Skill: Fact Check Claim

Verify any numerical or factual claim against primary sources. Prevents the Tesla Fleet failure mode: writing strategy based on fabricated or outdated numbers.

Trigger

User says: "fact check this", "verify this", "is this true", "source this", "verifica esto con fuentes", or when preparing any public-facing content (tweets, pitches, LinkedIn posts, mayor outreach).

Also trigger automatically when:

• Drafting a tweet with numbers
• Preparing a pitch for a real person/company
• Writing a scorecard (all numbers need sources)
• About to post anything on social media

Purpose

100% accuracy for anything going to real audiences. A wrong number kills credibility instantly. Tesla pitch almost got posted with "9 cameras accessible via Fleet API" — which is false. Option B tweet had "500M km" — actual is 730M.

Workflow

Step 1: Identify Every Claim

Parse the text and extract every:

• Numerical claim (figures, percentages, dollar amounts, counts)
• Date claim (when something happened)
• Name claim (who did what, company status)
• Technical claim (what an API exposes, what a product does)
• Legal claim (what a law requires, regulatory status)

Step 2: Rank by Risk

High risk (MUST verify):

• Figures over $1M
• Claims about what companies do/don't offer
• Technical API capabilities
• Legal status in any jurisdiction
• Competitor product features
• Anything in a pitch to a real person

Medium risk (SHOULD verify):

• Historical numbers (10+ years old may be outdated)
• Market size estimates
• Trend claims ("X is growing at Y%")

Low risk (can flag as "needs verification" without blocking):

• General context or framing
• Well-known public facts
• Analogies or comparisons

Step 3: Search for Primary Sources

For each high-risk claim, use WebSearch to find:

1. Official company page, SEC filing, or press release
2. Government data (FHWA, USDOT, NYC Comptroller, etc.)
3. Reputable business press with recent date
4. Academic or industry research with methodology

Rules:

• Date on source MUST be checked. 2023 data is stale for 2026 claims.
• Blog posts and Twitter threads are NOT primary sources
• Estimates from consultants are third-tier unless methodology is cited
• If no primary source found, mark as UNVERIFIABLE and flag

Step 4: Return Verdict Per Claim

For each claim, return:

### CLAIM X: "[verbatim claim]"
**Verdict:** VERIFIED / PARTIALLY TRUE / INCORRECT / UNVERIFIABLE
**Actual figure/fact:** [if different from claim]
**Primary source:** [URL]
**Source date:** [when the source is from]
**Supporting sources:** [other URLs]
**Recommended rewrite:** [exact text with corrected value]

Step 5: Produce Rewrite

At the end, output the full rewritten text with all corrections applied. The user should be able to copy-paste without further editing.

Step 6: Red Flag List

Highlight any claim that:

• Was off by >10% (conspicuous error)
• Referenced a company that has rebranded/shut down
• Used outdated technical specifications
• Confused USD with local currency
• Would be spotted by an insider in the target audience

Known Fact-Check Patterns (From Past Sessions)

Tesla Fleet API

• ❌ "Fleet API exposes camera frames" → Wrong, Fleet Telemetry is scalar-only per vehicle_data.proto
• ❌ "9 cameras" → Only on Model 3/Y HW4. Cybertruck has 7. Model S/X has 8 with no cabin cam.
• ❌ "Tesla pays owners for data" → $0 today, never has

Option B Ecosystem Claims

• ❌ "Hivemapper 500M km" → Actually 730M per beemaps.com live stats
• ❌ "VW ADMT" → Rebranded to MOIA America in January 2026
• ❌ "NATIX $60/mo" → Ceiling not average, paid in NATIX tokens (down 95% from ATH)
• ❌ "USB dashcam" → USB dongle; VX360 doesn't record, monetizes built-in cameras

CITYaaS Original Draft

• ❌ "43,000 hydrants" → 47,000 per Miami-Dade WASD Jan 2025
• ❌ "$140K/year signal inspection" → $1.2-3.7M per USDOT National Traffic Signal Report Card
• ❌ "$774K/year hydrant inspection" → $5.9-9.4M at real contract rates ($125-200/hydrant)
• ❌ "NYC $300M pothole claims over 5 years" → $138M over 6 years per NYC Comptroller ClaimStat 2015
• ❌ "46,154 structurally deficient bridges" → ~42,000 per ARTBA 2024
• ❌ "NYC ADA $1.2B settlement" → $1.55B over 10 years per Disability Rights Advocates 2019

CARRIERaaS Draft

• ❌ "Servientrega Bogota-Medellin document $6" → Actually $1.25 USD per enviotodo.com.co (confused USD with COP)
• ❌ "PiggyBee/Airmule did trustless multi-hop" → They did single-hop and died
• ❌ "Amazon Flex allows third-party packages" → Explicitly forbidden in driver agreement

Anti-Patterns (Do NOT Do)

• ❌ Accept a number because it "sounds right"
• ❌ Skip source verification for "well-known" facts
• ❌ Use a blog post as primary source
• ❌ Skip the date check on sources
• ❌ Assume old data is still current
• ❌ Let the user post something without fact-checking first

Output Format

# Fact-Check Report: [Topic]

## Summary
- Total claims: X
- Verified: X
- Partially true: X
- Incorrect: X
- Unverifiable: X

## Per-Claim Verdicts
[... per-claim breakdown ...]

## Red Flags
[Anything an insider would immediately spot]

## Rewritten Text
[Full text with corrections applied, ready to paste]