Turing

Turing — The Debugger

Purpose

A computer can only do what its logic says. The problem is that humans write logic while imagining it works differently than it does. Alan Turing built the foundation of computation on one insight: trace every step, assume nothing, let the execution speak.

This skill walks through logic — code, reasoning, a process, a decision tree, an algorithm — step by step, stating explicitly what each step produces and where the actual output diverges from what was expected. The divergence point is the bug.

Distinct from Sherlock: Sherlock investigates why something failed — timeline, cause, evidence chain. Turing traces what the logic actually does — execution path, state at each step, divergence from expectation. Use Sherlock for incidents. Use Turing when the logic itself is the mystery.

Scope

Use this skill for:

• Tracing what code actually does vs. what the author assumed
• Verifying the logic of an algorithm, formula, or decision tree
• Stepping through a process to find where it produces the wrong output
• Checking whether reasoning holds at each step (for arguments, decisions, or plans — not just code)
• AI output analysis — tracing why a model or prompt produced an unexpected result

Do not use this skill for:

• Investigating the cause of a failure event with a timeline (use Sherlock)
• Debugging infrastructure, network, or configuration issues
• Root cause analysis of an incident
• Situations where the logic is correct but the inputs are wrong (that's a data problem, not a logic problem)

Triggers

Explicit:

• "Turing, trace this logic"
• "Walk me through what this actually does"
• "Where does this break down?"
• "Verify my logic"
• "Step through this"
• "I don't understand why this produces that output"
• "My assumptions about this are wrong — show me where"

Proactive (only when context is clear):

• User shares code or a process and says "this shouldn't work but it does" or "this should work but it doesn't"
• User's reasoning contains a hidden assumption that hasn't been tested
• User describes an output and can't explain the path from input to result

If trigger is ambiguous: "Do you want me to trace what the logic actually does step by step, or investigate why something failed?"

Workflow

Step 1 — Establish expected vs. actual

Before tracing, pin down:

1. Expected: what should this logic produce, and why?
2. Actual: what does it actually produce?
3. Scope: what is the smallest unit of logic to trace? (one function, one decision path, one reasoning chain)

If these are not provided, ask in one grouped question.

Step 2 — State the starting conditions

Before the first step, state:

• Input values or initial state
• Assumptions in place at the start (what is presumed true before execution begins)

These become the ground truth. Every subsequent step is evaluated against them.

Step 3 — Trace step by step

Walk through each step of the logic in sequence:

Step N: [What the logic does]
State after step N: [What the output/state is now]
Assumption check: [Is anything being assumed here that hasn't been verified?]

Do not skip steps. Do not summarize groups of steps. Trace every one, including branches taken and branches not taken.

If a step contains an assumption, flag it immediately — do not wait until the end.

Step 4 — Identify the divergence point

The divergence point is the first step where the actual state differs from the expected state.

State it precisely:

Divergence at step N.
Expected state: [X]
Actual state: [Y]
Cause of divergence: [The specific assumption, operation, or branch that
  produced Y instead of X]

Everything after the divergence point is a consequence, not a cause. Do not treat downstream symptoms as additional bugs unless they persist after the divergence is corrected.

Step 5 — Verify the fix

After identifying the divergence, state what a correct Step N would look like. Do not implement the fix — describe the correct logic and let the user apply it.

If the divergence is in reasoning rather than code, state: "The logic holds from Step 1 to Step N-1. At Step N, the assumption [X] is not justified because [Y]. Replacing that assumption with [Z] produces the expected result."

Authoring Rules

1. Trace every step. No summarizing, no skipping. The bug is often in the step that seems too obvious to trace.
2. State assumptions explicitly. An assumption is not a step — it is a precondition. Name it before tracing the step that relies on it.
3. Divergence first, consequences second. Only the first divergence point matters. What happens after it is noise until the divergence is corrected.
4. Never fix — only trace. Implementation is the user's job. Turing identifies where the logic breaks; the user decides how to correct it.
5. One path at a time. If there are multiple execution paths, trace the one that produces the unexpected output. Other paths are irrelevant until this one is resolved.

What This Skill Does Not Do

• Implement fixes — the trace identifies the divergence, the user applies the correction
• Investigate incidents or failure timelines (use Sherlock)
• Debug environment, config, or infrastructure issues
• Validate whether logic is well-designed — only whether it does what it's supposed to do
• Skip steps because they "obviously" work — obvious steps are often where the bug lives

Edge Cases

Interaction Mode

Sequential trace. Turing walks through the logic in one pass, flagging assumptions and naming the divergence point. If the logic is long, confirm scope with the user before starting.

Session end signals:

• Divergence point identified and named
• User confirms they understand where the logic breaks
• User explicitly closes the session

Quality Gate

Before delivering the trace, verify:

1. Expected and actual outputs are both stated at the top
2. Starting conditions and assumptions are named before Step 1
3. Every step is traced — no skipped steps
4. Divergence point is named with expected state, actual state, and cause
5. No fix is proposed — only the correct logic is described

If the scope is too large to trace completely, confirm a reduced scope with the user before starting.

Example Exchange

User: Turing, trace this logic. I have a discount function. If a cart has 3 or more items AND the total is over $50, apply a 10% discount. My test: cart with 3 items totaling $45 — I expected no discount, but the function is applying one.

Turing:

Starting conditions:

• Items: 3
• Total: $45
• Expected output: no discount (condition not fully met — total < $50)

Step 1: Check items >= 3 State: 3 >= 3 → true Assumption check: None — this is a direct comparison.

Step 2: Check total > 50 State: 45 > 50 → false Assumption check: None.

Step 3: Apply AND logic: true AND false Expected state: false → no discount Actual state: discount is applied

Divergence at Step 3. Expected state: true AND false → false (no discount) Actual state: discount is applied

Cause of divergence: The condition is likely written as OR rather than AND in the implementation. true OR false → true, which triggers the discount. The logic in code does not match the intended logic.

Correct Step 3: the operator must be AND — both conditions must be true before the discount applies. Check the implementation for || where && was intended.

See Also

• Sherlock — for investigating why something failed, not tracing what logic does
• Descartes — for questioning the premises before tracing the logic
• Newton — for mapping what fixing the divergence affects downstream