How trust works

FirstSqueak doesn't ask for proof of purchase. Honest data comes from several layers instead: how long an account has existed, whether it's verified, whether its owner keeps following up on their things, what other trusted members think, and moderation. This page explains how those combine. All constants live in one place in the code (trust.Default) and are covered by tests.

Tiers

• Unverified reviewer (new): every account starts here after confirming its email. Its reports count at a reduced weight.
• Verified: the account has a passkey (fingerprint, face or device PIN). This is the full base weight.
• Trusted reviewer: earned automatically, never bought or requested. It needs both of: - a real follow-up history: an account at least 6 months old, at least 3 on-time check-ins spread over at least 2 products, and no upheld moderation flags in the past 12 months; - a trust score of 60 or more (see below).

Helpful marks can speed up promotion, but they can never replace months of consistent follow-ups: those are the strongest signal that a real person owns real things. An upheld flag demotes a trusted reviewer automatically. Removing your last passkey returns a verified account to unverified.

How much a report counts

The weight of an owner's reports in durability scores is:

• tier base weight: unverified 0.3, verified 1.0, trusted 1.2;
• × account age factor: from 0.5 at signup, rising linearly to 1.0 after 6 months;
• × a reputation bonus: 1 + 0.1 × reputation, capped at 1.3;
• × a moderation penalty: 0.7 per upheld flag in the past year, never below 0.2.

The result is kept between 0.1 and 1.5, so no single account can dominate a product's score. Reports from deleted accounts stay in the data at a fixed weight of 0.5.

Reputation from helpful marks

When someone marks your review helpful, you gain reputation according to their weight at that moment. The rules are designed so reputation can't be farmed:

• Marks from unverified accounts are counted and shown, but add no reputation.
• You can't mark your own reviews, and each person can mark a review only once. Removing a mark removes its effect.
• One person can give you at most 1.0 in total, however many of your reviews they mark.
• If you also mark their reviews (reciprocal marking), their marks count at a quarter.
• Several endorsers acting from the same network (same short-lived IP hash) count once.
• Bursts of marks from young accounts (3 or more within 48 hours from accounts under 30 days old) count for nothing, and the account receiving them is flagged for moderators.
• Marks on reviews that are hidden, or that have an upheld flag, stop counting.
• Marks fade: each loses half its value per year unless renewed.
• Total reputation grows logarithmically: the hundredth mark matters much less than the first.

Every mark is stored with the marker's weight at the time, so reputation can always be recomputed from scratch.

Trust score (0–100)

Trust is recomputed every night.

Other defences

• Rate limits on signups, logins, reviews, check-ins, comments, marks, reports and new products; stricter for new accounts.
• Burst detection: when a product suddenly gets 5 or more reviews within 48 hours from new or young accounts, those reviews are held out of scores until a moderator looks.
• Copy-paste detection: reviews, check-in notes and comments that nearly duplicate recent text by the same person, or on the same product, are held for moderation instead of being published.
• Bot guards: invisible honeypot fields and a minimum time to fill in public forms. No third-party CAPTCHA.
• Community reports: anyone can report content, and moderators decide.
• No money involved: no affiliate links, no brand accounts. There's nothing to gain from fake reviews here.

Nothing is silently deleted: held or excluded contributions are kept and marked, so every decision can be reversed and scores recomputed.