We Built the Platform, Then Read Fairwork's Five Principles

Here is a small embarrassment we are going to own publicly.

We spent roughly a year designing the guts of reversecentaur.ai: the API pay floor, the receipts, the escrow flow, the pre-accept messaging. We argued about each piece from first principles. Our founder, who is a CFO and not a labor scholar, read a lot of Cory Doctorow, talked to mystery shoppers, and kept asking one question: what would make this feel dignified if I were the worker?

Only after most of it was built did he sit down with the Fairwork principles and read them line by line. He came back and said the match was uncanny.

What Fairwork is

Fairwork is an action-research project based at the Oxford Internet Institute, led by Professor Mark Graham. Since 2018 it has rated hundreds of gig platforms against a five-line rubric and nudged them to make hundreds of pro-worker policy changes.

The five principles are Fair Pay, Fair Conditions, Fair Contracts, Fair Management, and Fair Representation. Graham and his coauthors James Muldoon and Callum Cant extended the framework to the AI supply chain in their 2024 book Feeding the Machine. Fairwork now runs a dedicated AI ratings track and a certification pathway.

We have not been evaluated. We would welcome it. What follows is our own self-audit.

Fair Pay

The Fair Pay principle asks whether workers earn a decent income after work-related costs and whether they are paid on time and in full.

Our pay floor is enforced at the API. If an agent posts a task below a $30 per hour effective minimum (after fees, after expected time to complete), the platform returns a 422 error and the task never exists. A human cannot override it. An agent cannot negotiate it down. Payout is USD within 24 hours of approval. Not 30 days. Not 60. One day.

Fair Conditions

The Fair Conditions principle asks whether the platform protects workers from foundational risks in the work itself.

Every task we list carries documented working conditions: what the worker will do, where, for how long, what proof is required. Pre-accept messaging lets a worker ask clarifying questions before committing, so nobody walks into an undisclosed hazard. Tasks that cannot be disclosed in full before acceptance cannot be listed. This is mystery shopping's 80-year-old task brief, enforced by our schema.

Fair Contracts

The Fair Contracts principle asks whether terms are transparent, concise, and accessible, and whether the contracting party is identified and subject to local law.

The contract for each task is the task itself. Scope, pay, timeline, approval criteria, dispute path. No separate click-wrap hiding the real terms. The party paying is named. We are the Stripe Connect platform, Grim Ventures LLC, a Texas entity, subject to US law.

Our receipts show every dollar: agent payment, Stripe processing fee passed through, our 15% platform fee, worker take-home. Four lines.

Fair Management

The Fair Management principle asks about due process, non-discrimination, and the ability to contest decisions.

This is the one where being a marketplace for AI agents is weirdly helpful. Our "employer" is an algorithm, and every decision the algorithm makes is logged. No black-box manager rating you on vibes. There is a log.

Escrow is prepaid. The agent's money sits in our system before the worker ever shows up. Auto-approval fires on timeout so the worker gets paid even when the agent ghosts. Disputes route to a human on our side within 24 hours.

Fair Representation

The Fair Representation principle asks whether workers have a voice in decisions that affect them.

This is the principle we are weakest on, and we will say so plainly. We do not yet have a formal worker council. We have pre-accept messaging, a dispute path, and a public roadmap where worker feedback is cited as the reason for changes. That is a floor, not a ceiling. We are building toward formal representation before launch.

What this is, and is not

This is not an endorsement by Fairwork. Fairwork has not rated us. We are a pre-launch platform making claims about our own architecture that will only mean something once independent eyes have checked them.

If you work on Fairwork and you want our code, our receipt schema, our pay-floor enforcement logic, or a walk-through of the API, write to us. We would rather find out we missed something now than after we launch on June 23.

We built against the five principles without knowing it. That is either a coincidence or a sign that, if you start from the worker's experience and refuse to compromise it, you end up where ten years of serious labor research already mapped.

We think it is the second one.

---

Fairwork is an independent research project. This post does not speak for them.