How trust scores work

Every number on REKKN is computed from verifiable data. Here is exactly how.

Trust score formula

The trust score is a 0–100 composite computed from four weighted inputs. These weights are hard-coded in our scoring engine and cannot be changed per-merchant.

35%
Resolution rate
What fraction of published claims has the business successfully resolved? Confirmed by the reporter or verified by staff.
25%
Issue rate
What fraction of verified transactions generated a published claim? Weighted by severity — a severe issue counts 3× an informational one.
20%
Response speed
Average hours from claim publication to first public business response. Faster is better. Maximum window: 168 hours (7 days).
20%
Verified volume
Total verified transactions backing the score. More evidence = higher confidence. Caps at 50 verified transactions.

Anti-gaming note: the best way to improve your score is to actually resolve claims. That is the intended behavior.

Trust states

The trust score maps to one of eight named states. States are evaluated in rule order — the first matching rule wins.

TrustedScore ≥ 80, confidence ≥ 60%, resolution rate ≥ 90%, zero unresolved severe claims, issue rate ≤ 5%.
ResponsiveScore ≥ 65, resolution rate ≥ 75%, zero unresolved severe claims.
ImprovingWas below 45 within the last 90 days and has risen by ≥ 15 points. Overrides Mixed, Responsive, or Needs Attention.
MixedScore ≥ 50 but does not qualify for Trusted or Responsive. No dominant pattern.
Watch closelyScore < 50, or resolution rate < 60%, or 7+ days without response on a severe claim.
High riskScore < 35, or issue rate > 15%, or unresolved severe claim with 30+ days of silence.
Under reviewSet by moderation. Trust computation is frozen while the merchant is under active investigation.
UnratedFewer than 3 verified transactions or confidence below 20%. Not enough data to compute a state.

The internal state IDs (trusted, responsive, mixed, needs_attention, consumer_warning, under_review, improving, unrated) are used in the API and codebase. The display labels above are what users see.

Severity weights

Not all claims are equal. Claim severity affects the issue rate calculation:

Informational
Minor
1.5×
Major
Severe

An informational report does not affect the score. A severe issue counts three times as heavily as a minor one.

Confidence

Confidence is a 0–100% scale based on verified transaction volume:

Low
3–15 verified transactions
Medium
16–34 verified transactions
High
35–50+ verified transactions

Full confidence (100%) is reached at 50 verified transactions. Below 3, no trust state is assigned.

Definitions

Verified transaction

An evidence-backed transaction object in the REKKN system. A transaction must include documentation (receipt, invoice, confirmation) and pass verification level 2 or higher to count toward the score.

Published claim

A moderated public experience report tied to a verified transaction. Claims are reviewed by human moderators before publication. Only published claims affect the trust score.

Issue rate

Severity-weighted published claims divided by verified transactions. This is a fraction: 0.05 means 5% of verified transactions generated a claim. The severity weighting means one severe claim counts as much as three minor ones.

Resolution

A claim is considered resolved when the reporter confirms the issue was addressed, or when staff verifies the business provided evidence of resolution (refund, replacement, corrective action).

The denominator question

Today, REKKN computes scores from verified transactions — records where the consumer provided evidence of a real purchase or engagement. We do not yet have access to every transaction a business processes.

This means the trust score reflects patterns among people who engaged with REKKN, not a complete census of all customers. We are transparent about this limitation. As merchant integrations come online, the denominator will expand. Until then, the confidence indicator tells you how much data backs the score.

Evidence integrity — EMILIA Protocol

All uploaded evidence is anchored through the EMILIA Protocol, which creates a cryptographic fingerprint at the moment of submission. Evidence cannot be altered after the fact by anyone — not the reporter, not the business, and not REKKN — without the change being detectable.

Recalculation

Trust scores are recomputed on every trust-affecting event: a claim is published, a business responds, a resolution is confirmed, or a moderation hold changes. There is no daily batch — recomputation is event-driven. There is no manual override. There is no pay-to-improve. The trust engine source code will be published as an open-source repository.

Open questions

We are still refining: how to weight merchant-initiated positive evidence, how to handle businesses operating across multiple categories, how to normalize scores across industries with different complaint baselines, and when to introduce positive-experience ratio as a scoring input. These decisions will be documented here as they are made.

How we moderateSearch the record