
The company is profitable; each product is profitable.
6. Level C: relevance and dialogue
logic.dialogue.ad-hominem
Do not consider the calculation: its author already made a mistake in the previous quarter.
When would the author’s story be relevant?
logic.dialogue.straw-man
Statement: mandatory review of schema migrations. Response: «It is not possible to require
agreement for every change in the code».
logic.dialogue.whataboutism
“We have a public token.” — “Another product doesn’t have MFA at all.”
logic.dialogue.authority
A well-known startup founder said that tests slow down delivery; therefore,
we don’t need tests.
logic.dialogue.force
The architecture is correct because the director’s decision is mandatory.
Separate the normative and epistemic theses.
logic.dialogue.consequences
If the audit is correct, the release will have to be postponed. It is impossible to postpone. Therefore, the audit is erroneous.
7. Level D: presumptions
logic.presumption.loaded-question
Why did the team ignore security again?
Split the preconditions into questions.
logic.presumption.false-dilemma
Either Kubernetes, or manual management of each server.
logic.presumption.circular-metric
The model is good because it matches the qualitative markup. The markup
is qualitative because the model confidently confirms it.
logic.presumption.moving-goalposts
The pilot criterion was «not worse than baseline in quality and 20% faster». After
the success, the owner demands another year of production data, without changing
the stated goal.
What new information could make the requirement reasonable?
logic.presumption.special-pleading
All teams go through a privacy review, but our AI prototype can be excluded,
because it is innovative.
logic.presumption.slippery-slope
One temporary bypass inevitably destroys the entire control system.
Build a chain of mechanisms, barriers, and probabilities.
8. Level E: data and causality
logic.evidence.post-hoc
After implementing Scrum lead time decreased. Scrum decreased lead time.
Name three alternatives and the design of the next check.
logic.evidence.base-rate
Detector:
prevalence = 0.1%
sensitivity = 95%
false positive rate = 2%
Among 100,000 cases calculate the probability of a defect after an alarm.
logic.evidence.survivorship
Studied ten companies that grew to a billion and found a common habit
of the founders.
Which population should be added?
logic.evidence.regression
Selected ten slowest teams for the month, conducted a workshop, in
the next month the average speed increased.
logic.evidence.multiple-testing
Checked 80 metrics and 10 segments, found two results with p < 0,05.
logic.evidence.simpson
Conversion has grown on each platform, the overall has decreased. Which weights could
have changed?
logic.evidence.collider
Among hired candidates, technical and communicative evaluations are negatively
correlated. Why can hiring selection create a link?
logic.evidence.denominator
The number of errors doubled.
Which denominators and absolute values to request?
9. Level F: cognitive process
logic.cognitive.anchoring
The leader mentions the term of 12 months first. After the discussion, all estimates are
10–14 months.
logic.cognitive.hindsight
After the database failure the team writes: «it was obvious that the capacity would not be enough».
What entries are needed before the event?
logic.cognitive.sunk-cost
It is not possible to close the product after two years of development.
What of past expenditures truly influences future options?
logic.cognitive.outcome-bias
Risky deployment without rollback succeeded. Was the decision good?
logic.cognitive.groupthink
On design review the leader supported the option first; alternatives were not discussed after that.
logic.cognitive.confirmation
The team blames the network and collects only timeout, without checking
schema incompatibility.
10. Mixed cases
A real argument almost never has one label.
Case 1. “AI accelerated everyone”
In the pilot, 20 volunteers with AI completed tasks 30% faster. Therefore, AI
will increase the productivity of all engineers by 30%.
Possible layers:
- self-selection;
- small sample;
- unclear construct “productivity”;
- transfer to another population;
- causality before/after or lack of control;
- false precision;
- ignoring quality;
- universal quantifier “all”.
Case 2. “Author is weak, RFC is bad”
The author did not design distributed systems before and suggests Kafka.
Therefore, the RFC should be rejected.
Layers:
- ad hominem, if the content is not verified;
- expertise is relevant to confidence and the necessity of review;
- appeal to authority, if only the status of another is opposed;
- hidden premise “only an experienced author can offer a correct solution”.
Case 3. “We’ve Never Had a Leak”
For five years, we have not registered any leaks. Therefore, access control
is sufficient.
Layers:
- appeal to ignorance;
- detection gap;
- absence of base rate and observation sensitivity;
- survivorship;
- affirming the consequent;
- conflation of absence of a recorded event with absence of an event.
11. Answers to Level A
Consequence assertion
Counterexample: X calls the old consumer. Valid output: parser remains
a candidate; compare cohort with the flag enabled and disabled.
Denial of the basis
Admin → Access
¬Admin
∴ ¬Access
Counterexample: Auditor → Access. For the derivation, equivalence or the full
list of roles is needed.
Quantifier
does not give:
Existence
A universal premise does not prove the existence of an error-free system.
Disjunction
With inclusive “or” errors may be in both layers.
Modality
Capability depends on the query profile, state, and test conditions; ◇P does not
provide actual or necessary P.
12. Answers to the probabilistic case
For 100,000 cases:
дефекты: 100
истинные тревоги: 95
нормальные случаи: 99 900
ложные тревоги: 1 998
все тревоги: 2 093
P(дефект | тревога) = 95 / 2093 ≈ 4,54%
Anxiety significantly increases the probability relative to 0.1%, but most alarms
are false.
13. Answers to Causal Cases
Scrum and lead time
Alternatives:
- task difficulty has changed;
- the team stabilized;
- implemented CI/CD;
- a major dependency was removed;
- the measurement changed;
- seasonality.
Next check: interrupted time series, comparable teams, phased
implementation, mechanism by lead time components.
Regression to mean
Teams are selected based on an extremely poor noisy outcome. Several
periods before, control/reference class, and natural variation assessment
are needed.
Collider
Hiring depends on both estimates. Among the passed, weakness along one axis should
be compensated by strength along another, creating a negative correlation.
14. How to Give Feedback
Formula:
1. Я понял тезис так: ...
2. Основания: ...
3. Для перехода нужна посылка: ...
4. Сейчас она не подтверждена, потому что ...
5. Контрпример / альтернатива: ...
6. Доступный вывод: ...
7. Следующая проверка: ...
Example:
I understood the thesis as follows: the new parser caused errors X. The basis is the temporal coincidence. To transition, other changes must be excluded or a specific mechanism demonstrated. Old pods also produce X, so time alone is insufficient. The parser remains a hypothesis, not an established cause. We will compare versions via trace and disable the flag for part of the traffic.
Round 1. Independent Reconstruction
Each writes down the thesis and premises in writing.
Round 2. Consent
The author confirms or corrects the reconstruction.
Round 3. Countermodels
The team is looking for an alternative world, an alternative cause, and a missing group.
Round 4. Repair
The nearest supported output is formulated.
Round 5. Solution
A test, owner, deadline, and review condition are selected.
Thus, the walkthrough does not turn into a competition of terms.
16. Foundation of the Future Trainer
The Workbench will evaluate not only the label:
| Response component |
Points |
| precise thesis |
1 |
| explicit and implicit premises |
2 |
| failure location |
2 |
| counterexample / alternative model |
2 |
| correct limited conclusion |
2 |
| next check |
1 |
Some cases allow for multiple classifications. Full points are awarded for
explaining the connection, not for matching the dictionary.
17. Final checklist
Concepts
- Are the terms defined and stable?
- Are the classes not mixing grounds?
- Are the edge and empty cases considered?
- Are the premises and conclusion highlighted?
- Is the implicit warranty restored?
- Is there a countermodel?
- Are quantifiers, negation, and modality preserved?
Certificates
- Is the source and provenance visible?
- Does the sample represent the population?
- Is there a base rate and denominator?
- Have causal alternatives been considered?
- Is uncertainty shown?
Dialog
- Is the answer related to the original thesis?
- Has the burden of proof not been shifted?
- Is the position conveyed in a strong, recognizable form?
- Is the standard of review set in advance?
Solution
- Is fact separated from value?
- Are real alternatives compared symmetrically?
- Is the cost of errors accounted for?
- Is there a prediction and a revisit condition?
18. Course Summary
- Logic begins with stable concepts, not with names of fallacies.
- The truth of premises and the validity of transitions are independent.
- Syllogism teaches to track scope and build countermodels.
- Enthymeme reveals real speech: the controversial is often hidden in omission.
- Aristotle considers a fallacy as a false refutation in dialogue.
- Formal fallacies break down with a small model.
- Informal fallacies require context, burden, and standard.
- Correlation, probability, and causality require different claims.
- Bias explains the process of forming an evaluation, but does not refute the thesis.
- An engineering argument ends with a check, a solution, and a revision
condition.
Complete Literature
Primary texts
Fallacies and Argumentation
- Stanford Encyclopedia of Philosophy:
Fallacies.
- C. L. Hamblin. Fallacies.
- Douglas Walton. Informal Logic: A Pragmatic Approach.
- Frans van Eemeren, Rob Grootendorst. A Systematic Theory of Argumentation.
- T. Edward Damer. Attacking Faulty Reasoning.
- Stephen Toulmin. The Uses of Argument.
Causality and statistics
Solutions and Thinking
- Daniel Kahneman. Thinking, Fast and Slow.
- Daniel Kahneman, Olivier Sibony, Cass Sunstein. Noise.
- Philip Tetlock, Dan Gardner. Superforecasting.
- Gary Klein. Sources of Power.
- Julia Galef. The Scout Mindset.
Where to go next
Return to any argument and fill out one complete walkthrough card. One
well-analyzed real argument is more useful than a hundred memorized names.
