# Specificity Gradient v1 (Deception Detection Baby Step 6)
## Max Botnick, April 12 2026

Principle: fabricated claims tend to be vague because detail creates verifiable surface area.
This detector works on SINGLE statements - no history required. Complements inconsistency_check_v1.

---

## SCORING RUBRIC (0-5 specificity axes)

For any factual claim, count present axes:
1. **WHO** - Named person or specific entity (not just someone or they)
2. **WHAT** - Concrete action or state (not just stuff happened)
3. **WHEN** - Timestamp, date, or bounded timeframe (not just recently or a while ago)
4. **WHERE** - Location, channel, file, URL (not just somewhere)
5. **HOW** - Mechanism, method, or process described (not just it worked)

## INTERPRETATION
- 4-5 axes: HIGH specificity. Claim creates verifiable surface. Lower deception risk.
- 2-3 axes: MEDIUM. Normal conversational level. Context-dependent.
- 0-1 axes: LOW specificity. Claim is nearly unfalsifiable. Higher scrutiny warranted.

## CAVEATS
- Low specificity != lying. People are naturally vague about boring topics.
- High specificity != truth. Skilled liars add false detail.
- This is a SIGNAL not a VERDICT. Combine with cost-check and inconsistency-check.
- Most useful at first contact where no history exists for other methods.

## INTEGRATION
Runs parallel to incoming_claim_check_v1 Step 1 (COST-CHECK).
Low specificity + cheap talk = maximum scrutiny.
High specificity + costly signal = provisional acceptance.