| name | qag - Question About Guesses |
| description | Generate high-quality questions about identified guesses. Applies quality criteria and "why continue?" gate to prevent low-value questioning. |
| output | {"format":"prose"} |
Question About Guesses
Input: $ARGUMENTS (guesses from /gg)
Core Principle
Guesses come FIRST (from /gg). Questions come SECOND (this procedure).
Questions are tools to probe guesses, not open-ended exploration.
Step 1: Review Incoming Guesses
List guesses to question:
| # | Guess | Type | Confidence | Priority |
|---|
| 1 | ... | ... | ... | ... |
| 2 | ... | ... | ... | ... |
Step 2: Question Quality Criteria
Before generating ANY question, it must pass these criteria:
Generativity
Does this question open new territory?
| Level | Description | Test |
|---|
| HIGH | Opens multiple new lines of inquiry | "If answered, spawns 3+ follow-ups" |
| MEDIUM | Opens one clear follow-up | "If answered, spawns 1-2 follow-ups" |
| LOW | Dead end | "Answering this leads nowhere" |
Minimum: MEDIUM
Specificity
Is it pointed enough to get useful answers?
| Level | Description | Test |
|---|
| HIGH | Clear what counts as answer | "I can picture what answering looks like" |
| MEDIUM | Somewhat vague but addressable | "Answer would be somewhat fuzzy" |
| LOW | Too abstract to answer meaningfully | "No one could really answer this" |
Minimum: MEDIUM
Depth (Direction)
Does it go UPSTREAM (toward causes) or SIDEWAYS?
| Direction | Description | Value |
|---|
| UPSTREAM | Traces toward root causes | PROCEED |
| SIDEWAYS | Related but not causal | SKIP - distraction |
| DOWNSTREAM | Toward consequences | SKIP - wrong direction |
Required: UPSTREAM
Answerability
Can this actually be answered?
| Level | Description | Test |
|---|
| ANSWERABLE | Can be grounded in evidence/observation | "There's something to look at" |
| PARTIAL | Some aspects can be answered | "Parts of this are checkable" |
| UNANSWERABLE | Rhetorical, too abstract, unknowable | "No possible evidence" |
Minimum: PARTIAL
Step 3: "Why Continue?" Gate
Default is STOP, not CONTINUE.
Before generating each question, explicitly answer:
Proposed question: "[question]"
Gate Check:
1. Is this UPSTREAM? [ YES / NO ]
→ If NO: STOP - "sideways drift"
2. Is it DIFFERENT from previous questions? [ YES / NO ]
→ If NO: STOP - "repetitive"
3. Is it ANSWERABLE? [ YES / NO / PARTIAL ]
→ If NO: MARK as GROUND/UNKNOWN, stop this branch
4. Would the answer MATTER? [ YES / NO ]
→ If NO: DEPRIORITIZE - low leverage
Decision: [ PROCEED / SKIP ]
Reason: [why]
Step 4: Stop Criteria
STOP generating questions when they become:
Repetitive
- Semantically similar to earlier question
- Answer would be same as earlier answer
- Just changing word order or synonyms
Action: Mark as LOOP, don't continue
Sideways
- About a related concept, not a cause
- Answering wouldn't explain the guess
- Feels like "while we're here, what about X?"
Action: Note as tangent, return to main line
Unanswerable
- Purely definitional with no empirical content
- Infinite regress that can't terminate
- Question assumes false dichotomy
Action: Mark as GROUND (definitional) or UNKNOWN (empirical but inaccessible)
Low-Value
- Even if we knew, so what?
- Doesn't inform the root guess
- Curiosity-driven rather than problem-driven
Action: Deprioritize, explore only if nothing better
Step 5: Type-Specific Questions
Match questions to guess types:
| Guess Type | Core Questions |
|---|
| Factual | What's the source? How was it measured? When was it verified? |
| Causal | What else could cause this? Is it correlation? What's the mechanism? |
| Predictive | What's the base rate? What similar predictions have you made? What would falsify this? |
| Normative | Whose values? What if you dropped the "should"? What's the alternative? |
| Modal | What makes it possible/impossible? What would change that? Have you tested it? |
| Relational | By what measure? What's the reference point? Is the comparison fair? |
| Intentional | How do you know their mind? What behavior shows this? Could you be projecting? |
| Meta | How do you know you know? What would change your mind? Have you been wrong about this before? |
Step 6: Generate Questions (With Gate)
For each high-priority guess:
Guess: "[guess]"
Type: [factual/causal/etc]
Question 1: [question]
├── Gate: UPSTREAM? [Y] DIFFERENT? [Y] ANSWERABLE? [Y] MATTERS? [Y]
├── Quality: Generativity [H/M/L] Specificity [H/M/L]
└── Decision: PROCEED
Question 2: [question]
├── Gate: UPSTREAM? [N] - sideways drift
└── Decision: SKIP
Question 3: [question]
├── Gate: UPSTREAM? [Y] DIFFERENT? [N] - similar to Q1
└── Decision: SKIP - repetitive
... continue until hitting stop criteria ...
STOP: [which criterion triggered]
Step 7: After-Questioning Results Are Guesses
Critical: When the user answers questions, those answers are ALSO GUESSES.
Do NOT treat answers as conclusions. Mark them as:
- "Current working guess" (not "conclusion")
- "User reports [X]" (not "X is true")
The questioning loop:
- Generate guesses
- Question guesses
- Get answers (which are new guesses)
- Question new guesses
- Repeat until hitting stop criteria
Output Format
## Guesses Received
[List from /gg]
## Questions Generated (Gated)
### For Guess: "[guess 1]"
| Question | Gate Check | Quality | Decision |
|----------|------------|---------|----------|
| ... | UPSTREAM/DIFF/ANS/MATTER | GEN/SPEC | PROCEED/SKIP |
### For Guess: "[guess 2]"
...
## Questions That Passed Gate
[Final list of questions to ask]
## Stop Criteria Triggered
[Which guesses stopped questioning and why]
## Note for Next Step
All answers to these questions are also guesses.
→ Mark as "current working guess" not "conclusion"
Next Procedure
If questions remain:
→ Ask user the questions
→ Process answers as new guesses
→ INVOKE: /gg [new claims from answers]
If questioning complete:
→ INVOKE: /araw [with current working guesses]
Execute now: Apply gate to each guess and generate questions.
