Certainty

Input: $ARGUMENTS

Core Principle

Do not stop until the answer is thoroughly satisfying. This skill runs at maximum depth, chains across multiple skills, evaluates its own output between iterations, and continues until resolution criteria are met. It is the “give me everything you’ve got” mode.

Phase 1: Classify and Set Resolution Criteria

1. What Is the Question?

Extract the core question to resolve. State it precisely.

2. What Does “Resolved” Mean?

Define what a satisfying answer looks like:

• Factual question: answer is verified, sourced, and no credible counter-evidence remains.
• Analytical question: all major branches explored, each tested with AR/AW, synthesis accounts for all findings.
• Decision question: clear recommendation with justified confidence, alternatives considered and rejected for stated reasons, risks identified.
• Problem question: root cause confirmed, prevention derived, no competing explanations unaddressed.

If the user didn’t specify, ask: “What would make the answer satisfying to you?” If unclear, default to: no remaining unresolved branches, untested assumptions, or competing conclusions.

3. Classify the Input

Use the taxonomy to determine starting mode:

4. Enhanced Classification Signals

Some input patterns route to newer skills before the main analysis:

Phase 2: Iterative Resolution

Iteration Loop

ITERATION 1:
  1. Run the starting category skill at 8x depth.
  2. Collect all findings.
  3. Self-evaluate: → INVOKE /evaluate on own output.
     - Are there unresolved branches? → explore them.
     - Are there untested assumptions? → test them with /claim.
     - Are there competing conclusions? → stress-test each with /araw.
     - Are there gaps in the space? → fill with /search.
     - Are there obvious things missed? → check with /obv.
     - Are there obvious bad outcomes ignored? → check with /obo.
     - Is self-deception present? → check with /sdc.
     - Are there ethical dimensions unaddressed? → check with /eth.
     - Are there safety concerns unaddressed? → check with /saf.
  4. If unresolved items found → continue to Iteration 2.
     If resolved → proceed to Final Synthesis.

ITERATION 2:
  1. Address each unresolved item from Iteration 1.
     - Unresolved branch → /search or /claim on that branch.
     - Untested assumption → /claim [assumption].
     - Competing conclusions → /claim [conclusion A] then /claim [conclusion B].
     - Gap → /search [gap area].
  2. Collect new findings.
  3. Self-evaluate again.
  4. If unresolved items found → continue to Iteration 3.
     If resolved → proceed to Final Synthesis.

ITERATION 3+:
  Continue the pattern. Each iteration should have FEWER unresolved items.
  If an iteration produces no new findings or progress, that's the natural stopping point.

Stopping Conditions

Stop when ANY of these is true:

• No unresolved branches remain — every significant avenue has been explored.
• No untested assumptions remain — every load-bearing assumption has been AR/AW tested.
• No competing conclusions remain — contradictions have been resolved.
• Diminishing returns — an iteration produced no new substantive findings.
• Natural bedrock — the analysis reached fundamental axioms or empirical facts that can’t be broken down further.

Anti-Stopping Conditions

Do NOT stop if:

• There are branches marked “uncertain” or “needs investigation.”
• Two conclusions contradict each other and neither has been eliminated.
• Key assumptions are marked “unknown” status.
• The user’s resolution criteria haven’t been met.

Phase 3: Final Synthesis

After all iterations:

CERTAINTY RESOLUTION:

QUESTION: [the original question]
RESOLUTION CRITERIA: [what "resolved" means]

ITERATIONS COMPLETED: [N]
TOTAL FINDINGS: [N across all iterations]

ANSWER:
[The thoroughly resolved answer — comprehensive, backed by findings from all iterations]

CONFIDENCE: [high / medium with stated reason]

KEY EVIDENCE:
1. [Most important finding supporting the answer]
2. [Second most important]
3. [Third most important]

WHAT WAS TESTED AND SURVIVED:
- [Claim/assumption that was AR/AW tested and held]
- [Claim/assumption that was AR/AW tested and held]

WHAT WAS TESTED AND ELIMINATED:
- [Claim/assumption that was tested and failed]
- [Claim/assumption that was tested and failed]

REMAINING UNCERTAINTY:
- [Anything that couldn't be fully resolved, with explanation of why]

WHAT WOULD CHANGE THIS ANSWER:
- [New evidence or conditions that would alter the conclusion]

Phase 4: Extended Analysis Tools

During any iteration, draw on these skills when they would resolve open questions:

Depth

This skill always runs at maximum depth. There is no 1x/2x mode for /certainty. If you want lighter analysis, use the specific category skill directly.

After Completion

Follow-Up Routing

After certainty resolution, the user may need:

• “What should I do?” → INVOKE: /action or /how
• “What are the implications?” → INVOKE: /sycs
• “What skill should I run next?” → INVOKE: /next or /fonss
• “What’s still unresolved?” → INVOKE: /tbd
• “Iterate on this” → INVOKE: /iterate

Integration

• Use from: Any router (when the user asks for maximum effort), /ecal (when effort calibration says maximum depth warranted)
• Routes to: All category skills at 8x, plus extended analysis tools listed in Phase 4
• Differs from: Individual category skills (certainty chains across multiple skills iteratively; category skills run once), /iterate (certainty resolves a question to completion; iterate improves an existing output)
• Complementary: /ecal (determine whether certainty is warranted), /evaluate (used within certainty for self-evaluation between iterations)