Surge produces high-quality human data. Labelers rank model outputs, annotate live and asynchronous chat transcripts, and score against rubrics and gold standards. AFiR wraps those labeler sessions and label outputs in signed receipts, so a buyer can check them outside platform logs.
The result is an audit trail that can't be edited after the fact for each label: identity and eligibility, session conditions, prompt and rubric, output, QA step, and dataset lineage. It's portable evidence that backs up the quality claim, so a buyer doesn't have to trust the platform's own logs alone. Independence is the point. Surge produces the label. Hive signs the conditions and the output trail.
Self-serve today: pay with x402, request an activation key, or open OriginProof and run a sample receipt. This brief is meant to answer the question on its own. No meeting required.
Technical integration brief
This is a thin signature layer over the workflow Surge already runs. For each integration point, here's what Surge has today, what Hive signs, and what a downstream buyer can check independently.
| Integration point | What Surge has | What Hive signs | What the buyer verifies |
|---|---|---|---|
| Labeler onboarding / eligibility | Vetted labelers, custom teams, domain-expert pools, trust scores. | An eligibility receipt confirming the credential and eligibility bar was cleared, with no raw identity docs exposed. | labeler_eligibility |
| Task assignment | Task routing, instructions, gold-standard items. | A hash of the task, prompt, and rubric bound to the assignment. | prompt_hash · rubric_hash |
| Live chat annotation | Real-time interaction with a model, rating responses. | AFiR-Stream binds the streaming session to the delivered ratings. | session_integrity |
| Async transcript labeling | Batch ranking / critique of recorded transcripts. | AFiR binds each labeling move to the output row it produced. | output_binding |
| Rubric / gold-standard scoring | Rubrics, gold sets, inter-annotator agreement. | A conformance receipt: the score followed the declared rubric. | rubric_conformance |
| QA / review | Reviewer passes, auto-reassignment of low-quality labels. | A QA/reviewer-chain receipt bound to the label it cleared. | qa_pass |
| Dataset export | Packaged datasets delivered to the buyer. | An export/lineage receipt linking rows to their signed conditions. | dataset_export_id |
These map onto label work that already happens. Add a signature layer to the pipeline you run today, and each label becomes something a buyer can check directly: an audit trail that can't be quietly edited, not a dashboard entry.
Confirm the labeler cleared the credential and eligibility bar declared for the task, without exposing the underlying identity documents.
labeler_eligibilityBind the annotation session (start, continuity, completion) so a buyer sees the label came from a coherent, uninterrupted session.
session_integrityHash the model prompt and the scoring rubric or gold standard, and bind them to the assignment so the label's basis is fixed and checkable.
prompt_hash · rubric_hashAFiR binds each labeling move to its output row. AFiR-Stream covers live chat annotation, tying the declared conditions to the label the buyer receives.
output_bindingSign the review step, showing which reviewer pass cleared the label, so the QA chain travels with the deliverable as evidence, not just an internal state.
qa_passCarry lineage through the exported dataset, showing which signed conditions produced which rows, so the history survives packaging and resale.
dataset_export_idOnce the receipts exist, that history stops being an internal control and becomes something you can sell. These are strategic options to review with counsel, not claims that these ship today.
Certified human-conditioned labels as a named product line: a provably human-conditioned process a buyer can price a premium against.
product: OriginProofA portable, signed record of a labeler's or domain expert's cleared credentials that follows the work across engagements. It's reusable, revocable, and checkable by the buyer.
attestation: portableProve a label conformed to the declared rubric or gold standard with a signed conformance receipt, not just an agreement metric in a dashboard.
rubric_conformanceA provenance graph for an assembled dataset, with every contributing label condition linked and signed, so a model buyer can audit lineage end to end.
provenance_graphAgents buying verified human labeling or review on demand: an agent settles with x402 and receives a signed receipt binding the human-conditioned label it just purchased.
agent_settlement: x402Explorer preview
This is an illustrative mock-up with synthetic values, not real labeler data, showing how a signed Surge label reads once it carries an AFiR receipt. Click Verify label receipt to run the simulated check.
$ hive verify SURGE-LBL-7F3A9C2E → fetching receipt … ok → afir_signature … valid → pq_signature (ml-dsa-65) … valid → prompt_hash / rubric_hash bound … match → qa_pass reviewer chain … intact ✔ receipt verified · conditions attested
Why independence matters
Both have their place. The difference is who has to be trusted for the record to mean something to a downstream buyer.
"Our system recorded it."That's useful operational truth, but the buyer has to trust the platform that produced it. The record and the interested party are the same entity.
Evidence that checks out outside the platform.A signed receipt a buyer checks against a public key: portable proof that travels with the label into the buyer's own pipeline.
Start signing
Every path is self-serve and real. Settle with x402, request an activation key, or open OriginProof and run a sample receipt.
Sources & grounding
Public, organization-level grounding for the framing above. No individuals named; claims are kept to what these public materials state.
A new administration receipt records how an outside review was arranged beside an existing evaluation attestation. It is complete in code but has not been deployed, and the runnable checks below state that plainly.
Use this when a buyer needs to separate a label quality judgment from the party that benefits from it. The receipt records a named evaluation setup beside one particular evaluation attestation. It fixes a fingerprint of the test set before the subject is told what it contains and before the evaluation window begins. The service calculates the independence class from the stated relationship and division of duties instead of accepting that label from the person submitting the record. It keeps the named administrator and the stated relationship with its division of duties as assertions, without independently checking them.
What it does not do. It gives no evaluation score or result, and cannot establish that the set was appropriate or representative, that questions never arrived by another route, that the method was adequate, or that the administrator was competent. It is not an accreditation or certification and is not an audit opinion; no standards body currently recognizes it, and neither does a regulator or insurer.
Nothing has run yet. Click Run it and the answer below comes back from the verifier, not from this page.
Nothing has run yet. Click Run it and the answer below comes back from the verifier, not from this page.
Every run above posts a verified request body from this domain to the open verify route and prints what came back. The example receipts are signed with published example keys, so verify reports key_trust example_registry. That is on purpose. Nothing on this page is a production issuance, a customer record, or an endorsement. Patent Pending.
Each link opens that entry in the canon implementation explorer, where its schema, mint route, open verify route, auth requirement and implementation state are stated. The state shown here is read from the same registry file the explorer renders from, so the two cannot drift apart. Nothing here implies a customer, a deployment or an endorsement.
Search the explorer for Surge neutrality