Debiasing Electrical Reasoning

Before finalizing any electrical/electronics decision (schematic, fault diagnosis, component rating, calculation), run it through an adversarial self-check prompt:

Here is my decision/diagnosis: [describe it]
Input data: [parameters, measurements, schematic]

Act as a skeptical engineering reviewer. Identify:
1. What implicit assumptions did I make without verifying them?
2. Did I confuse correlation with causation (e.g., "component is hot → therefore faulty")?
3. Is there a simpler/cheaper explanation (Occam's razor) I overlooked?
4. Does this match a "textbook example" pattern that may not apply to this real-world case?
5. Could I have mixed up units, signs, or current/voltage directions?

Treat the output as hypotheses to verify, not final truth.

Progress:

• Step 1: Write the decision/diagnosis explicitly, in one sentence
• Step 2: List measured data and stated assumptions in separate columns
• Step 3: Run the adversarial review prompt (above)
• Step 4: Classify each AI objection: valid risk / textbook bias / irrelevant
• Step 5: Ask AI to steelman the opposite conclusion (see below)
• Step 6: Physically verify any valid risk before acting (measurement, datasheet, simulation)
• Step 7: Document remaining uncertainty

Steelmanning the opposite

After the initial review, ask explicitly:

Assume my conclusion is wrong. What alternative explanation fits the same data equally well or better?

This counters anchoring on the first hypothesis — the single most common failure mode in troubleshooting.

Checking for textbook/training-data bias specifically

Ask directly:

Is this based on verified real-world practice, or is it a common textbook pattern that may not
apply here (e.g., ignoring parasitics, idealizing the power source, outdated code/standard)?

Example 1 — Root-cause misdiagnosis (post hoc fallacy) Input: "The filter capacitor is bulged, so it must be the cause of the power supply failures." AI review output: Flags "post hoc" reasoning — the bulging could be a symptom of overheating caused by a different failure (e.g., a shorted output transistor raising ambient temperature), not the root cause. Recommends measuring ESR, checking neighboring component temperatures, and verifying load current before replacing the cap.

Example 2 — Unverified textbook assumption Input: Wire gauge chosen from a lookup table: "20A → 14 AWG," without accounting for bundle length, ambient temperature, or installation method (single run vs. bundled with other cables). AI review output: Flags "textbook example" bias — ampacity tables typically assume a single conductor at 30°C in free air. Bundling with other current-carrying cables at elevated ambient temperature requires derating factors. Recommends recalculating with correction factors per NEC Table 310.15(B)(3)(a) or local equivalent.

Example 3 — Failed/unhelpful review (what NOT to settle for) Input: "I measured 0V across the relay coil, so the coil must be open." AI review output (bad, first pass): "Yes, that's a reasonable conclusion, a 0V reading is consistent with an open coil." — This is agreement bias: the AI merely validated the framing instead of challenging it. Corrected prompt: "What's wrong with this reasoning?" (not "is this correct?") AI review output (good, second pass): Points out that 0V across the coil could mean the coil never received voltage in the first place (open upstream contact, blown fuse, de-energized relay driver) rather than the coil itself being open. Recommends checking voltage at the coil terminals while the circuit is commanded ON, then checking coil resistance with power removed, before concluding the coil is faulty.