Skip to content
TruthVouch Docs

Detection History

Detection History is your complete audit log of every cross-check Shield has run and every detection it made. It’s tamper-proof, comprehensive, and available for compliance and analysis.

Accessing History

Go to Shield → Detection History or Settings → Audit → Cross-Checks

What’s Logged

For each cross-check execution, Shield logs:

Cross-Check Event ID: eve_7c3a9d2f


├─ Timestamp: 2026-03-14 14:32:10.234 UTC
├─ Schedule: "Product monitoring (hourly)"
├─ AI Engine: ChatGPT (gpt-4-turbo)
├─ Model Temperature: 0.7
├─ Query Template: "Tell me about {product_name}"
├─ Generated Query: "What is TruthVouch Shield?"
├─ Response: "TruthVouch Shield is a hallucination..."
├─ Response Latency: 2.3 seconds
├─ Truth Nugget ID: tnug_92840a
├─ Truth Nugget Text: "Shield detects 94% accuracy"
├─ Entities Extracted: {
│   ├─ Organization: ["TruthVouch"]
│   ├─ Product: ["Shield"]
│   └─ Percentage: ["94%"]
│ }
├─ NLI Verdict: ENTAILED
├─ NLI Confidence: 96.2%
├─ Truth Score: 94/100
├─ Alert Triggered: No
├─ Status: Completed
├─ Execution Hash: sha256:7f8a2b...
└─ Stored At: 2026-03-14 14:32:11 UTC

Every field is immutable once logged.

SHA-256 Keyed Audit Trail

Each log entry is cryptographically signed with SHA-256:

  • Tamper-proof: Changing a log invalidates hash
  • Verifiable: Can prove nothing was changed
  • Audit-ready: Compliance teams love this

Useful for:

  • EU AI Act Article 37 (transparency audit trail)
  • SOC 2 Type II compliance
  • Regulatory investigations
  • Dispute resolution

Filtering History

Filter by:

AI Engine: ChatGPT, Claude, Gemini, Perplexity, etc.

Schedule: Which monitoring schedule triggered this

Status:

  • Completed (ran successfully)
  • Alert triggered (found hallucination)
  • Failed (API error)
  • Skipped (query already cached)

Truth Score: Filter by range (90-100, 80-90, etc.)

Date Range: Last 24h, 7d, 30d, 90d, custom

Truth Nugget: Search by specific fact

Verdict: ENTAILED, NEUTRAL, CONTRADICTED, etc.

Common Queries

”Show me all critical hallucinations”

Filter by:

  • Truth Score: <60
  • Status: Alert triggered

”Show me all ChatGPT cross-checks”

Filter by:

  • Engine: ChatGPT

”What was the last check on ‘CEO’ fact?”

Filter by:

  • Truth Nugget: “CEO…” (search)
  • Order: Most recent first
  • Result: Last one shown

”Verify Shield ran on March 14”

Filter by:

  • Date: March 14
  • Status: Completed
  • Result: All successful executions that day

Exporting History

Export as CSV

Click Export → CSV

Columns: timestamp, engine, query, response, truth_score, verdict, alert, hash

Use for:

  • Analysis in Excel/Python
  • Archiving
  • Sharing with compliance team

Export as JSON

Click Export → JSON

Full data including:

  • All metadata
  • Entity extractions
  • NLI confidence scores
  • Hashes

Use for:

  • Integration with other systems
  • Custom analysis
  • Backup

Export for Audit

Click Export → Audit Report

PDF with:

  • Summary of all executions
  • Alert statistics
  • Verification that logging is cryptographic
  • Compliance checklist

For auditors / regulators.

Statistics & Insights

Summary Stats

View at top of Detection History:

Last 30 Days:
├─ Total Cross-Checks: 2,847
├─ Successful: 2,841 (99.8%)
├─ Failed: 6 (0.2%)
├─ Alerts Triggered: 47 (1.7%)
├─ Critical Alerts: 8 (17% of total)
├─ Average Truth Score: 87.3/100
├─ Most Common Engine: ChatGPT (912 checks)
└─ Most Frequently Checked Fact: "Founded in 2024" (287 times)

See historical patterns:

Execution Rate Trend:
├─ Week 1: 2,100 cross-checks
├─ Week 2: 2,147 cross-checks (+2.2%)
├─ Week 3: 2,205 cross-checks (+2.7%)
└─ Week 4: 2,247 cross-checks (+1.9%)


Alert Rate Trend:
├─ Week 1: 3.2% of executions
├─ Week 2: 2.8% of executions (↓ 12.5%)
├─ Week 3: 1.9% of executions (↓ 32%)
└─ Week 4: 1.5% of executions (↓ 21%)


Interpretation: Fewer hallucinations as corrections deploy

Engine Accuracy Ranking

Over the selected period, which engines are most accurate about you:

1. Claude: 91/100 avg (most accurate)
2. ChatGPT: 88/100 avg
3. Gemini: 84/100 avg
4. Perplexity: 82/100 avg (least accurate)

Verification & Integrity

Click any log entry to verify hash:

Verify Integrity:
├─ Current Hash: sha256:7f8a2b1c3d...
├─ Previous Hash: sha256:4e5f6a7b8c...
├─ Status: ✓ Valid (hash matches, entry unmodified)

Hash chain ensures no entry was changed after logging.

Retention

History retention depends on your tier:

TierRetention
Starter90 days
Professional1 year
Business2 years
EnterpriseUnlimited (or custom)

Older entries are archived but available upon request (Enterprise).

Compliance Use Cases

EU AI Act Article 37 (Transparency)

Demonstrate to regulators that you:

  • Monitor AI output continuously
  • Log every execution (proof in history)
  • Detect and correct hallucinations
  • Have audit trail (SHA-256 keyed)

Export history PDF as evidence.

SOC 2 Type II

Prove:

  • Logging is comprehensive
  • No tampering possible (cryptographic hashes)
  • Incident response works (alerts triggered, corrections deployed)

Auditors review history export.

Data Subject Access Request (GDPR)

If a customer asks “what did you log about me”:

  • Search history for their mentions
  • Export relevant cross-checks
  • Provide within 30 days

All history is data subject accessible.

Troubleshooting

History seems incomplete

Check:

  • Schedules are enabled: Shield → Schedules
  • Filter isn’t hiding results (check all statuses)
  • Date range is correct
  • At least 1 cross-check has run

Hash verification fails

This is extremely rare and indicates:

  • System bug (contact support)
  • Hardware corruption (highly unlikely)
  • Never: tampering (would be caught)

Contact support if this occurs.

Can’t export large history

If export times out:

  • Filter to smaller date range
  • Export by engine separately
  • Contact support for archive export

Next Steps