Trust engine
How Few Found decides what to trust
Evidence over claims, fixed rules over opinions, and a person for every decision that matters. This page shows how information moves through Few Found, how our AI learns, and exactly what runs today.
Three rules we never break
- Money never buys trustNo sponsored positions, no paid shortcuts through a check.
- Machines flag, people decideAutomation explains what it found; a person decides about a business.
- Every status can be lostRecords are dated, expire, and are withdrawn if standards slip.
The whole system, in four steps
Everything on Few Found follows the same path, whether it is a business asking to be checked or a message between two members.
- Live
1. What comes in
What a business tells us, the evidence it links to, member reports, and what members write to each other.
- Live
2. Automatic checks
Fixed rules and automated checks run the same way for everyone. They flag; they explain why.
- Live
3. A decision
Clear rule breaks are stopped at once. Anything about trust in a business is decided by a person.
- Live
4. A visible result
A dated public record, a published or refused message, and a notice that says what happened and why.
Checking a business
From a claim to a public record
When a business applies to be checked, this is what happens to what it shares.
- Live
1. The business shares evidence
Official business details, the people behind it, links to real work and one client reference.
- Live
2. Automated checks prepare a report
A report for the reviewer, with a reason for every flag. It never approves or rejects anyone.
- The website loads
- The website names the business
- How old the domain is
- The contact email matches the business
- Evidence links work
- Client references are complete
- No duplicate accounts
- Live
3. A person decides
A reviewer reads the report and the evidence and contacts the reference before deciding.
- Live
4. The record goes public
Anyone can open the record and see what was checked and when. It expires, and it can be lost.
Applying for Few Found Verification is free during launch: start an application. Anyone can check a record by its ID at fewfound.co/verify. What a business shares for review stays private to our reviewers; the public sees what was checked and when, never the material itself.
Keeping conversations safe
Spam and abuse stop before they reach anyone
Strikes are only for spam, scams and abuse towards other members. An incomplete profile never earns one. A first strike is a warning, a second pauses messaging and posting for 7 days, and a third suspends the account until a person reviews it.
- Live
1. A member writes
A message, comment, post, brief or application.
- Live
2. Rules check it first
Before anyone else sees it. Ordinary business talk always goes through.
- Insults, slurs and threats
- Asking for money to apply or join
- Link dumping and hidden links
- The same message to many members
- Several members reporting the same account
- Live
3. Refused, with a reason
Abuse, scams and repeated spam add a strike. 3 strikes suspend the account; each expires after 90 days.
- Live
4. Anyone can appeal
A person reads every appeal. Strikes found to be wrong are removed.
The full rules are in Terms §9. Members see their own standing, and appeal, under Account standing in the dashboard.
How our AI learns
It learns from real decisions, and earns its place slowly
General AI models are good at language but cannot check a registry, follow a network of fake reviews, or be held to a fixed rule. So our AI is one part of a pipeline where fixed rules and people come first.
- In development
1. Collect
Public pages about fraud patterns and industry standards. Our crawler obeys robots.txt and slows itself down; most sites say no, and we respect that.
- Live
2. Label
Every refused item is recorded with the rule and the outcome. Every appeal decision corrects the record. These become the examples the model learns from.
- In development
3. Train and test
The model is trained on labelled examples and tested on cases it has never seen before it is allowed to help.
- In development
4. Assist
It suggests and flags. Fixed rules still override it, and people still make every decision about a business.
What it will never do
- Money never buys a position, a check result or a faster decision.
- The AI never approves or rejects a business on its own.
- We never use what members write to each other for advertising, and never show it to other members.
- We keep only a short excerpt of refused content, for the notice, the appeal and the learning record.
- We say what we did not check.
Protecting the platform
Security that is switched on today
Every email is confirmed
A one-time code proves the address before an account works.
Passwords are never stored
Only a slow, salted hash (bcrypt). Codes are cryptographically random.
Every form is rate limited
Sign-in, sign-up, messages, posts, reports and appeals each have limits.
Sessions can be ended at once
A password reset signs out every device, and a suspended account loses access immediately.
Strict browser security
HTTPS only, a content security policy, and no framing by other sites.
Members can report and block
Reports go to the Few Found team, never to the member reported.
Status
Live today, and what we are building
| Part | Status | Notes |
|---|---|---|
| Automated checks on business applicationsPrepare a report for the reviewer. | Live | Prepare a report for the reviewer. |
| Human review of every business decisionNo exceptions. | Live | No exceptions. |
| Spam, scam and abuse screening with strikes and appealsMessages, comments, posts, briefs, applications. | Live | Messages, comments, posts, briefs, applications. |
| Labelled learning record from refusals and appealsCollected from day one. | Live | Collected from day one. |
| Government registry checks (GSTIN, MCA, Udyam)Tested offline; not yet connected. | In development | Tested offline; not yet connected. |
| Fake-review ring detection on a trust graphTested on synthetic data. | In development | Tested on synthetic data. |
| Trained moderation and verification modelsWaiting for enough labelled real examples. | In development | Waiting for enough labelled real examples. |
Questions about how we check?
Ask us anything about a check, a record or a strike. A person on the Few Found team replies.