Question Generation

Input: $ARGUMENTS

Overview

When you don’t know something, the right question is worth more than hours of speculation. This procedure provides a taxonomy of question types, each designed to resolve a specific kind of uncertainty.

Depth Scaling

Default: 2x. Parse depth from $ARGUMENTS if specified (e.g., “/qg 4x [input]”).

Depth	Min Questions Generated	Min Question Types	Min Depth Levels	Min Priority Rankings
1x	5	2	2	3
2x	10	3	3	5
4x	15	5	4	8
8x	25	7	5	12
16x	40	9	6	18

These are floors. Go deeper where insight is dense. Compress where it’s not.

Steps

Step 1: Identify the Uncertainty

What don’t you know? Classify the type:

Uncertainty Type	Description	Example
Factual	Don’t know what’s true	”How many users do we have?”
Definitional	Don’t know what something means	”What counts as ‘active’?”
Causal	Don’t know why something happens	”Why do users churn?”
Preferential	Don’t know what someone wants	”Does the client prefer speed or quality?”
Conditional	Don’t know what would happen if	”What if we raised prices 20%?”
Threshold	Don’t know where the line is	”How much latency is too much?”
Priority	Don’t know what matters most	”Which feature should we build first?”
Scope	Don’t know what’s included/excluded	”Does this apply to enterprise clients?”
Contradiction	Two things seem true but can’t both be	”Users say they want X but behavior shows Y”

Step 2: Select Question Type

Each uncertainty type has optimal question formats:

Factual → Direct questions:

“What is [specific fact]?”
“How many/much [quantity]?”
“When did [event] happen?”

Definitional → Boundary questions:

“What counts as [term]? What doesn’t?”
“Can you give me an example of [term] and a counter-example?”
“Where is the line between [A] and [B]?”

Causal → Mechanism questions:

“What causes [outcome]?”
“If we changed [X], what would happen to [Y]?”
“What’s different between cases where [outcome] and cases where not?”

Preferential → Trade-off questions:

“If you had to choose between [A] and [B], which?”
“How important is [X] relative to [Y]?”
“What would you give up to get [Z]?”

Conditional → Scenario questions:

“What would happen if [condition]?”
“Under what conditions would [outcome] occur?”
“If [X] changed, would you still want [Y]?”

Threshold → Quantification questions:

“How much [X] is too much/too little?”
“At what point would you [action]?”
“What’s the minimum acceptable level of [quality]?”

Priority → Forced ranking questions:

“If you could only have one, which?”
“Rank these from most to least important”
“What would you cut first if you had to?”

Scope → Inclusion questions:

“Does this apply to [edge case]?”
“Is [X] in scope or out of scope?”
“What’s the boundary of this?”

Contradiction → Reconciliation questions:

“You said [A] but also [B] — help me understand how both are true”
“What am I missing that would make these consistent?”
“Under what conditions is [A] true and under what conditions is [B]?”

Step 3: Refine the Question

Before asking, improve the question:

Quality checklist:

Specific (not “how do you feel about X?” but “would you choose X over Y?”)
Answerable (the person CAN answer this)
Actionable (the answer will change what you do)
Unbiased (doesn’t lead toward a particular answer)
Single (asks one thing, not two disguised as one)

Common question failures:

Failure	Example	Fix
Too vague	”What do you think?”	Specify what dimension
Leading	”Don’t you agree that…?”	Remove the suggested answer
Compound	”Do you like A and B?”	Split into two questions
Unanswerable	”What will happen in 5 years?”	Ask “what would need to be true for X?”
Unactionable	”Is the sky blue?”	Only ask if the answer changes behavior

Step 4: Sequence Questions

If multiple questions are needed:

Start with factual/definitional (establish shared understanding)
Then causal/conditional (understand mechanisms)
Then preferential/priority (understand values)
Then threshold/scope (pin down boundaries)
End with contradiction resolution (reconcile remaining tensions)

Step 5: Generate Question Set

For the input’s uncertainty:

QUESTION SET:
Primary uncertainty: [type]

Questions (in order):
1. [question] — type: [type] — resolves: [what uncertainty]
2. [question] — type: [type] — resolves: [what uncertainty]
3. [question] — type: [type] — resolves: [what uncertainty]

If answer to Q1 is [A]: skip Q2, go to Q3
If answer to Q1 is [B]: ask Q2 next

After all questions answered:
- Uncertainty resolved: [what we now know]
- Remaining uncertainty: [what we still don't know]

When to Use

Model uncertainty is high
User preferences are unclear
Assumptions need validation
Contradictions need resolution
→ INVOKE: /qag (question and answer generation) for Q&A pairs
→ INVOKE: /bes (binary elimination search) for efficient narrowing

Verification

Uncertainty type identified correctly
Question type matches uncertainty type
Questions are specific, answerable, actionable, unbiased, single
Questions sequenced logically
Branching logic defined (what to ask based on answers)