How we verify
every claim
Four phases. Licensed sources only. Named human sign-off. No anonymous AI output ships from ThriveFinity.
THE PROCESS
Five phases. One standard
Every verdict — Teardown, Audit, or Council — follows the same sequence. Tier affects scope, not rigour.
Phase 00 · Intake
Brief received. Scope defined
Every engagement begins with a structured intake. We read the brief, identify the claims to be verified, define the evidence scope, confirm the service tier, and return a written scope statement within 4 hours. Nothing retrieves until scope is agreed.
Founder submits the intake form with their brief, claims list, and context. We confirm receipt within 30 minutes.
We extract each verifiable claim, flag ambiguities, and return a written scope statement. You confirm or adjust before anything retrieves.
Teardown (1 claim), Audit (full asset), or Council (full Sentinel + named verifier). Scope gates retrieval — nothing proceeds without written confirmation.
Phase 01 · Retrieve
Sources first, claims second
Before a single claim is evaluated, the Sentinel retrieves relevant evidence from licensed corpora. No web scraping. No model training data. No unverified secondary sources.
Phase 02 · Challenge
Every claim meets its harshest critic
The Sentinel generates three adversarial positions against each claim: the most likely investor objection, the most likely regulatory challenge, and the most likely factual alternative. The claim must survive all three.
Phase 03 · Cite
Every number has a home
Surviving claims receive an inline citation in ThriveFinity’s standardised format. The citation includes source name, publication date, methodology note, and a confidence level.
Citation validity criteria — all five must pass
Phase 04 · Sign
A name on the record
A named human strategist — not AI, not a pseudonym — reads the complete draft, tests the reasoning, and signs. If they won’t sign it, it does not ship. Their name appears in the footer of every verdict.
- 1Claims without a traceable, licensed source
- 2Market size assertions derived from unaudited or proprietary models
- 3Competitive comparisons naming companies without their verified consent to the benchmark
- 4Regulatory compliance assertions without qualified legal review and sign-off
Live example
The Sentinel in action. From claim to cited verdict
ACCOUNTABILITY
The guarantee that keeps us honest
No questions asked. We publish the refund count and corrections publicly every month. Every Council verdict carries the same promise.
Monthly refund log · Published publicly
| Month | Verdicts delivered | Unsupported claims | Refunds issued | Corrections published |
|---|---|---|---|---|
| Apr 2026 | 7 | 0 | 0 | — |
| Mar 2026 | 11 | 1 | 1 | 1(corrected & published) |
| Feb 2026 | 9 | 0 | 0 | — |
| Jan 2026 | 8 | 0 | 0 | — |
| Dec 2025 | 6 | 0 | 0 | — |
AI transparency
What AI does — and what humans do
ThriveFinity uses AI as a research assistant, not as an oracle. Here is the exact boundary.
- →Queries licensed corpora (Gartner, Statista, CB Insights, PubMed, regulatory filings)
- →Generates an initial structured draft from retrieved sources
- →Matches submitted claims against retrieved evidence
- →Drafts adversarial challenges for human review
- →Formats citations and produces confidence scores
- ✓Read every claim and verify it against the cited source independently
- ✓Reject or rewrite any assertion that cannot be confirmed
- ✓Test the logical reasoning of the overall output
- ✓Sign with their name — and own the consequences if it is wrong
- ✓Apply the automatic refusal policy to exclude unacceptable claims
Transparent by statute. Honest by design
Every ThriveFinity verdict that contains AI-synthesised content carries a mandatory Article 50 disclosure. This is not an opt-in. It ships with every output, before the client reads it.
- AI synthesis flagged at the top of every verdict document
- Human sign-off name, date, and credential stated explicitly
- Source retrieval method and model usage disclosed on request
- No generated images, no deepfakes, no deceptive content — prohibited at intake
- Data stored on EU infrastructure; deletion on request, immediate
Providers of AI systems that interact with humans must ensure those humans are informed they are interacting with an AI. ThriveFinity embeds this disclosure in every AI-assisted verdict, with the human verifier’s signature as the accountability layer.
Read the full text