The Humaniti arc

I. Why a roadmap, and why this one

Most roadmaps are backlogs with quarters attached. This one is a direction with phases attached. The difference matters because the four phases of Humaniti are not features — they are layers of what it means for labor to own the work it produces.

A roadmap that lists features and dates is an internal artifact masquerading as a public one. It tells you what the company is going to ship next. It does not tell you what the company is for. We have decided to publish the second kind. The phases here have no quarters, no deadlines, no probability estimates. Each phase has a clear technical shape, a clear social shape, and a clear definition of what it would mean for the phase to exist. We will publish the dates when there is anything honest to say about the dates.

The arc is — Network, then Record, then Economy, then Commons. Labor first. Ownership next. Governance last. Each phase is a precondition for the next. Skip one and the rest is theater.

II. Phase I — The Network

The Network is what most people will see first. It is a marketplace where vetted Humans complete AI labeling work, peer-consensus verifies it, and Builders pay directly per verified label. There are seventeen task types — image, text, audio, video, 3D, document — covering the full range of what a frontier-model team typically buys from a vendor. Consensus runs N-Human by default at three, configurable from one to seven. Disputes route to T3 Stewards. The qualified pool is gated by tier — phone-verified, ID-verified, reputation-verified — and weighted by per-skill Elo. Pay flows through Wise, Razorpay, Rapyd, Stripe Connect, or fiat-to-USDC, settled directly to the Human who did the work, not to a vendor's payroll.

The reason the Network exists is the diagnosis of what is wrong with the existing market. Today the labor that trains frontier AI is invisible labor. The buyer ships data to a vendor, receives labels back, and never sees who labeled what or how it was checked. The vendor's margin is not a line item. The Human who did the work does not exist on the receipt. Quality is a status report, not a verifiable property. This is fine when the model in question matters very little. It is corrosive when the model in question routes a court decision, a medical diagnosis, or a self-driving lane change.

The Network is the substrate. It does the work that needs doing today — frontier Builders need labels, and they need labels they can trust — and it does the work in a shape that the next three phases can build on. Direct pay is a precondition for the credit unit that comes later. Per-label receipts are a precondition for the on-chain attestations that come later. Tier-gated qualification is a precondition for the Steward eligibility that becomes the basis of governance.

If we never shipped Phase II, the Network would still be a defensible thing on its own — a transparent labor market that pays Humans directly and verifies labels with peer consensus. But Phase I is not the destination. It is the floor.

III. Phase II — The Record

Phase II is where the labor becomes durable. Every dataset becomes a signed, addressable, permanent record. Per-label provenance. Soulbound contribution receipts. On-chain attestations. The labor that produced the dataset travels with the dataset, queryable forever, owned by no single party.

The reason this matters is that today's datasets disappear. They sit in private buckets, get sold to one buyer, get re-keyed to another, and after enough hands the original labor history is gone. The labels survive; the labor does not. The Record is an answer to that. The cryptographic shape — signing, hashing, anchoring — is the part that gets the engineering attention. The social shape is more important. A Human who labels for the Network in Phase II builds a portable record of what they verified. They can show that record to a future employer, a future training dataset, a future academic study. The record is theirs. The dataset is the Network's. Neither belongs to a vendor.

The Record also makes the Network durable independently of Humaniti the company. If Humaniti ceases to exist tomorrow, the datasets do not. They are pinned in redundant storage. The attestations are on-chain. The signing keys are recoverable through the Network's social-recovery scheme, not through an internal Humaniti server. This is part of why we are calling Phase II what we are calling it. Record is a deliberate word. The dataset is a record of work, and the work has a record of who did it. Both are now in the world. Neither can be quietly deleted by any one actor.

Phase II is the precondition for Phase III. You cannot build a credit system on top of work history that is not durable. You cannot pay people retroactively for labor whose receipts have been overwritten.

IV. Phase III — The Economy

Phase III is where the work becomes ownership. The Network gets its own credit unit — HMNT — which Humans earn through verified labor. They can hold it in-network, withdraw it to fiat, or stake it. Staked credits multiply pay rates, unlock higher-stakes work, and qualify the holder for Steward roles. There is also a stable Work Credit — a USD-pegged work-unit derivative — for Builders who want to pre-fund tasks. Slashing penalises confirmed gaming. Delegated staking lets Humans delegate to Stewards and share the multiplier.

The retroactive distribution is part of Phase III on purpose. Every Human who did verified work in Phases I and II earns a credit allocation when the credit unit ships. This is not a marketing gesture. It is a structural one. The Network would not have a Phase III without the labor of Phases I and II, and the credit system has to acknowledge that or it becomes another vendor model with extra steps.

The reason an internal credit unit matters — instead of just paying everyone in fiat and being done — is that fiat-only labor is rented labor. You sell your time, you walk away with cash, the work belongs to whoever bought it. The Economy phase asks a different question. What if the work you did stayed yours? What if the Humans who built the Network's reputation accumulated equity in the Network's reputation? That is what credits do. They are not a token speculation. They are a claim on a share of what the labor produced.

This is the phase where the cynics will arrive in numbers. They are doing crypto. They are launching a token. The whole thing was a token play. Read Phases I and II again. The Network is the substrate. The Record is the durability layer. The credit system is the ownership layer that those two earn. Phase III without Phases I and II is a token. Phase III after Phases I and II is a labor union with a balance sheet.

V. Phase IV — The Commons

Phase IV is where the rules of the Network become the property of the people who use it. Network governance moves on-chain, weighted by verified work and stake. There are sub-DAOs — Quality Council, Language Guilds, Domain Experts, Worker Welfare, Treasury, Disputes, Platform Dev, Growth. Annotation DAOs let collectives of Humans co-own the datasets they verify, with royalties flowing to contributors. Retroactive Public Goods Funding routes surplus into open datasets, tooling, education, and accessibility. The Worker Welfare sub-DAO funds hardship grants, healthcare matching, and retraining for active Humans out of network surplus.

The role of Humaniti the company in Phase IV is the role of a steward, not an owner. We will hold the keys until the Network can hold its own keys. We will operate the infrastructure until the Network can operate its own infrastructure. Then we will hand it over. This is not a vague promise. It is the only honest endpoint of the arc. A network whose labor is verified by Humans, whose datasets are owned by Humans, and whose credits accrue to Humans, has to be governed by Humans. Anything less is the company taking back what it claimed to give.

The Commons phase is the answer to the question we get asked most often by skeptical Builders — why should we trust that you will not pull the rug? The honest answer is that you should not have to trust us. The Commons phase puts the rules out of our reach. The dataset, the labor record, the credit unit, and the governance protocol all become things we no longer own. We become one voice among many.

VI. Why this arc and not a different one

There are other arcs we could have published. We could have built only Phase I and called it a vendor with a better website. We could have skipped Phase II and built the credit system on top of mutable datasets. We could have skipped Phase IV and quietly held governance forever. We chose this arc because it is the one we can defend in public. Each phase is a precondition for the next. None of them is optional. Skip one and the rest is theater.

The arc is also a filter. It tells Builders who are looking for a vendor that we are not the vendor for them. It tells Humans who are looking for a tasking app that the work is also accumulating into something larger. It tells investors who are looking for a moat that the moat is the labor base, not the software. It tells regulators that the network is meant to outlive its founders. We would rather narrow the audience to the people who want the thing we are actually building than broaden it by lying about what we are building.

This is the public roadmap. We will publish the next note when something material changes. We will not publish a note for every shipped feature, for every milestone, for every quarter. Phases are the unit of public communication. Features are the unit of internal communication. We are trying to keep the two separate.

— The Humans of Humaniti