Genesis / Prototype Stage / PROTOCOL

Every handoff has a responsible role.

Proposed workflow. Scroll or select a step to explore its inputs, action, outputs and evidence. On mobile, expand ordinary vertical cards.

PROTOCOL / 01

Every handoff has a responsible role.

Proposed workflow. Scroll or select a step to explore its inputs, action, outputs and evidence. On mobile, expand ordinary vertical cards.

Synthetic Demo · Not production records

01 / 06

Task Registry

A submitter registers the task

Read the full explanation

Task Registry

A submitter registers the task

Input: Licensed UTF-8 samples, rule version and resource limits.
Action: Check permissions and acceptance conditions; freeze the task definition.
Output: A Task ready for assignment.
Evidence: task_id, input_digest, policy_version.

Illustrative JSON · Not an execution log

{
  "synthetic": true,
  "production_evidence": false,
  "task_id": "synthetic-task-01",
  "state": "registered",
  "input_ref": "sample-001",
  "license_ref": "synthetic-license",
  "rule_version": "demo-1",
  "input_digest": null
}

02 / 06

Coordinator

A coordinator assigns a bounded lease

Read the full explanation

Coordinator

A coordinator assigns a bounded lease

Input: Frozen tasks, available controlled Nodes and permissions.
Action: Lease by capability; record generation, retries and expiry.
Output: A Task Attempt and lease.
Evidence: attempt_id, generation, lease_expires_at.

Illustrative JSON · Not an execution log

{
  "synthetic": true,
  "production_evidence": false,
  "task_id": "synthetic-task-01",
  "attempt_id": "synthetic-attempt-01",
  "node_id": "synthetic-node-01",
  "state": "leased",
  "generation": 1,
  "lease_policy": "illustrative-lease-1"
}

03 / 06

Node Runner

A contributor device reads and runs

Read the full explanation

Node Runner

A contributor device reads and runs

Input: A valid lease, fixed Runner and authorized input.
Action: Process text offline under the designated rule and return structured output.
Output: An Artifact or explicit failure reason.
Evidence: runner_version, input/output_digest, state events.

Illustrative JSON · Not an execution log

{
  "synthetic": true,
  "production_evidence": false,
  "attempt_id": "synthetic-attempt-01",
  "state": "completed",
  "runner_version": "illustrative-runner-1",
  "network_access": false,
  "artifact_id": "synthetic-artifact-01",
  "output_digest": null
}

04 / 06

Verifier

A designated verifier reviews

Read the full explanation

Verifier

A designated verifier reviews

Input: Structured artifact, permissions and reference rule.
Action: Check versions, structure and known answers; distinguish pass, fail and error.
Output: A Verification record.
Evidence: verifier_id, policy_version, evidence references.

Illustrative JSON · Not an execution log

{
  "synthetic": true,
  "production_evidence": false,
  "verification_id": "synthetic-review-01",
  "artifact_id": "synthetic-artifact-01",
  "state": "illustrative-pass",
  "policy_version": "illustrative-policy-1",
  "acceptance": "not-decided"
}

05 / 06

Acceptance

A coordinator decides separately

Read the full explanation

Acceptance

A coordinator decides separately

Input: Valid Verification, task role and frozen acceptance policy.
Action: Decide whether the unique acceptance slot is filled; record the reason.
Output: Accepted, Rejected or awaiting recovery.
Evidence: acceptance_id, policy_version, decision_reason.

Illustrative JSON · Not an execution log

{
  "synthetic": true,
  "production_evidence": false,
  "acceptance_id": "synthetic-accept-01",
  "verification_id": "synthetic-review-01",
  "state": "illustrative-accept",
  "policy_version": "illustrative-policy-1",
  "reason": "separate coordinator decision"
}

06 / 06

Contribution Store

Store evidence-linked contributions

Read the full explanation

Contribution Store

Store evidence-linked contributions

Input: Accepted artifacts and provenance relationships.
Action: Transactionally link the task, attempt, artifact, review and acceptance.
Output: A queryable, revocable contribution index.
Evidence: contribution_id, provenance, revocation.

Illustrative JSON · Not an execution log

{
  "synthetic": true,
  "production_evidence": false,
  "contribution_id": "synthetic-contribution-01",
  "acceptance_id": "synthetic-accept-01",
  "state": "illustrative-recorded",
  "evidence_refs": [
    "synthetic-task-01",
    "synthetic-attempt-01",
    "synthetic-artifact-01",
    "synthetic-review-01",
    "synthetic-accept-01"
  ],
  "economic_rights": false
}

01 / 06

Task Registry

A submitter registers the task

Input: Licensed UTF-8 samples, rule version and resource limits.
Action: Check permissions and acceptance conditions; freeze the task definition.
Output: A Task ready for assignment.
Evidence: task_id, input_digest, policy_version.

Illustrative JSON · Not an execution log

{
  "synthetic": true,
  "production_evidence": false,
  "task_id": "synthetic-task-01",
  "state": "registered",
  "input_ref": "sample-001",
  "license_ref": "synthetic-license",
  "rule_version": "demo-1",
  "input_digest": null
}

Dashboard — A submitter inspects the evidence. An inspectable task view.

PROTOCOL / 02

Follow an outcome back to its origins.

Contribution Graph is a proposed provenance index. Select an entity for minimal fields; switch views to focus on tasks, artifacts or verification.

Synthetic Demo · Not production records

Minimal field example

Task

Frozen inputs, rules and acceptance policy.

The task freezes authorized samples, a rule version and an acceptance policy.

{
  "task_id": "synthetic-task-01",
  "input_ref": "sample-001",
  "rule_version": "demo-1",
  "acceptance_policy": "illustrative-policy-1",
  "synthetic": true,
  "production_evidence": false,
  "status": "illustrative-only",
  "schema_version": "demo-1",
  "summary": "The task freezes authorized samples, a rule version and an acceptance policy.",
  "content_digest": null,
  "evidence_refs": [
    "sample-001",
    "synthetic-license"
  ]
}

Provenance does not automatically establish correctness, real-world identity, original ownership or reward rights. All fields below are synthetic examples.

Minimal field example

Contributor

A participant’s claims and permission references.

{
  "contributor_id": "synthetic-person-01",
  "role": "operator",
  "permission_ref": "synthetic-license",
  "synthetic": true,
  "production_evidence": false,
  "status": "illustrative-only",
  "schema_version": "demo-1",
  "summary": "The contributor supplies a controlled execution role, not a verified real-world identity.",
  "content_digest": null,
  "evidence_refs": [
    "synthetic-license"
  ]
}

Node

A controlled device’s declared identifier; not hardware proof.

{
  "node_id": "synthetic-node-01",
  "contributor_id": "synthetic-person-01",
  "capability_claim": "CPU / offline text",
  "synthetic": true,
  "production_evidence": false,
  "status": "illustrative-only",
  "schema_version": "demo-1",
  "summary": "The Node declares offline CPU text processing, not a real connected device.",
  "content_digest": null,
  "evidence_refs": [
    "synthetic-person-01"
  ]
}

Task

Frozen inputs, rules and acceptance policy.

{
  "task_id": "synthetic-task-01",
  "input_ref": "sample-001",
  "rule_version": "demo-1",
  "acceptance_policy": "illustrative-policy-1",
  "synthetic": true,
  "production_evidence": false,
  "status": "illustrative-only",
  "schema_version": "demo-1",
  "summary": "The task freezes authorized samples, a rule version and an acceptance policy.",
  "content_digest": null,
  "evidence_refs": [
    "sample-001",
    "synthetic-license"
  ]
}

Attempt

One execution attempt under a lease.

{
  "attempt_id": "synthetic-attempt-01",
  "task_id": "synthetic-task-01",
  "generation": 1,
  "state": "illustrative",
  "synthetic": true,
  "production_evidence": false,
  "status": "illustrative-only",
  "schema_version": "demo-1",
  "summary": "The attempt records one lease and state changes; a retry needs a new generation.",
  "content_digest": null,
  "evidence_refs": [
    "synthetic-task-01",
    "synthetic-node-01"
  ]
}

Artifact

A structured result identified by its content digest.

{
  "artifact_id": "synthetic-artifact-01",
  "attempt_id": "synthetic-attempt-01",
  "media_type": "application/json",
  "digest": "sha256: <calculated at runtime>",
  "synthetic": true,
  "production_evidence": false,
  "status": "illustrative-only",
  "schema_version": "demo-1",
  "summary": "The artifact links an attempt to content references; no valid digest is fabricated here.",
  "content_digest": null,
  "evidence_refs": [
    "synthetic-attempt-01"
  ]
}

Verification

A review outcome and evidence under a specified policy.

{
  "verification_id": "synthetic-review-01",
  "artifact_id": "synthetic-artifact-01",
  "verifier_id": "synthetic-verifier",
  "policy_version": "illustrative-policy-1",
  "outcome": "illustrative-pass",
  "synthetic": true,
  "production_evidence": false,
  "status": "illustrative-only",
  "schema_version": "demo-1",
  "summary": "Verification connects an artifact to reference rules; a pass does not create acceptance.",
  "content_digest": null,
  "evidence_refs": [
    "synthetic-artifact-01"
  ]
}

Acceptance

The coordinator’s separate acceptance decision.

{
  "acceptance_id": "synthetic-accept-01",
  "verification_id": "synthetic-review-01",
  "decision": "illustrative-accept",
  "reason": "synthetic policy example only",
  "synthetic": true,
  "production_evidence": false,
  "status": "illustrative-only",
  "schema_version": "demo-1",
  "summary": "The coordinator makes a separate policy-based acceptance decision with a reason.",
  "content_digest": null,
  "evidence_refs": [
    "synthetic-review-01"
  ]
}

Contribution

A provenance link between an accepted artifact and contributors.

{
  "contribution_id": "synthetic-contribution-01",
  "acceptance_id": "synthetic-accept-01",
  "contributor_id": "synthetic-person-01",
  "economic_rights": false,
  "synthetic": true,
  "production_evidence": false,
  "status": "illustrative-only",
  "schema_version": "demo-1",
  "summary": "The contribution index links accepted work to a contributor without assigning economic rights.",
  "content_digest": null,
  "evidence_refs": [
    "synthetic-accept-01",
    "synthetic-person-01"
  ]
}

PROTOCOL / 03

Improvement needs comparison.
Comparison needs evidence.

Freeze the baseline, candidate, samples, labels and thresholds before running B, C, C, B. Repetition checks determinism; it does not create independent samples or a confidence interval.

Synthetic / Illustrative Data · Not UNORYN production results

Evidence missing

Each metric uses a shared scale; exact values appear below. Evidence completeness is synthetic checklist coverage, not a probability of correctness.

Synthetic / Illustrative Data · Not UNORYN production results
	Baseline	Candidate
Quality · F1	0.72	0.84
Latency · ms	100	96
Cost · illustrative units	1.00	0.95
Evidence completeness	4 / 4	2 / 4

Final conclusion

Insufficient evidence

Aggregate scores cannot replace row-level outputs, run records and review.

Evidence eligibility: Missing row-level outputs and review

Inspect B / C / C / B run evidence

{
  "synthetic": true,
  "complete": false,
  "order": [
    "B",
    "C",
    "C",
    "B"
  ],
  "runs": [
    {
      "role": "B",
      "f1": 0.72,
      "latency_ms": 100,
      "cost_units": 1
    },
    {
      "role": "C",
      "f1": 0.84,
      "latency_ms": 96,
      "cost_units": 0.95
    },
    {
      "role": "C",
      "f1": 0.84,
      "latency_ms": 96,
      "cost_units": 0.95
    },
    {
      "role": "B",
      "f1": 0.72,
      "latency_ms": 100,
      "cost_units": 1
    }
  ]
}

Regression

Each metric uses a shared scale; exact values appear below. Evidence completeness is synthetic checklist coverage, not a probability of correctness.

Synthetic / Illustrative Data · Not UNORYN production results
	Baseline	Candidate
Quality · F1	0.72	0.66
Latency · ms	100	95
Cost · illustrative units	1.00	0.96
Evidence completeness	4 / 4	4 / 4

Final conclusion

Regression: quality loss

F1 loss exceeds the frozen tolerance of 0.02.

Evidence eligibility: Complete, deterministic, reviewed (synthetic)

Inspect B / C / C / B run evidence

{
  "synthetic": true,
  "complete": true,
  "order": [
    "B",
    "C",
    "C",
    "B"
  ],
  "runs": [
    {
      "role": "B",
      "f1": 0.72,
      "latency_ms": 100,
      "cost_units": 1
    },
    {
      "role": "C",
      "f1": 0.66,
      "latency_ms": 95,
      "cost_units": 0.96
    },
    {
      "role": "C",
      "f1": 0.66,
      "latency_ms": 95,
      "cost_units": 0.96
    },
    {
      "role": "B",
      "f1": 0.72,
      "latency_ms": 100,
      "cost_units": 1
    }
  ]
}

Improvement verified

Each metric uses a shared scale; exact values appear below. Evidence completeness is synthetic checklist coverage, not a probability of correctness.

Synthetic / Illustrative Data · Not UNORYN production results
	Baseline	Candidate
Quality · F1	0.72	0.80
Latency · ms	100	96
Cost · illustrative units	1.00	0.95
Evidence completeness	4 / 4	4 / 4

Final conclusion

Improvement verified: threshold reached

Complete evidence, resource guardrails passed and absolute F1 gain ≥ 0.05.

Evidence eligibility: Complete, deterministic, reviewed (synthetic)

Inspect B / C / C / B run evidence

{
  "synthetic": true,
  "complete": true,
  "order": [
    "B",
    "C",
    "C",
    "B"
  ],
  "runs": [
    {
      "role": "B",
      "f1": 0.72,
      "latency_ms": 100,
      "cost_units": 1
    },
    {
      "role": "C",
      "f1": 0.8,
      "latency_ms": 96,
      "cost_units": 0.95
    },
    {
      "role": "C",
      "f1": 0.8,
      "latency_ms": 96,
      "cost_units": 0.95
    },
    {
      "role": "B",
      "f1": 0.72,
      "latency_ms": 100,
      "cost_units": 1
    }
  ]
}

Improvement not confirmed

Each metric uses a shared scale; exact values appear below. Evidence completeness is synthetic checklist coverage, not a probability of correctness.

Synthetic / Illustrative Data · Not UNORYN production results
	Baseline	Candidate
Quality · F1	0.72	0.74
Latency · ms	100	98
Cost · illustrative units	1.00	1.05
Evidence completeness	4 / 4	4 / 4

Final conclusion

Improvement not confirmed: below threshold

Complete evidence and resource guardrails passed, but absolute gain < 0.05.

Evidence eligibility: Complete, deterministic, reviewed (synthetic)

Inspect B / C / C / B run evidence

{
  "synthetic": true,
  "complete": true,
  "order": [
    "B",
    "C",
    "C",
    "B"
  ],
  "runs": [
    {
      "role": "B",
      "f1": 0.72,
      "latency_ms": 100,
      "cost_units": 1
    },
    {
      "role": "C",
      "f1": 0.74,
      "latency_ms": 98,
      "cost_units": 1.05
    },
    {
      "role": "C",
      "f1": 0.74,
      "latency_ms": 98,
      "cost_units": 1.05
    },
    {
      "role": "B",
      "f1": 0.72,
      "latency_ms": 100,
      "cost_units": 1
    }
  ]
}

Resource limit exceeded

Each metric uses a shared scale; exact values appear below. Evidence completeness is synthetic checklist coverage, not a probability of correctness.

Synthetic / Illustrative Data · Not UNORYN production results
	Baseline	Candidate
Quality · F1	0.72	0.84
Latency · ms	100	145
Cost · illustrative units	1.00	1.30
Evidence completeness	4 / 4	4 / 4

Final conclusion

Regression: resource guardrail failed

A timeout or overspend in either paired run rules out improvement even when quality increases.

Evidence eligibility: Complete, deterministic, reviewed (synthetic)

Inspect B / C / C / B run evidence

{
  "synthetic": true,
  "complete": true,
  "order": [
    "B",
    "C",
    "C",
    "B"
  ],
  "runs": [
    {
      "role": "B",
      "f1": 0.72,
      "latency_ms": 100,
      "cost_units": 1
    },
    {
      "role": "C",
      "f1": 0.84,
      "latency_ms": 145,
      "cost_units": 1.3
    },
    {
      "role": "C",
      "f1": 0.84,
      "latency_ms": 145,
      "cost_units": 1.3
    },
    {
      "role": "B",
      "f1": 0.72,
      "latency_ms": 100,
      "cost_units": 1
    }
  ]
}

Illustrative frozen policy: min_gain 0.05; maximum quality loss 0.02; every candidate run ≤ 120 ms and ≤ 1.20 cost units. All values explain logic only.

CONTRIBUTIONS / 04

Different work.
Different records.

Four proposed categories of contribution. Genesis starts with one task chain; four open marketplaces have not been implemented.

Compute

Execute authorized tasks within a bounded environment.

Example: A CPU Runner checks one offline text sample.
What is recorded: Node, Attempt, version, accepted artifacts and execution events.
Current status: Planned · Controlled execution awaits validation.
What is not yet proven: Actual hardware use, real operator and outsourced execution.

Knowledge

Provide data and feedback with clear permission, provenance and version.

Example: Maintain licensed samples with expected labels.
What is recorded: Permission manifest, provenance, version, digests and revisions.
Current status: Proposed · Input permissions and provenance fields are specified.
What is not yet proven: Originality, complete licensing and continuing permission.

Intelligence

Provide rules, code, methods or model improvements.

Example: Submit a text-rule candidate for comparison.
What is recorded: Authorship claims, rule versions, candidate artifacts and comparison evidence.
Current status: Proposed · Baseline/candidate comparison awaits a prototype.
What is not yet proven: Real improvement, causal contribution and generalizability.

Validation

Rerun, check and evaluate other results.

Example: A designated verifier reruns a fixed reference rule.
What is recorded: Verifier, policy version, control relationships, outcome and evidence.
Current status: Planned · A designated verifier is to be implemented.
What is not yet proven: Independence, resistance to collusion and verification cost.

SYSTEM / 06

Make the boundaries explicit first.

The proposed Genesis stack includes FastAPI, a Python CLI Node, SQLite or another relational database, and controlled artifact storage addressed by digest.

01 / Submission & management

Task RegistryDashboard

02 / Coordination services

CoordinatorVerifierContribution Store

Freeze policies, lease tasks, designate reviews, decide acceptance and store records.

03 / Contributor device

Node Runner

Read inputs and return artifacts in a bounded environment; no arbitrary code execution.

Proposed architecture. Not deployed, load-tested or security-audited. The coordinator and designated verifiers remain inside the trust boundary.

Every handoff has a responsible role.