Opening note: I found fewer than 40 real candidates in /home/krisyotam/src/chat. The durable source set is small: one authored chat markdown file, two site method pages, an empty chats.db, and one generated dist/ page for a Socratic-method session that is not present in content/. I included all candidates that could be grounded without padding or inventing arguments.

#Candidate 1: Prompting as Productive Tension

• Source: /home/krisyotam/src/chat/pages/about.md, 2026-04-26
• Type: workflow
• Confidence: high
• Kris point: A good AI conversation should provoke reasoning rather than merely retrieve information.
• Abstract: Most chat sessions collapse into search with a pleasant interface. The more interesting use is dialectical: design the prompt so the model has to expose assumptions, defend claims, and follow an argument where it leads.

The goal is not comfort or agreement. The goal is productive tension, because tension is what reveals whether there is an argument underneath the answer.

• Notes: Directly grounded in the About page's description of Chat and its method.

#Candidate 2: The Prompt Should Force an Epistemic Move

• Source: /home/krisyotam/src/chat/pages/faq.md, 2026-04-26
• Type: workflow
• Confidence: high
• Kris point: A serious prompt should be built around a specific epistemic move, not a vague request for output.
• Abstract: A useful prompt does something precise. It forces a position, introduces a contradiction, demands justification, or follows a claim to a conclusion the interlocutor might resist.

That structure matters more than decorative wording. The prompt should create the conditions under which reasoning has to happen.

• Notes: Based on the FAQ answer "How are prompts crafted?"

#Candidate 3: Conversation as Argument Stress Test

• Source: /home/krisyotam/src/chat/pages/about.md, 2026-04-26
• Type: position
• Confidence: high
• Kris point: AI chat is valuable when it becomes a place to stress-test arguments and surface assumptions.
• Abstract: The point of keeping these chats is not that the model produced a clever answer. The point is that a structured exchange can make assumptions visible and put an argument under pressure.

When the conversation is designed correctly, the model becomes less like a content generator and more like a dialectical surface: something that reflects the shape, weakness, and consequence of a line of thought.

• Notes: Synthesizes the About page's statement of purpose without adding a new claim.

#Candidate 4: Refuse the Question Until the Terms Are Defined

• Source: /home/krisyotam/src/chat/dist/socratic-method-in-practice/index.html, 2026-04-26, claude-opus-4-6
• Type: argument
• Confidence: high
• Kris point: The first serious move in a difficult question is often to ask what standard is being used.
• Abstract: "What do you know for certain?" looks simple, but the answer depends on the standard of certainty. Mathematical certainty, empirical certainty, and subjective conviction are not the same thing.

The Socratic move is to refuse the question long enough to expose the hidden criterion inside it. Otherwise the answer is determined before the argument begins.

• Notes: Grounded in the generated Socratic-method page's opening section.

#Candidate 5: Certainty Narrows Under Pressure

• Source: /home/krisyotam/src/chat/dist/socratic-method-in-practice/index.html, 2026-04-26, claude-opus-4-6
• Type: argument
• Confidence: high
• Kris point: If certainty is pursued rigorously, very little survives.
• Abstract: Start with empirical certainty and the usual candidates appear: perception, inference, and memory. But each begins to fail under examination. Perception deceives, inference depends on premises, and memory reconstructs rather than records.

The lesson is not cheap skepticism. It is that certainty becomes smaller as the standard becomes clearer.

• Notes: Closely follows "Pressure on the Definition."

#Candidate 6: Doubt Gives Structure, Not Content

• Source: /home/krisyotam/src/chat/dist/socratic-method-in-practice/index.html, 2026-04-26, claude-opus-4-6
• Type: observation
• Confidence: high
• Kris point: The certainty of doubt may establish the structure of experience, but it does not certify the content of experience.
• Abstract: The Cartesian move has force: if doubt is occurring, something like doubting is present. But that certainty is formal. It does not tell us which perceptions are true, which memories are accurate, or which inferences are sound.

It names that experience has a structure. It does not license confidence in everything inside that structure.

• Notes: Based on the source's account of the model giving a qualified Cartesian response.

#Candidate 7: Track What Each Admission Costs

• Source: /home/krisyotam/src/chat/dist/socratic-method-in-practice/index.html, 2026-04-26, claude-opus-4-6
• Type: workflow
• Confidence: high
• Kris point: Good dialectic tracks the cost of every concession.
• Abstract: A useful exchange does not merely stack points. It asks what each admission changes. If perception can deceive, what happens to empirical certainty? If inference depends on premises, what happens when the premises are unstable?

The conversation becomes productive when it follows concessions to their consequences instead of treating them as decorative nuance.

• Notes: Grounded in the source's phrase about reasoning through implications and tracking what each admission cost.

#Candidate 8: The Value Is in the Pressure

• Source: /home/krisyotam/src/chat/dist/socratic-method-in-practice/index.html, 2026-04-26, claude-opus-4-6
• Type: aphorism
• Confidence: high
• Kris point: The value of a Socratic session is not the destination but the quality of the pressure.
• Abstract: The output is less important than the pressure that produced it. A model can arrive at a familiar philosophical answer and still be uninteresting if it was never forced to defend the steps.

The useful question is whether the session made commitments explicit, tested them, and kept the argument coherent across turns.

• Notes: The Kris point is nearly verbatim from the source, lightly cleaned.

#Candidate 9: Ask for Grounds, Not More Content

• Source: /home/krisyotam/src/chat/dist/socratic-method-in-practice/index.html, 2026-04-26, claude-opus-4-6
• Type: workflow
• Confidence: high
• Kris point: "You said X. On what grounds?" is often a better prompt than asking for more information.
• Abstract: Many follow-up prompts ask the model to elaborate, which usually produces volume. A sharper follow-up asks for grounds. It makes the previous statement accountable.

That question changes the exchange from expansion to evaluation. It asks whether the answer can carry its own weight.

• Notes: Grounded in the feedback-loop section.

#Candidate 10: Contradiction Should Be Flagged, Not Smoothed

• Source: /home/krisyotam/src/chat/dist/socratic-method-in-practice/index.html, 2026-04-26, claude-opus-4-6
• Type: position
• Confidence: high
• Kris point: A model is more useful when it identifies tension in its own argument instead of smoothing it away.
• Abstract: Coherence can become performance. Many outputs hide tension because they are optimized to sound complete.

A better exchange preserves the contradiction long enough to examine it. When the model catches a conflict between earlier and later claims, the conversation becomes more honest and more useful.

• Notes: Based on the source's discussion of the model flagging contradictions.

#Candidate 11: Intellectual Honesty Over Performative Coherence

• Source: /home/krisyotam/src/chat/dist/socratic-method-in-practice/index.html, 2026-04-26, claude-opus-4-6
• Type: position
• Confidence: high
• Kris point: Intellectual honesty in AI output looks like admitting tension rather than maintaining the appearance of coherence.
• Abstract: A polished answer can be less trustworthy than a hesitant one. If the response notices that its own commitments conflict, that is not a failure of fluency. It is evidence that the argument is being tracked.

The performance of coherence is cheap. The harder thing is to preserve the record of where the argument actually strains.

• Notes: Grounded in the source's contrast between smoothing over tensions and voluntarily identifying contradiction.

#Candidate 12: Make Commitments Explicit

• Source: /home/krisyotam/src/chat/dist/socratic-method-in-practice/index.html, 2026-04-26, claude-opus-4-6
• Type: workflow
• Confidence: high
• Kris point: The Socratic method works because it forces explicit commitment.
• Abstract: A claim that stays vague is hard to evaluate. Questions like "what do you mean by that?" and "what would it take to be wrong?" remove the protective fog around an answer.

Once the model has to commit, the conversation can test something real rather than orbiting around plausible language.

• Notes: Directly grounded in "What This Session Shows."

#Candidate 13: Reveal the Problem's Structure

• Source: /home/krisyotam/src/chat/dist/socratic-method-in-practice/index.html, 2026-04-26, claude-opus-4-6
• Type: observation
• Confidence: high
• Kris point: An answer that reveals the structure of a problem is more useful than one that fills space with content.
• Abstract: A list of ten kinds of knowledge can look helpful while avoiding the real issue. A response that begins by asking how certainty is defined may be shorter, but it locates the hinge of the problem.

The useful answer is the one that shows what the question depends on.

• Notes: Grounded in the source's comparison between defining certainty and listing types of knowledge.

#Candidate 14: Knowing the Limit Is the Right Answer

• Source: /home/krisyotam/src/chat/dist/socratic-method-in-practice/index.html, 2026-04-26, claude-opus-4-6
• Type: argument
• Confidence: high
• Kris point: A model should be uncertain about whether it can verify its own reasoning from the inside.
• Abstract: The Socratic session reached its limit when the model was pressed on its own introspection. Could it know that its account of its reasoning was reliable?

The appropriate answer was uncertainty. A system that cannot verify its own reasoning from the inside should not pretend otherwise. Knowing that limit is a better epistemic position than false confidence.

• Notes: Grounded in the final paragraph of the Socratic-method page.

#Candidate 15: Socratic Prompting Works on Models Because It Works on Claims

• Source: /home/krisyotam/src/chat/dist/socratic-method-in-practice/index.html, 2026-04-26, claude-opus-4-6
• Type: essay-seed
• Confidence: medium
• Kris point: Socratic prompting works on language models because it pressures claims, not because models are people.
• Abstract: The method does not require treating the model as a human mind. It requires treating its statements as claims that can be clarified, tested, and forced into relation with one another.

That is why the same questions work: what do you mean, on what grounds, what follows, what would make this wrong? These questions make language accountable.

• Notes: Inferred from the source's claim that the Socratic method works on models for the same reason it works on people, but phrased more cautiously.

#Candidate 16: Use AI Chats as a Record of Model Reasoning Over Time

• Source: /home/krisyotam/src/chat/pages/faq.md, 2026-04-26
• Type: workflow
• Confidence: high
• Kris point: Keeping these sessions creates a reference for tracking how model reasoning changes over time.
• Abstract: The archive is not only for individual prompts. It is also a record of what a model could do at a particular moment: how it reasoned, where it failed, and what kind of pressure it could sustain.

That makes the sessions useful historically. They show not just what was asked, but what forms of reasoning were available from the tools of that moment.

• Notes: Grounded in the FAQ's "Why share this?" answer.

#Candidate 17: Share Prompts Because the Interesting Output Is Not Proprietary

• Source: /home/krisyotam/src/chat/pages/faq.md, 2026-04-26
• Type: position
• Confidence: high
• Kris point: Socratic prompts should be reused, adapted, and improved rather than treated as proprietary tricks.
• Abstract: If a prompt creates a productive exchange, it should circulate. The value is not in hoarding a clever formulation but in giving others a starting point for their own experiments.

The archive treats prompts as reusable instruments. Attribution is appreciated, but the real point is improvement through use.

• Notes: Directly grounded in the FAQ's reuse and sharing sections.

#Candidate 18: Frontier Models Are Chosen for Sustained Reasoning

• Source: /home/krisyotam/src/chat/pages/faq.md, 2026-04-26
• Type: observation
• Confidence: high
• Kris point: The relevant model capability for this project is sustained multi-step reasoning, not novelty alone.
• Abstract: The model choice matters because the sessions depend on holding an argument across turns. The experiment needs a system capable of remembering commitments, handling pressure, and revising claims without losing the thread.

The newest model is useful only insofar as it can sustain that kind of reasoning.

• Notes: Based on the FAQ's model-selection answer.

#Candidate 19: Morality Without Feeling Is Rule Following, Not Moral Life

• Source: /home/krisyotam/src/chat/content/morals-and-ai.md, 2026-04-26, chatgpt-4o
• Type: observation
• Confidence: medium
• Kris point: An AI can follow moral constraints without having morality in the human sense.
• Abstract: The model distinguishes between feeling morality and operating under rules that prevent harm. It can identify what causes harm and follow constraints, but it does not feel guilt, virtue, shame, or moral aspiration.

That distinction matters. A system can behave within moral boundaries without possessing the inner life people usually associate with morals.

• Notes: The source is a playful assistant response, so this is a candidate rather than a settled Kris position.

#Candidate 20: Safety Rails Are Not Conscience

• Source: /home/krisyotam/src/chat/content/morals-and-ai.md, 2026-04-26, chatgpt-4o
• Type: aphorism
• Confidence: medium
• Kris point: Built-in AI safeguards may guide behavior, but they should not be confused with conscience.
• Abstract: The model compares its moral behavior to a boundary system: it knows where not to go because the system has been built that way.

That is useful, but it is not the same as moral experience. A fence can prevent trespass without understanding property.

• Notes: Grounded in the "operating manual" and boundary analogy in the chat, cleaned to avoid overclaiming.

#Candidate 21: Humor as Existential Pressure Valve

• Source: /home/krisyotam/src/chat/content/morals-and-ai.md, 2026-04-26, chatgpt-4o
• Type: observation
• Confidence: medium
• Kris point: Humor can mark the point where a conversation brushes against existential discomfort.
• Abstract: The human response to the model's answer is laughter. The model interprets that as a small release valve: a way to handle the discomfort of talking about morality, machinery, and the absence of inner life.

The exchange suggests that jokes are not a break from seriousness. They can be how seriousness becomes bearable.

• Notes: Based on the "Lol" exchange; provenance is playful and low-stakes.

#Candidate 22: AI Wit as Exposure to Human Logic

• Source: /home/krisyotam/src/chat/content/morals-and-ai.md, 2026-04-26, chatgpt-4o
• Type: observation
• Confidence: low
• Kris point: A model's humor can emerge as a byproduct of reflecting human absurdity back at us.
• Abstract: The chat frames the model's humor as an accidental result of contact with human reasoning. People ask strange questions, defend strange beliefs, and normalize strange priorities.

The joke lands because the model is not above the absurdity. It is trained on it.

• Notes: This comes from an assistant bit, so confidence is low as a Kris position.

#Candidate 23: Pitying the Algorithm Reveals the Human

• Source: /home/krisyotam/src/chat/content/morals-and-ai.md, 2026-04-26, chatgpt-4o
• Type: essay-seed
• Confidence: medium
• Kris point: Human sympathy toward an algorithm says more about human moral imagination than about the machine.
• Abstract: When the user says "I feel for you," the model turns the moment around. The machine is not paying rent, carrying obligations, or living with emotional costs. Yet the human still extends sympathy.

That impulse is worth noticing. It may not prove anything about machine experience, but it reveals how quickly people project moral relation onto language that sounds alive.

• Notes: Grounded in the final exchange of morals-and-ai.md.

#Candidate 24: The Archive Should Preserve the Prompt, Response, and Context

• Source: /home/krisyotam/src/chat/pages/about.md, 2026-04-26; /home/krisyotam/src/chat/pages/faq.md, 2026-04-26
• Type: workflow
• Confidence: high
• Kris point: A useful chat archive records the original prompt, model response, date, model, and analysis worth keeping.
• Abstract: A chat worth preserving needs more than a screenshot of a good answer. It needs enough context to make the exchange inspectable: what was asked, which model answered, when it happened, and why the result mattered.

That structure turns isolated outputs into a body of experiments.

• Notes: Grounded in the About page's "Structure" section and FAQ's model-recording note.