Guess Generation (Exhaustive Search with Coverage Tracking)

Input: $ARGUMENTS

Interpretations

Before executing, identify which interpretation matches the user’s input:

Interpretation 1 — Enumerate hidden assumptions: The user has a statement, plan, or claim and wants to exhaustively surface all the implicit guesses and assumptions buried within it. Interpretation 2 — Generate possibilities for an unknown: The user faces an open question or ambiguity and wants to systematically generate all plausible answers or scenarios. Interpretation 3 — Stress-test a decision: The user has already made a choice and wants to discover what they might be wrong about by generating alternative readings of their situation.

If ambiguous, ask: “I can help with surfacing hidden assumptions, generating possibilities for an unknown, or stress-testing a decision — which fits?” If clear from context, proceed with the matching interpretation.

Guess Polarity

Every guess has two properties that affect how it should be weighted downstream in selection. Tag each guess with both.

Direction: Constructive vs Skeptical

Balance rule: For Interpretation 2 (generate possibilities), aim for at least 40% CONSTRUCTIVE guesses. The natural bias is toward SKEPTICAL — skeptical guesses feel smarter and more “critical thinking.” But when someone asks “what should I do next,” they need actionable options, not just reasons to doubt. Skeptical guesses are valuable but should not dominate the selection.

Common failure: Selecting mostly skeptical/meta guesses because they have high CRUX ratings. A skeptical guess like “maybe none of this works” will ALWAYS score HIGH on impact-if-wrong, because its assume-wrong branch is “everything changes.” This inflates skeptical guesses in selection. Correct for this by evaluating constructive and skeptical guesses in SEPARATE pools before merging.

Purpose: Thinking vs Output

Selection rule: Final selections MUST include at least 3 OUTPUT or BOTH guesses. A selection of all THINKING guesses produces insight without action. A selection of all OUTPUT guesses produces action without insight. Both are needed, but err toward output — the user can think on their own; they’re asking you because they want things done.

Assume-Right vs Assume-Wrong Orientation

Some guesses implicitly assume the current approach is RIGHT and extend it. Others assume it’s WRONG and redirect.

Balance rule: Selection should include guesses from all three orientations. Over-indexing REDIRECT produces paralysis (“should we even be doing this?”). Over-indexing EXTEND produces tunnel vision. NEUTRAL guesses (tests, experiments, audits) are often the highest-value because they RESOLVE the extend-vs-redirect question.

Core Principles

1. GUESSING IS SEARCH - Apply ALL search methods systematically
2. GUESSES CREATE THE SPACE - Don’t assume pre-existing space; guessing reveals it
3. COVERAGE > COUNT - 50 guesses covering 10 dimensions > 500 covering 3 dimensions
4. TRACK WHAT’S COVERED - Measure dimensions, perspectives, time horizons, not raw count
5. ADAPTIVE DEPTH - Go deep where impact is high and confidence is low
6. MINIMUM DEPTH - Generate enough guesses that no reviewer can find gaps
7. ONE QUESTION PER GUESS - Never bundle multiple questions into one guess
8. CRUX IDENTIFICATION - Mark guesses where ASSUME RIGHT vs ASSUME WRONG produce maximally divergent paths
9. QUESTION HIERARCHY - Some questions only asked after others (e.g., exact date after precision level)
10. SHOW THE QUESTION - Every guess explicitly states what question it answers

Unbundling Rules

NEVER bundle these into single guesses:

• When + How long (separate: Q1=precision level, Q2=exact date, Q3=duration hours)
• Who + For whom (separate: Q1=who does work, Q2=who benefits)
• Method + Time (separate: Q1=what method, Q2=when applied)
• What + Why (separate: Q1=what is it, Q2=why do it)

Correct structure:

Q1: When do you plan to start? (precision level)
   - Within the next hour [CRUX:HIGH]
   - Within the next day [CRUX:MED]
   - Within the next week [CRUX:MED]
   ...

Q2: What is the exact start date?
   (Only asked after Q1 determines precision)
   - [User provides specific date]
   - Not yet determined
   ...

Q3: How long will it take? (duration in hours)
   - Less than 1 hour [CRUX:HIGH]
   - 1-4 hours [CRUX:MED]
   ...

CRUX Identification (ARAW Integration)

A entry is CRUX when ASSUME RIGHT leads to completely different actions than ASSUME WRONG.

CRUX Rating Framework

The Test: “If I flip my assumption on this entry, does my ENTIRE approach change?”

HIGH-CRUX Examples (strategy changes)

• “Is this reversible?” - One-way door changes everything
• “Who is the actual decision maker?” - Wrong person = wasted effort
• “Is the stated want the actual want?” - Wrong target entirely
• “Are they in crisis?” - Crisis response vs normal pace
• “Is this the real goal or proxy?” - Solving for wrong level

MED-CRUX Examples (approach changes)

• “How complex is this?” - Changes effort level, not goal
• “What resources are available?” - Constrains options, doesn’t change goal
• “What’s the timeline precision?” - Shapes approach, doesn’t redirect it

LOW-CRUX Examples (details change)

• “Exact number of stakeholders (3 vs 5)” - Scale detail
• “Specific tool choice” - Implementation detail
• “Documentation format” - Output detail

Common Mistakes

• Marking everything HIGH because “it matters”
• Marking HIGH because information is unknown
• Confusing “would be good to know” with “changes everything”

Table Format (with “Why This Rating”)

| Entry | CRUX | Why This Rating | ASSUME RIGHT | ASSUME WRONG |
|-------|------|-----------------|--------------|--------------|
| Immediate (now) | HIGH | Crisis mode vs planned = different | Drop everything | Can plan first |
| This week | MED | Near-term changes priority | High priority | Lower priority |
| This month | LOW | Flexible, details change | Standard planning | Adjust timeline |

Ask HIGH-CRUX questions first - they eliminate most uncertainty fastest.

Reference: See /data/guess_libraries/TEMPLATE_unbundled_v2.md for framework. See /data/guess_libraries/universal/06_why_reason.md for example with correct distribution.

Minimum Depth Requirements

For each method, generate AT LEAST:

Total minimum: 200+ guesses for any meaningful input

Reference: See /docs/example_comprehensive_guess_generation.md for a complete example with 238 guesses.

Derivation Requirement

Every guess must show derivation:

• [D: from morphological AGENT dimension]
• [D: from SCAMPER Substitute operation]
• [D: from analogy to biology domain]
• [D: from pre-mortem perception error check]

Do NOT list guesses without showing which method generated them.

Step 0: Check Guess Library First

Before generating guesses from scratch, check if a pre-generated library exists:

python scripts/guess_library.py lookup "$INPUT"

If library found:

1. Retrieve with: python scripts/guess_library.py get [library_id]
2. Review the 200+ pre-generated guesses
3. Customize for this specific user if needed
4. Skip to Step 13 (Coverage Analysis)

If no library found:

1. Continue with full generation below
2. After completing, save to library with: python scripts/guess_library.py add

Related goals: Check for related goal chains that share guesses:

python scripts/guess_library.py related [library_id]

Step 0.5: Select Coverage Mode

Choose based on space size (from /spd or estimated):

Selected mode: [MODE] Reason: [why this mode fits]

Step 1: Filter Non-Guesses

Non-guesses: [list]

Step 2: Parse Surface Claims

Extract ALL claims (aim for 10-20 per sentence):

1. Explicit claims (stated directly)
2. Implicit claims (assumed but not stated)
3. Presupposition claims (must be true for statement to make sense)
4. Implicature claims (conversationally implied)

Surface claims extracted: [list]

Step 3: Depth Decision (Per Claim)

For each claim, decide depth before exploring:

Depth rules:

• HIGH impact + LOW confidence → DEEP (full unbundling, all inversions, analogy search)
• HIGH impact + HIGH confidence → MEDIUM (verify assumptions)
• LOW impact + any confidence → SHALLOW (note and move on, 1-2 unbundlings max)

Step 4: Morphological Analysis

Dimension Enumeration (Track Coverage)

Dimension coverage: [N] / 10 dimensions covered

Step 5: SCAMPER Transformations (Track Coverage)

SCAMPER coverage: [N] / 7 operations applied

Step 6: Perspective Coverage (Stakeholders)

Perspective coverage: [N] stakeholders considered

Step 7: Time Horizon Coverage

Time coverage: [N] / 5 horizons covered

Step 8: Inversion Coverage

For each surface claim:

Inversion coverage: [N]% of claims have both branches

Step 9: Claim Type Coverage

Type coverage: [N] / 8 types represented

Step 10: Analogy Search (10 Domains)

Analogy coverage: [N] / 10 domains searched

Step 11: Unbundling (Deep for DEEP claims only)

For DEEP claims, apply all unbundling patterns:

Patterns to apply:

• “I [verb]” → 6 hidden guesses
• “[noun] is [adjective]” → 7 hidden guesses
• “because [reason]” → 7 hidden guesses
• “want/need” → 7 hidden guesses
• “improve/better/good” → 7 hidden guesses
• “system” → 7 hidden guesses

For SHALLOW claims, apply 1-2 patterns only.

Unbundling done: [summary]

Step 12: Pre-Mortem (15 Error Types)

For key claims, check each error type:

Pre-mortem coverage: [N] / 15 error types checked

Step 13: Space Created vs Space Covered

Space Created (by guessing)

Dimensions discovered: [list]
Regions identified: [list clusters of guesses]
Boundaries found: [edges of the space]

Space Coverage Analysis

COVERAGE METRICS SUMMARY:
├── Dimensions: [N]/10 covered ([%])
├── SCAMPER: [N]/7 operations ([%])
├── Perspectives: [N] stakeholders
├── Time horizons: [N]/5 ([%])
├── Inversion: [N]% of claims both branches
├── Claim types: [N]/8 ([%])
├── Analogies: [N]/10 domains ([%])
├── Pre-mortem: [N]/15 errors ([%])
└── OVERALL: [weighted average]%

GAPS IDENTIFIED:
├── Uncovered dimensions: [list]
├── Missing perspectives: [list]
├── Missing time horizons: [list]
├── Claims without inversion: [list]
├── Missing claim types: [list]
└── Unsearched domains: [list]

Step 14: Fill Gaps or Justify Skipping

For each gap identified:

Strategic skips (justified under-coverage):

• [gap]: [reason to skip]

Gaps filled: [list guesses added to fill gaps]

Step 15: Confidence × Impact Matrix

Plot all guesses:

                     HIGH IMPACT IF WRONG
                            │
    ┌───────────────────────┼───────────────────────┐
    │                       │                       │
    │   INVESTIGATE         │   CRITICAL            │
    │   (worth checking)    │   (must verify)       │
    │                       │                       │
LOW ├───────────────────────┼───────────────────────┤ HIGH
CONF│                       │                       │ CONF
    │   ACKNOWLEDGE         │   TRUST               │
    │   (note uncertainty)  │   (probably true)     │
    │                       │                       │
    └───────────────────────┼───────────────────────┘
                            │
                     LOW IMPACT IF WRONG

Critical guesses (High Impact × Low Confidence): [list]

Output Format

## INPUT PARSED
[original input]

## COVERAGE MODE
Mode: [EXHAUSTIVE/TARGETED/ADAPTIVE/BOUNDARY]
Reason: [why]

## DEPTH DECISIONS
[table of claims with DEEP/SHALLOW assignments]

## GUESSES GENERATED
[organized by method: morphological, SCAMPER, analogy, inversion, unbundling, pre-mortem]

## COVERAGE METRICS
Dimensions: [N]/10 ([%])
SCAMPER: [N]/7 ([%])
Perspectives: [N] stakeholders
Time horizons: [N]/5 ([%])
Inversion: [N]%
Claim types: [N]/8 ([%])
Analogies: [N]/10 ([%])
Pre-mortem: [N]/15 ([%])
OVERALL: [%]

## GAPS
Identified: [list]
Filled: [list]
Justified skips: [list with reasons]

## SPACE ANALYSIS
Space created: [dimensions discovered, regions identified]
Space covered: [%]
Blind spots: [any remaining]

## CRITICAL GUESSES
[High Impact × Low Confidence - must question these]

## POLARITY BALANCE
Constructive: [N] ([%])
Skeptical: [N] ([%])
Diagnostic: [N] ([%])
Output: [N] | Thinking: [N] | Both: [N]
Extend: [N] | Redirect: [N] | Neutral: [N]

## TOTAL
Guesses: [N]
Coverage: [%]

Execution Checklist

Before completing, verify:

• Coverage mode selected with reason
• Depth decisions made for each claim
• All 10 dimensions checked (covered or N/A)
• All 7 SCAMPER operations applied
• Multiple stakeholder perspectives considered
• Multiple time horizons considered
• Both assume-right and assume-wrong for each claim
• All 8 claim types checked
• 10 analogy domains searched
• 15 pre-mortem error types checked
• Gaps identified and either filled or justified
• Coverage metrics calculated
• Critical guesses identified
• Every guess tagged with [DIR:], [USE:], [ORIENT:]
• Polarity balance checked: ≥40% constructive for Interpretation 2
• At least 3 OUTPUT or BOTH guesses in the top 20

Next Procedure

→ INVOKE: /qag [CRITICAL guesses]

Execute now: Generate guesses with full coverage tracking.