When humans review automated decisions

1. Why we publish this

Per EU Directive 2024/2831 Article 7, no significant employment-affecting decision involving a platform worker may be made by an automated system without human oversight. This document explains where the line sits at Svaggy: which decisions are made by software alone, which involve a human, and what that human actually does.

2. Decisions made by software alone

The following decisions are made entirely by software, with no human in the loop:

• Order assignment — choosing which rider receives an offer

• Forward-dispatch timing — deciding to dispatch a rider 5 minutes before the merchant marks the order ready

• Cascade advancement — moving an offer from one rider to the next after a 30-second timeout or decline

• Score updates — recomputing your acceptance, completion, and rating after each offer or delivery

• Surge window calculations — computing payouts including the surge bonus

• Notification fan-out — deciding which push or email to send for a state transition

• Inbox content generation — formatting "delivery complete" and "payout sent" messages

These are all consequential to your day, but none of them, individually, has a long-term effect on your standing on the platform. They are the operational machinery of the platform; they fail closed (the order parks for manual dispatch) or are immediately reversible (a misfired notification is annoying, not material).

3. Decisions that REQUIRE human review

The following decisions, even when initially flagged by an automated system, require a human ops member's explicit sign-off before taking effect:

• Suspension of your account (temporary or pending review)

• Termination of your account

• Issuing a written warning

• Reducing your scoring threshold or rate

• Manually pushing an offer to you outside the cascade (an ops decision, not algorithmic)

• Resolving a dispute about a delivery (customer claim, merchant claim, or your appeal)

• Handling a safety incident (any SOS-triggered event)

• Approving or rejecting a KYC document submission

• Rejecting your initial onboarding application

• Responding to an appeal you file

For each of these, the human reviewer:

• Receives the automated system's flag with the underlying data and reasoning

• Has authority to override the recommendation

• Records their decision and reasoning in the audit log (queryable for compliance)

• Cannot delegate the decision back to software

For terminations specifically, two human reviewers are required — the initial reviewer and a senior signer. Either can refuse, in which case the termination does not happen.

4. Decisions where a human reviews on your request

These are decisions made by software that you can ask a human to review:

• A specific offer that the system skipped you for

• A specific delivery that was counted as an exception against your completion rate

• A customer rating you believe is unfair

• A coaching message you believe is unjustified

To request human review, see the appeals disclosure.

5. What "human review" actually means

A human reviewer at Svaggy:

• Is an ops staff member trained on the rider lifecycle, dispatch system, and Finnish labour-law minimum standards

• Is not the original decision-maker (or, for terminations, not either of the two reviewers in the original decision)

• Has access to the full context: your scores, the offer history, customer-side data, audit logs, prior coaching or warnings, and the algorithmic reasoning trace

• Documents their decision in the same audit log riders can see

• Cannot retaliate against you for the appeal itself

Reviewers operate from a written rubric, available on request via legal@svaggy.com.

6. When this disclosure changes

If we change which decisions require human review (in either direction — moving more work to software or adding more human-review checkpoints), we will:

• Bump this document's version

• Notify you in-app at least 30 days in advance

• Re-prompt acknowledgement for material changes